1 of 54

NLP Essentials

Overview

By Jinho D. Choi (2025 Edition)

Natural Language Processing (NLP) is a dynamic field within Artificial Intelligence focused on developing computational models to understand, interpret, and generate human language. As NLP technologies become increasingly embedded in our daily lives, understanding its fundamentals is crucial for both leveraging its potential and enhancing our interaction with language-based systems.

This course is designed to build a robust foundation in the core principles of modern NLP. We begin with text processing techniques that show how to manipulate textual data for model development. We then progress to language models, enabling computational models to comprehend and generate human language, and vector space models that convert textual data into machine-readable formats. Advanced topics include distributional semantics for creating context-aware word embeddings, and contextual encoding for analyzing word relationships within their surrounding text.

The latter part of the course focuses on practical application through team projects. Students will have the opportunity to work with cutting-edge NLP technologies, such as large language models, to develop real-world applications. This hands-on approach encourages creativity and innovation, with students proposing their own ideas and presenting demonstrations to their peers.

Learning assessment combines concept quizzes to reinforce theoretical understanding with hands-on programming assignments that develop practical implementation skills. By the conclusion of the course, students will have gained the knowledge and skills to navigate and contribute to the rapidly evolving field of NLP.

Prerequisites

Introduction to Python Programming
Introduction to Machine Learning

Sections

Syllabus

CS|QTM|LING-329: Computational Linguistics (Spring 2025)

General

GitHub: https://github.com/emory-courses/nlp-essentials/
Book:
Time: MW 2 - 3:45 PM
Location: MSC W201

Instructors

- Associate Professor of Computer Science, Data and Decision Sciences, Linguistics
- Office Hours and Location: MW 4:00 - 5:00 PM, Emerson Hall 500

Grading

Homework: 65%
Team Project: 35%
Your work is governed by the . Honor code violations (e.g., copies from any source, including colleagues and internet sites) will be referred to the Emory Honor Council.

Homework

Each topic will include homework that combines concept quizzes and programming assignments to assess your understanding of the subject matter.
Assignments must be submitted individually. While discussions are allowed, your work must be original.
Late submissions within a week will be accepted with a grading penalty of 15%. They will not be accepted after the solutions are discussed in class.

Concept Quizzes

Each section incorporates questions to explore the content more comprehensively, with their corresponding answers slated for discussion in the class.
While certain questions may have multiple valid answers, the grading will be based on the responses discussed in class, and alternative answers will be disregarded. This approach allows us to distinguish between answers discussed in class and those generated by AI tools like ChatGPT.

Programming Assignments

You are encouraged to use any code examples and invoke APIs provided in this book.
Feel free to create additional functions and variables in the assigned Python file. For each homework, ensure that all your implementations are included in the respective Python file located under the corresponding directory.
Usage of packages not covered in the assigned chapter is prohibited. Ensure that your code does not rely on the installation of additional packages, as we will not be able to execute your program for evaluation if external dependencies are needed.

Team Project

You are expected to:
- Form a team of 3-4 members.
- Present a to share your proposed idea, and write a .

Project Grading

Team members receive the same grade for the pitch presentation, live demonstration, and demonstration video.
Peer evaluations from other teams factor into your team grade.
Your feedbacks to other teams are graded individually.

For the project and final reports, you are required to indicate the contribution percentage of each team member, which impacts the individual grades for the assignment.

Contribution

If your team of two members received 4 out of 5 points, for example, and you indicate that your contribution was 60% while your teammate's was 40%, the points are distributed as follows:

You receive: 4 (team points) ⨉ 60 (your contributions) / 60 (max contributions) = 4 points.
Your teammate receive: 4 (team points) ⨉ 40 (your teammate's contributions) / 60 (max contributions) = 2.67 points.

This approach ensures that the grading reflects the effort and input of each team member, promoting fairness and accountability.

Consensus

Each team is required to submit a single, agreed-upon chart detailing the contribution percentages of all members for each team assignment. This means that you and your teammates must reach a consensus on the contribution rates before submitting your work.
Open communication and transparency are essential in this process. Disagreements should be resolved within the team, ensuring that the final submission reflects the true division of labor and contributions.

By adhering to these guidelines, you not only produce a strong research paper but also develop key skills in teamwork and fair assessment of contributions.

Schedule

CS|QTM|LING-329: Computational Linguistics (Spring 2025)

Date

Topic

Assignment

01/14

01/19

Martin Luther King Day

01/21

Attendance is mandatory for all Project Pitches and Live Demonstrations sessions.

Development Environment

This guide will help you set up your development environment by installing required tools: Python programming language, GitHub for version control, and PyCharm IDE.

Python

Install version 3.14.x or higher. Earlier versions may not be compatible with this course.

Homework

HW0: Getting Started

Task 1: Getting Started

In this assignment, you will:

Set up your development environment,
Install your first Python package using pip within a virtual environment,
Run a test program to verify the installation and environment configuration, and
Commit your changes and push them to your GitHub repository.

This will ensure your development workspace is properly configured for this course.

Package Installation

Once you set up the :

Open Terminal in PyCharm:
1. Click the Terminal icon () at the bottom left, or
2. Select the [View] > [Tool Windows] > [Terminal] menu).

Test Program

Create the project structure:
1. Create a new Python package called in your nlp-essentials directory.
2. Inside src, create a package.

Commit & Push

Create a file:
1. Create the file in your nlp-essentials root directory
2. Add the following lines to exclude unnecessary files:

Submission

Submit the URL of your GitHub repository to Canvas.

Task 2: Project Ideas

Share your team project concept by filling out the form in Canvas (about 100-150 words). Your description will be posted on the page to help classmates discover shared interests and form teams.

Rubric

GitHub Setup (0.2 points):
- Private repository created.
- All instructors added as collaborators.

Text Processing

Text processing refers to the manipulation and analysis of textual data through techniques applied to raw text, making it more structured, understandable, and suitable for various applications.

Sections

Frequency Analysis

If you are not acquainted with Python programming, we strongly recommend going through all the examples in this section, as they provide detailed explanations of packages and functions commonly used for language processing.

Frequency Analysis

Frequency Analysis examines how often each word appears in a corpus. It helps understand language patterns and structure by measuring how often words appear in text.

Word Counting

Consider the following text from Wikipedia about Emory University (as of 2023-10-18):

Our task is to determine the number of word tokens and unique word types in this text.

Q1: What is the difference between a word token and a word type?

A simple way of accomplishing this task is to split the text by whitespace and count the resulting strings:

L1: Import the class, a special type of a , from the package.
L3: Use to indicate the parameter type (str) and the return type (Counter).
L4: the corpus

L1: The corpus can be found .
L4: Save the total number of word tokens in the corpus, which is the of counts values.
L5: Save the number of unique word types in the corpus, which is the of counts

When running this program, you may encounter FileNotFoundError. To fix this error, follow these steps to set up your working directory:

Go to [Run] > [Edit Configurations] in the menu.

Word Frequency

In this task, we aim to retrieve the top-k most or least frequently occurring word types in the text:

L1: Sort words in counts in descending order and save them into dec as a list of (word, count) , sorted from the most frequent to the least frequent words (, , ).
L2: Sort words in counts in ascending order and save them into asc as a list of (word, count) tuples.

Notice that the top-10 least-frequent word list contains unnormalized words such as "Atlanta," (with the comma) and "Georgia." (with the period). This occurs because the text was split only by whitespaces without considering punctuation. Consequently, these words are recognized separately from the word types "Atlanta" and "Georgia". Therefore, the counts of word tokens and types processed above do not necessarily represent the distributions of the text accurately.

Q2: How can we interpret the most frequent words in a text?

Save Output

Finally, let us save all word types in alphabetical order to a file:

L2: Open outfile in mode (w).
L4: Iterate over unique word types (keys) of counts in alphabetical order.
L5: Write each word followed by a newline character to fout

L1: Creates the file if it does not exist; otherwise, its previous contents will be completely overwritten.

References

Source:
, The Python Standard Library - Built-in Types
, The Python Standard Library - Built-in Types

Regular Expressions

Regular expressions, commonly abbreviated as regex, form a language for string matching, enabling operations to search, match, and manipulate text based on specific patterns or rules.

Online Interpreter:

Homework

HW1: Text Processing

Task 1: Chronicles of Narnia

Your goal is to extract and organize structured information from C.S. Lewis's series, focusing on book metadata and chapter statistics.

Language Models

A language model is a computational model designed to understand, generate, and predict human language. It captures language patterns, learns the likelihood of a specific term occurring in a given context, and assigns probabilities to word sequences through training on extensive text data.

N-gram Models

An n-gram is a contiguous sequence of n items from text data. These items can be:

Words (most common in language modeling)
Characters (useful for morphological analysis)
Subword tokens (common in modern NLP)
Bytes or other units

For the sentence "I'm a computer scientist.", we can extract different n-grams:

1-gram (unigram): {"I'm", "a", "computer", "scientist."}
2-gram (bigram): {"I'm a", "a computer", "computer scientist."}
3-gram (trigram): {"I'm a computer", "a computer scientist."}

In the above example, "I'm" and "scientist." are recognized as individual tokens, which should have been as ["I", "'m"] and ["scientist", "."].

Q1: What are the advantages of splitting "I" and "'m" as two separate tokens, versus recognizing "I'm" as one token?

Unigram Estimation

Given a large corpus, a unigram model assumes word independence and calculates the probability of each word as:

Where:

: the count of word in the corpus.
: the vocabulary (set of all unique words).
The denominator represents the total word count.

Let us define the Unigram type:

L1: .
L3: A dictionary where the key is a unigram and the value is its probability.

Let us also define a function unigram_count() that takes a file path and returns a Counter with all unigrams and their counts in the file as keys and values, respectively:

We then define a function unigram_estimation() that takes a file path and returns a dictionary with unigrams and their probabilities as keys and values, respectively:

L3: Calculate the total count of all unigrams in the text.
L4: Return a dictionary where each word is a key and its probability is the value.

Finally, let us define a function test_unigram() that takes a file path as well as an estimator function, and test unigram_estimation() with a text file :

L1: Import the type from the typing module.
L3: The second argument accepts a function that takes a string and returns a Unigram.
L4: Call the estimator with the text file and store the result in unigrams.

L2: Pass the unigram_estimation() function as the second argument.

This distribution shows the most common unigrams in the text meeting the conditions in L8, dominated by the first-person pronoun "I", followed by proper nouns - specifically character names such as "Aslan", "Lucy", and "Edmund".

Q2: What advantages do unigram probabilities have over ?

Bigram Estimation

A bigram model calculates the conditional probability of the current word given the previous word as follows (: the total occurrences of in the corpus in that order, : a set of all word types occurring after ):

Where:

: the total occurrences of in the corpus in that order.
: a set of all word types occurring after .

Let us define the Bigram type:

A dictionary where the key is and the value is a nested dictionary representing the unigram distribution of all given , or a for .

Let us also define a function bigram_count() that takes a file path and returns a dictionary with all bigrams and their counts in the file as keys and values, respectively:

L1: Import the class from the package.
L2: import the Bigram from the package.
L5: Create a defaultdict with Counters as default values to store bigram frequencies.

We then define a function bigram_estimation() that takes a file path and returns a dictionary with bigrams and their probabilities as keys and values, respectively:

L8: Calculate the total count of all bigrams with the same previous word.
L9: Calculate and store the probabilities of each current word given the previous word.

Finally, let us define a function test_bigram() that takes a file path and an estimator function, and test bigram_estimation() with the text file:

L2: Call the bigram_estimation() function with the text file and store the result.
L5: Create a bigram list given the previous word, sorted by probability in descending order.
L6: Iterate through the top 10 bigrams with the highest probabilities for the previous word.

Q3: What NLP tasks can benefit from bigram estimation over unigram estimation?

References

Source: .
, Speech and Language Processing (3rd ed. draft), Jurafsky and Martin.

Maximum Likelihood Estimation

Maximum likelihood estimation (MLE) is a statistical method used to estimate the parameters of a probability distribution based on observed data. MLE aims to find the values of the model's parameters that make the observed data most probable under the assumed statistical model.

In the previous section, you have already used MLE to estimate unigram and bigram probabilities. In this section, we will apply MLE to estimate sequence probabilities.

Sequence Probability

Let us examine a model that takes a sequence of words and generates the next word. Given a word sequence "I am a", the model aims to predict the most likely next word by estimating the probabilities associated with potential continuations, such as "I am a student" or "I'm a teacher," and selecting the one with the highest probability.

The conditional probability of the word "student" occurring after the word sequence "I am a" can be estimated as follows:

The joint probability of the word sequence "I am a student" can be measured as follows:

Counting the occurrences of n-grams, especially when n can be indefinitely long, is neither practical nor effective, even with a vast corpus. In practice, we address this challenge by employing two techniques: and .

Q7: What are the key differences between conditional and joint probabilities in sequence modeling, and how are they practically applied?

Chain Rule

By applying the chain rule, the above joint probability can be decomposed into:

Thus, the probability of any given word sequence can be measured as:

The chain rule effectively decomposes the original problem into subproblems; however, it does not resolve the issue because measuring is as challenging as measuring .

Markov Assumption

The Markov assumption (aka. Markov property) states that the future state of a system depends only on its present state and is independent of its past states, given its present state. In the context of language modeling, it implies that the next word generated by the model should depend solely on the current word. This assumption dramatically simplifies the chain rule mentioned above:

The joint probability can now be measured by the product of unigram and bigram probabilities.

Q8: How do the Chain Rule and Markov Assumption simplify the estimation of sequence probability?

Initial Word Probability

Let us consider the unigram probabilities of and . In general, "the" appears more frequently than "The" such that:

Let be an artificial token indicating the beginning of the text. We can then measure the bigram probabilities of "the" and "The" appearing as the initial words of the text, denoted as and , respectively. Since the first letter of the initial word in formal English writing is conventionally capitalized, it is likely that:

This is not necessarily true if the model is trained on informal writings, such as social media data, where conventional capitalization is often neglected.

Thus, to predict a more probable initial word, it is better to consider the bigram probability rather than the unigram probability when measuring sequence probability:

This enhancement allows us to elaborate the sequence probability as a simple product of bigram probabilities:

The multiplication of numerous probabilities can often be computationally infeasible due to slow processing and the potential for decimal points to exceed system limits. In practice, logarithmic probabilities are calculated instead:

Q9: Is it worth considering the end of the text by introducing another artificial token, , to improve last-word prediction by multiplying the above product with ?

References

, Wikipedia

Homework

HW2: Language Models

Task 1: Bigram Modeling

Your goal is to build a bigram model using (1) Laplace smoothing with normalization and (2) initial word probabilities by adding the artificial token $w_0$ at the beginning of every sentence.

Implementation

Create a file in the directory.
Define a function named bigram_model() that takes a file path pointing to the text file, and returns a dictionary of bigram probabilities estimated in the text file.
Use the following constants to indicate the unknown and initial probabilities:

Notes

Test your model using .
Each line should be treated independently for bigram counting such that the INIT token should precede the first word of each line.
Use such that all probabilities must sum to 1.0 for any given previous word.

Task 2: Sequence Generation

Your goal is to write a function that takes a word and generates a sequence that includes the input as the initial word.

Implementation

Under , define a function named sequence_generator() that takes the following parameters:

A bigram model (the resulting dictionary of Task 1)
The initial word (the first word to appear in the sequence)
The length of the sequence (the number of tokens in the sequence)

This function aims to generate a sequence of tokens that adheres to the following criteria:

It must have the precise number of tokens as specified.
Not more than 20% of the tokens can be punctuation. For instance, if the sequence length is 20, a maximum of 4 punctuation tokens are permitted within the sequence. Use floor of 20% (e.g., if the sequence length is 21, a maximum of puncuation tokens are permitted).
Excluding punctuation, there should be no redundant tokens in the sequence.

The goal of this task is not to discover a sequence that maximizes the overall , but rather to optimize individual bigram probabilities. Hence, it entails a greedy search approach rather than an exhaustive one. Given the input word , a potential strategy is as follows:

Identify the next word where the bigram probability is maximized.
If fulfills all the stipulated conditions, include it in the sequence and proceed. Otherwise, search for the next word whose bigram probability is the second highest. Repeat this process until you encounter a word that meets all the specified conditions.
Make and repeat the #1 until you reach the specific sequence length.

Finally, the function returns a tuple comprising the following two elements:

The list of tokens in the sequence
The log-likelihood estimating the sequence probability using the bigram model. Use the logarithmic function to the base , provided as the math.log() function in Python.

Extra Credit

Create a function called sequence_generator_plus() that takes the same input parameters as the existing sequence_generator() function. This new function should generate sequences with higher probability scores and better semantic coherence compared to the original implementation.

Submission

Commit and push the language_models.py file to your GitHub repository.

Rubric

Task 1: Bigram Modeling (5 points)
Task 2: Sequence Generator (4.6 points), Extra Credit (2 points)
Concept Quiz (2.4 points)

Projects

Speed Dating

During this class hour, you will have the opportunity to meet your classmates in a series of brief, focused conversations. The goal is to identify potential teammates whose research interests and skills complement your own. This is not about finding your exact clone, but rather discovering peers whose strengths and passions could contribute to a well-rounded and dynamic project team.

Be concise, be curious, and most importantly, be yourself. After this session, you will have 2.5 weeks to form a team for collaboration on a team project.

Process

You will participate in 7 rounds of speed dating, with each round lasting 8 minutes and 2-minute transition time between rounds.
During each round:
- Form a group of three people with classmates.
- Each person gets 2 minutes to present and answer questions (6 minutes total).
After each round, split up and form new groups with people you have npt met yet.
By the end of the session, you will have met approximately 14 different classmates (about 25% of the class), helping you identify potential teammates that complement your skills and interests.

Make the most of each interaction by being concise about your interests and skills while asking thoughtful questions to understand your classmates' strengths and working styles.

Proposal Pitch

Each team will deliver an 8-minute presentation pitching their project proposal to the class. This presentation should effectively communicate your project's vision, methodology, and potential impact while demonstrating its feasibility within the semester timeframe.

Requirements

Duration: 8 minutes (strict time limit).
All team members should contribute to the presentation.
Content should cover key aspects of your written proposal, including:
- Project overview and objectives
- Problem statement and motivation

Submission

Submit your presentation slides in PDF to Canvas.

Rubric

Content & Technical Merit (1 point):
- Demonstrates clear understanding of the technical concepts and methodologies involved.
- Presents a well-researched problem statement with relevant background information.

Project Ideas (2024)

Spring 2024

Mara Adams

My vision is a model which could reference the major (at least American, but potentially other English) style guides (the AP Stylebook, Chicago, MLA, etc.) and, given a sentence with an point of ambiguous grammar/style, could give the solution according the different major style guides. I'm not married to this idea per se, but I like the idea of working with style guides in some way.

Nate Adams

I'd like to try and build a sentiment analysis tool, capable of classifying emotions into positive, neutral, or negative (or perhaps give a numerical rating). If possible, the model can be expanded to include other emotions like happiness, anger, disappointment, etc.

Calvin Brauer

Looking at words used on social media and comparing sentiments across different social platforms.

Wenxuan Cai

For the team project, I want to use NLP skills to catch keywords in some math papers. From my own experience, math research papers are hard to understand due to their nouns and abstract ideas. If we can get some key things out from the math paper, this will be helpful.

Michael Cao

N/A

Alec Chapman

For the team project at the end of the semester, I would like to build a Sentiment Analysis model for Cryptocurrency Trading. In the world of Crypto, many of the price movements are caused by sudden changes in the sentiment of investors. This often can be found on social media sites like Reddit and Twitter, where users often post their feelings about the currencies. A few years ago, a subreddit called WallStreetBets blew up for causing an insane price increase in Gamestop stock as well as some others. From this, I discovered the power of using Sentiment Analysis modeling on various social media sites to attempt to predict price movements of online currencies and stocks. For this project, I would like to analyze user posts on these social media sites to calculate the overall sentiment for specific Cryptocurrencies. I will then use this data to predict the incoming price changes that may occur.

Marcus Cheema

My project idea involves training a large language model on countless recipes, found online, which must be preprocessed accordingly to be usable. Once the chatbot is trained, a simple web platform would be used for accessibility and testing. This platform would be comprised of chats, and an area to enter text. The goal is to make a chatbot capable of creating usable recipes based on user inputs, dietary restrictions, and ingredient restrictions. I thought of this project idea with Dylan Parker.

Murphy Chen

Concept and Idea: Spam Detection Bot for Email Systems The potential concept for our team project is to develop an advanced Spam Detection Bot that is specifically designed for email systems.

Intelligent Linguistic Analysis: The bot will utilize natural language processing (NLP) techniques to analyze the content of emails. It will focus on identifying linguistic patterns, keywords, and phrase structures commonly associated with spam.

Machine Learning Integration: By employing machine learning algorithms, the bot will be able to learn and adapt to evolving spam tactics. This continuous learning approach ensures that the bot remains effective even as spammers change their strategies.

Feedback Mechanism: The bot allows the users to manually mark certain email as spam or not, which allows the bot to update itself to increase the accuracy of detection and reduce the possibility of false detection.

Real-time Processing and Efficiency: The bot is designed to process emails in real-time, ensuring that users' inboxes are promptly cleared of spam.

Security and Privacy Focus: In addition to spam detection, the bot will prioritize user security and privacy, ensuring that the user’s privacy is always the top priority.

Andrew Chung

I would like to make a model that can read influential people's tweets in the financial world to get some sort of sentiment, and determine if someoen should invest in that stock or not.

ex. If Warren Buffett were to tweet out "Apple is a terrible company", the model would be able to detect the sentiment of Apple as negative, and therefore not invest in the stock.

Henry Dierkes

I want to create a language identifier that can output the name of a language based on a text of a certain language as the input. To make this work I believe that we would first identify the types of characters used and narrow down the languages based on that. Secondly, we would tokenize the text into words and compare it to the dictionaries of each language we are still considering. Lastly, we would look at how many matches there are between the text and the dictionaries and output the most similar one. Maybe if it's below a certain threshold, the output could say that there's not enough data to suggest that the language is one we have access to.

Tung Dinh

For the project, I'm thinking about an AI diary app using GPT. This app will let students write about their day, and the AI will offer encouraging words and advice. It could also detect if the student is stressed and help them as a friend. The goal is to create a comforting space for students to reflect and relax.

Benjamin Dixon

Project Idea I would like to study the difference between the usage of “(disability-adj) person” and “person with (disability)” in the context of academic papers. For example, there is lots of discourse on the difference between saying autistic person or person with autism, and from my own experience on reading research papers about autism, both are used frequently. The project could involve gathering data about its usage frequency in scientific papers between academic fields (medical, psychology, sociology, anthropology, etc.), within the same papers (do research papers always use the same term, or do some use both?), by date (is there a difference in pre- and post- 2015, or some other relevant date?), or some other metric. Further work could be done with sentiment analysis technology to see if the papers use language that is favorable / amiable towards disability or disparaging towards disability, and could be correlated with their chosen phrase to describe disability (disabled person vs person with disability) to see if there is a significant difference.

Lindsay Esterman

I just switched into this class and haven't attended any lectures yet so I don't have a good concept of what a good project is, but an idea is to analyze tik toks to gauge that population's opinions on the 2024 election.

Calla Gong

I have an plan to do a text analyze on the answer in QUORA to distinguish answer from experts and other participants.

Hunter Grimes

One idea for a project which I might like to pursue would be an AI coach for video games. The idea would be to have the coach look at in game performance, potentially in real time, and give coaching about how to improve ala "pay more attention to objectives".

Frederic Dean Michae Guintu

Something that may be challenging to do as a team project is to look at computational linguistics based on languages other than English. An alternative that would fall within the scope of the class while also looking at nonstandard English would be analyzing various English dialects, with projects such as dialect identification, translation between dialects, or even a chatbot that understands multiple English dialects.

Molly Han

For the group project, I am interested in sentiment analysis, specifically concerning evaluations of business products, or social media posts. Additionally, I am also inclined towards topics associated with the detection of spam emails or messages.

Paige Hendricks

I am fairly open to the type of team project I would like to pursue this semester. Things such as sentiment analysis, story generation, and chatbots would all be interesting to me. I am not sure about the feasability, but the project I would be most interested in working on would be a text to speech tool. I would like to learn how speech generation works on such a level, and it is a tool I could imagine using once created. I know that this type of tool already exists, so I would ideally like to work on a less prevalent aspect such as adding emotionally expressive speech so that it doesn't sound as flat. I am also interested in the ways that characters or celebrities voices can be emulated, but I am not sure if this would have any legal hurdles.

Jerry Hong

N/A

Peter Jeong

I have thought of working on a sentiment analysis model which would classify customer review we see on e-commerce platform(whether they are positive, negative, or neutral). However, I also want to listen to other ideas so I am also open to new ideas.

Helen Jin

Project Idea: Using AI for recipe recommendation, modification, and generation. Opening the fridge often reveals leftover ingredients from previous meals, which can make deciding what to cook a challenge. Typically, we end up spending considerable time looking for recipes that match our tastes and the ingredients we currently have. It’s common to find that we’re missing a few items for the recipes we’re interested in, which can be quite inconvenient. However, by incorporating artificial intelligence, we could have a tool that acts like a consultative chef. This AI system would allow us to tailor our recipe searches more precisely, adjusting recipes to fit our personal requirements, like creating a low-calorie version of a certain dish or suggesting substitutions for missing ingredients.

Carl Kassabian

I am very unsure of what kind of project I would like to pursue. I think combining linguistics with visual data could be interesting though. Maybe some kind of coding that sorts or filters certain types of words or phrases, and then those are represented in some visually pleasing way. I think some kind of code that analyzes literature at a highly specific linguistic level could be very interesting. I am extremely flexible and would be willing to work on almost any project.

Michelle Kim

A possible project could be to use NLP techniques on unconventional data. Personally, I am interested in cross-cultural linguistics / multilingualism such as heritage language usage in diasporic settings. Some possible datasets could be English loan word usage in a non-English speaking countries (ie. in everyday life, music lyrics, social media, websites). For example, in South Korea and many other countries, popular music integrates English into various text such as music lyrics, advertisements and marketing, and everyday speech (based on age group). Another possible topic could be topics such as correcting gendered language to non-gendered language. Lastly, academic related topics: I think a fun project could be something like predicting final course grade based on a student writing sample, predicting / generating potential test questions based on a text, or predict / generate weak areas of students based on their code sample.

Will Kohn

A text summarization tool for simplifying complex readings for classes.

Andrew Lee

I have a group that I believe we will be working together, yet we have not yet decided on a topic for a project. An idea I have been thinking about is incorporating a character into game that will take the language input that the player puts in. After analyzing the type of writing that the players uses, the character will respond in the same way the player wrote. For example, if someone is using Shakespearean language, then the character will respond back the same way.

Sam Liu

Project Idea: Explain Attention Is All You Need to children: Design a system that explains and summarizes academic papers in a more comprehensive way, especially for those who do not have much background knowledge. It can reliably lay out the fundamental information from abstracts, introductions, methods, and findings from any academic paper without missing vital information, allowing readers to process the main ideas efficiently.

Andrew Lu

Though I still don't know much about NLP, for my team project, I think it would be interesting to try to work on a language model that is trained and works solely on inputs with perfect grammar in an attempt to see the effect of input "sanitization" on performance and model size.

Louis Lu

I'd like to apply NLP algorithms and some machine learning algorithms on a public available dataset to perform supervised classification task. For example, applying MLP on product review to distinguish helpful reviews against unhelpful ones. I would like to further compare and evaluate the performance of some large language models, such as Bert and GPT4 API.

Wenzhuo Ma

My idea is to make an "PolyGlotBot" which is a multilingual virtual assistant that helps Chinese learners practice and improve their skills. The model will give real-time feedback on grammar, pronunciation, and vocabulary when having the conversation with it. It will also have the interior function which adapts the user's proficiency level and personalizes learning content. This can help leaners learn better on his own rythm and on his own level.

Ja'Zmin McKeel

I'm not sure but I am open to exploring

Izana Melese

Project Idea: Create a tool that can figure out how people feel when they post on social media, no matter what language they're using. The idea is to make a system that helps us see the emotions behind online conversations in different languages, so users and businesses can get a sense of what people are expressing on a global scale.

Ruichen Ni

I would like to learn how could NLP techniques be applied to alternative data in finance and business.

Ellie Paek

Utilization of sentiment analysis would be something I'd be interested in working on; a project like conducting sentiment analysis on a dataset (such as poems) and using it to generate new data (such as new poems based on a key emotion) is something that I would be interested in. In addition, I would also be interested in leveraging LLMs to create something, such as recreating the personality of characters.

Dylan Parker

The team project that I have in mind is an LLM that recommends personalized recipes based on user requests, dietary restrictions, and ingredient availability. The system could also assist users in meal planning by suggesting balanced meal combinations and creating shopping lists if they do not have the necessary ingredients. My team would ideally collect many recipes online for our database to serve as the foundation for our recommendations. An algorithm would need to be developed for the meal recommendation system. We would then create a web-based application for user interactions with our system. If possible, it would be great to have a user "account" feature for further user personalization in the future. I thought of this project idea with Marcus Cheema.

Sherry Rui

Text Writing Editor: Use models like OpenAI's Davinci API for generating creative writing, including poetry, stories, or even scripts. The focus would be on fine-tuning the model's parameters and prompts to allow certain styles or themes.

Chengyu Shi

GitHub repository link: Team project: I'd like to know more about sentiment/emotion/opinion analysis of text. This semester I'd probably do something related to analysis techniques regarding sentiment lexicons and sentiment classification models.

Dennis Sun

N/A

Nicole Thomas

N/A

Michael Wang

The team project that we would like to pursue is to use ai (chatGPT) to train specifically for a task, such as writing, research, programming. For example, there is a plugin function in Canva supported by ChatGPT that is able to generate infographics and visualizations automatically through prompts.

Yoyo Wang

N/A

Lindsey Wendkos

I am interested in doing a project about sentiment analysis, where a model would be able to decide what the tone of a piece of text is. This is intriguing to me because even as people, tone is difficult to convey accurately through text, and even when we are able to determine the sentiment of a piece of text, it is not always clear exactly how that tone was communicated. I am curious to see how accurate a machine can be in determining something so emotion-based and not clear-cut. With that said, I am open to other project ideas as well!

Cashin Woo

N/A

Freddy Xiong

Two idea possibilities: 1)Sentiment Analysis Chatbot: A chatbot that detects the emotional and state of mind-being from a person, the idea behind this we are LARPing as Walmart or Target or another retailer, and we want to detect how the customer feels without having them directly state it (because that tends to anger them). Put it like this, if you're disappointed with either customer service or a product you just purchased, and you write to a bot wanting to express disappointment, but is instead asked how you feel about the service or product, that will just annoy you into stating you are angry or upset. The key to this tool is being able to interpret word choices and convert them into state of emotional well-being and satisfaction from a state of 1-5, 1 being extremely unsatisfactory, and 5 being extremely satisfactory, and utilizing this metric across 5 rubrics (Customer Service, Product Quality, Cleanliness of Stores, Location Convenience, Feeling of Safety) from a short conversation with the consumer. Along the way, the bot can also make recommendations, including for products and advices, after the conversation. 2)Spam Detection Bot: Email filter bot that would combine ANN with corpus of common spam emails, including ones I've fallen for (I've failed 100% of the Emory phishing emails, I'm sad to say, not an exaggerated stat, I've never not clicked on those baits). We'd create a distinction between actual legitimate bot emails (i.e. Job offers, important notifications, reminders) from both harmless and harmful spam. Differs from email spam on multi-lingual factor: my emails still have large amounts of non-english spam that filter through, but because the spam bot isn't as well-trained in that factor, I'm receiving garbage on fake chinese job referrals and german job offers (I know they're fake, because like the genius that I am, clicked on them and inquired, which resulted in me getting more spam emails). The spam detection bot would also create warnings on non-spam emails that border on spam (i.e. promotional emails), and offer the end user the opportunity to enable options to filter them out.

Simon Yu

Team Project Idea: Examine presidential inaugural addresses for in word tokens, types, diversity, etc. More points of analysis will definitely come up, but in general hope to look for different trends over time, individual differences between presidents, and other interesting observations.

Yunnie Yu

Our team would like to develop a system that can analyze and classify the sentiment of social media posts. We will choose one social media platform and focus on posts from a certain period or around a specific topic. Our goal is that the system can help businesses, organizations, and governments understand the public reaction and adjust policies/improve products.

Jingxuan Zhang

N/A

Joyce Zhang

Project Idea: I aim to analyze the linguistic nuances and sentiment differences in academic papers when referring to "people with disabilities" vs. "disabled people". I hope to understand if the choice of terminology correlates with varying sentiments and contextual frameworks in disability discourse within academic literature.

Chenming Zhou

N/A

Serena Zhou

Project Ideas (2024)

Spring 2024

Mara Adams

Nate Adams

Calvin Brauer

Looking at words used on social media and comparing sentiments across different social platforms.

Wenxuan Cai

Michael Cao

N/A

Alec Chapman

Marcus Cheema

Murphy Chen

Concept and Idea: Spam Detection Bot for Email Systems The potential concept for our team project is to develop an advanced Spam Detection Bot that is specifically designed for email systems.

Real-time Processing and Efficiency: The bot is designed to process emails in real-time, ensuring that users' inboxes are promptly cleared of spam.

Security and Privacy Focus: In addition to spam detection, the bot will prioritize user security and privacy, ensuring that the user’s privacy is always the top priority.

Andrew Chung

I would like to make a model that can read influential people's tweets in the financial world to get some sort of sentiment, and determine if someoen should invest in that stock or not.

ex. If Warren Buffett were to tweet out "Apple is a terrible company", the model would be able to detect the sentiment of Apple as negative, and therefore not invest in the stock.

Henry Dierkes

Tung Dinh

Benjamin Dixon

Lindsay Esterman

Calla Gong

I have an plan to do a text analyze on the answer in QUORA to distinguish answer from experts and other participants.

Hunter Grimes

Frederic Dean Michae Guintu

Molly Han

Paige Hendricks

Jerry Hong

N/A

Peter Jeong

Helen Jin

Carl Kassabian

Michelle Kim

Will Kohn

A text summarization tool for simplifying complex readings for classes.

Andrew Lee

Sam Liu

Andrew Lu

Louis Lu

Wenzhuo Ma

Ja'Zmin McKeel

I'm not sure but I am open to exploring

Izana Melese

Ruichen Ni

I would like to learn how could NLP techniques be applied to alternative data in finance and business.

Ellie Paek

Dylan Parker

Sherry Rui

Chengyu Shi

Dennis Sun

N/A

Nicole Thomas

N/A

Michael Wang

Yoyo Wang

N/A

Lindsey Wendkos

Cashin Woo

N/A

Freddy Xiong

Simon Yu

Yunnie Yu

Jingxuan Zhang

N/A

Joyce Zhang

Chenming Zhou

N/A

Serena Zhou

Project Ideas

Spring 2025

Project Groups by Theme

A. Career Development & Professional Tools

Resume & Cover Letter Analysis

Amoakohene, Humphrey: NLP system to analyze and rank resumes based on job descriptions
Kansal, Rhea: System to extract key skills from resumes and match with job descriptions
Wen, Yuanhuizi: AI-powered resume-to-job matching system using neural networks

Job Application Tracking

Arshavsky, Mark: Email-based internship application tracking system

B. Content Analysis & Verification

Media Bias & Fake News Detection

Dattatreya, Maya: Sentiment analysis for detecting tone and bias in media
Jerry Xing: Program to evaluate media bias levels
Kansal, Rhea: Fake news detection system

Academic & Research Analysis

Ahn, Eric: Academic paper simplifier for general audiences
Xu, Jack: Sentiment analysis tool for evaluating research paper quality
Xinyuan, Hu: Tool to evaluate research paper credibility

Review & Sentiment Analysis

Chen, Angelina: Bot detection in Amazon product reviews
Coulanges, Charlington: Sentiment analysis chatbot for platform reviews
Gyul Kim: Analysis of healthcare system and doctor reviews

C. Content Generation & Processing

Text Simplification & Accessibility

Pham, Chloe: Translation of academic text to accessible language
Yirdaw, Elshaday: Readability analysis tool with improvement suggestions
Yuxuan, Shi: Medical text simplification for patient education

Creative Writing & Story Generation

Zhang, Jingzhi: Interactive life story generator
Wu, Junting: Poetry style analysis and classification
Sixing, Wu: Context-aware reply generator for various scenarios

D. Educational Tools

Course Selection & Academic Planning

Hu, Yutong & Shah, Jiya: Natural language-based course recommendation system

Study Aids

Jarman, Robert: Pocket Teaching Assistant for lecture note processing
Yirdaw, Elshaday: Vocabulary building tool

E. Language & Translation

Language Understanding

Chang, Ridge: Idiomatic expression translator across languages
Dahiya, Rishita: Context-based word definition tool
Kim, Olivia: Hand gesture and NLP integration system

F. Content Classification

Genre & Style Analysis

Correa-Perez, Nicholas: Music genre and lyrics analysis
Jennings, Caleb: Author style prediction model
Jitngamplang, Varakorn: Music mood and theme analyzer

G. Code Analysis & Documentation

Code Quality

Carrier, Emma: Code syntax analysis and improvement tool
Shah, Riyan: Automated code documentation generator

H. Data Processing & Analysis

Financial Analysis

Jiang, Jenny: Financial news summarizer
Reicin, Noah: Financial document analyzer
Suh, EunGyul: Accounting fraud detector

Data Cleaning & Structure

Guan, Yifei: Police report information extractor
Yang, Junhyeok: Web-scraped data cleaner for LLMs
Tolmochow, Gregory: Natural language query interface for structured data

I. Health & Fitness

Fitness Planning

Chen, Xinmo: Personalized fitness recommendation system
Arularasu, Akhil: AI-powered meal planning assistant
Jarman, Robert: Gym companion app

J. AI & Machine Learning Research

AI Development

Gao Henry: AI text detection and human-like text generation
Jung, Seungwon: Zero-shot vs few-shot learning comparison
Hakam, Brian: AI debate simulator using persona-based LLMs

Detailed Individual Ideas

Ahn, Eric

The team project idea I had in mind was an academic paper simplifier that would aim to bridge the gap between complex academic research and general audiences. Lots of scholarly papers today are filled with technical jargon, which makes them difficult to understand for non-experts. In this way, by using NLTK for text preprocessing followed by NLP (using pre-trained models like GPT or BERT) for summarization to analyze research articles, we could preserve key information and provide plain-language summarizes, highlight essential points, and offer explanations for complex terms. Users, such as students or the general public, can input an academic paper and receive an easy-to-read summary via input from the following potential options: (1) web scraping from URLs (like PubMed), (2) uploading .txt/.docx/.pdf file, (3) and/or direct text input in the UI.

Ambrose, Daniel

Cover letters are definitely the most annoying and time consuming part about applying to jobs. There are methods to make a cover letter quicker like using ChatGPT, but the result is usually too general and recruiters could easily tell it's AI-generated. I want to create a project where you can input a job’s description and your general cover letter and it outputs a tailored cover letter that incorporates keywords from the job description. To create this project, I will need to use a mix of AI-generation and keyword matching through text processing. This project will mostly focus on computer science jobs to limit the variety of keywords. Keywords extracted from job descriptions will prioritize certain groups of keywords such as programming languages and the requirements section in the job description. Additionally, there could be functionality to rate how well the job's keywords match your cover letter's keywords.

Amoakohene, Humphrey

Building an NLP system to analyze and rank resumes based on job descriptions to make it easy to apply to spam apply for jobs without having to manually create new resumes

Arshavsky, Mark

In Data Science and CS, it’s quite normal for students to apply to tons of internships. From personal experience, it gets really hard to keep track of all the applications. Sometimes you might even miss an important status update that gets buried in your email. My project idea is to create a product that simplifies the internship application process for students by handling the “tracking” and “monitoring” step for them. My vision is for it to work by students connecting the product to their emails (as when you apply to an internship you usually get an email), but I am open to other interpretations! Then, the product will use NLP to parse the relevant emails, extracting data such as company name, date of application, position name, etc. and put it in an Excel sheet. The product will also track for updates to students’ applications and color-code accordingly. Perhaps red would be “Rejected”, yellow would be “Status Update”, and green would be “Offered”. I have been thinking of this idea for some time now, and I think it can make a real-world impact!

Arularasu, Akhil

I want to implement an AI-powered meal planning assistant that dynamically generates personalized meal plans based on user inputted dietary preferences, allergies, fitness goals, and daily caloric needs. By processing natural language inputs (eg: "I need a high-protein vegetarian meal plan with 2500 calories per day"), the system will create meals from one or many extensive recipe datasets, ensuring nutritional balance while considering restrictions like keto, vegan, vegetarian, gluten-free, etc.

Key Features include leveraging NLP for text generation, calory and nutrient optimization, dynamic recipe recommendations, grocery list generator, and more. Users can describe their dietary needs in natural language (eg: "low-carb, high-protein meals under 2000 calories"), and the system will parse these requests into structured data to generate meal plans. I feel that there is a lot of places the group can take this idea and run with it while utilizing NLP.

Carrier, Emma

An application that identifies poor coding syntax practices and gives suggestions to fix it. For example, if there isn't any commenting found throughout the code, it will suggest to add more comments in specific places (before functions, etc.). The user would input a file / code chunk to be analyzed and it can either return the code chunk accepting the suggestions or just give a list of suggestions for the user to change themselves. There could be an option for multiple programming languages because they all have different syntax and standards that programmers should follow. Could market it as a way to help new programmers or programmers learning a new language as a way to help them identify areas in which their code could improve. Would likely need to implement a parser to achieve this goal.

Chang, Ridge

I want to build a system that translates idiomatic expressions across languages in a way that actually makes sense, rather than just converting words literally. Expressions like “break the ice” don’t mean much if translated directly, but most language has its own way of saying the same thing or similar things. I want to create a tool that captures these nuances, ensuring that meaning—not just words—is preserved to the best of its ability. I’m interested in how idioms reflect cultural perspectives, and I want to explore how this project can improve communication between languages while keeping translations natural, authentic, and pragmatic.

Chen, Angelina

I think it would be interesting to look into Amazon product reviews and use NLP to detect whether a bot is responding or an actual human; this is due to the increased number of bot action on Amazon. Often bots will give fake positive reviews to help boost a particular item or to show that a user is active if they are a paid reviewer. Another idea is to analyze the most popular companies and the kinds of words that are used in advertisements/website/customer interaction to understand how they have good customer retention. There could be a analysis of popular and dying companies.

Chen, Xinmo

I want to analyze user input and provide personalized fitness recommendations. It will allow users to log their workouts in natural language, such as "I lifted 50 lbs for 3 sets of 8 reps today," and extract key data points using NLP. By analyzing past workout logs, the system will identify trends, detect plateaus, and suggest progressive overload strategies, such as increasing weight, reps, or intensity. Additionally, I also want to do/incorporate sentiment analysis to give user motivation and provide encouraging feedback. This model will probabaly rely on outside fitness-related datasets and integrating them in order to offer actionable insights for optimizing workout routines.

Correa-Perez, Nicholas

My idea for the team project would be an analysis between lyrics and music genre. It would take a list of songs from a list of different genres and see if common themes or even words come up in the search. Perhaps maybe even an ambitious goal could be to see if we can tie the type of music to its words (i.e. bubbly music has positive words). Genre would be described as the musical aspects of songs to have a consistent definition, and the lyrics would be analysis for overall theme and tone of the lyrics as a whole.

Coulanges, Charlington

I would like to build a sentiment analysis chatbot to distinguish reviews on given platforms (Yelp, Amazon, Rotten Tomatoes, etc.). By giving the data either a positive, neutral, or negative label, the majority of the labels will either recommend or not recommend the product using excerpts from the reviews analyzed. There are often times when a review site will have a decent amount of bot reviews to either tank or boost the rating. Thus, the chatbot will also be able to detect whether any review appears to be written by bots or a any non-human source.

Dahiya, Rishita

I've often been fascinated by the intricacies of the English language which may not be intuitive if English is not someone's first language. An example of such is how certain words change in definition by their context, for example the word "set", which means something different in each of these phrases: set the table, the sun set, to set the record, and so on. My goal in this project would be to develop a tool which is able to decipher the definition of the ambiguous words depending on the context it is used and as a result make machine language processing more precise and intuitive. This would make machines understand the human language better and would have practical applications in machine translations and search engines.

Dattatreya, Maya

My project idea would be to create a model that can perform sentiment analysis to detect the tone and bias in media and the news. I want to see how levels of emotional language used in different kinds of articles affect the engagement metrics of popular media and news sources. My goal is to understand how sentiment influences public perception and how media literacy can be improved.

Forbes, Alexander

I get spam emails daily, and I think it'd be an interesting final project to create a straightforward spam detection tool that flags suspicious emails or messages by using NLP techniques to identify spammy keywords, suspicious links, or recurring patterns. A feedback loop will let users confirm or override the model’s decisions, which could potentially improve accuracy over time. This project would combine classification algorithms with a user-friendly interface for simple adoption. Ultimately, I want to produce a packaged, practical tool that addresses an actual, real-world communication problem.

Gelb, Jake

Hi! I am open to all sorts of various projects and am not attached to any specific idea. Some areas of interest include spam bot detection, phishing and scam prevention, recipes tailored to dietary needs, sentiment analysis, and chat bot development. One specific idea is aa chatbot geared toward fantasy football users that analyzes trades and decision-making, helping users optimize their teams based on player performance, matchups, and trends. I am very excited for this course and am eager to collaborate on some impactful project.

Guan, Yifei

I'm currently working with a professor to aggregate data from the police department. Some department has organized data that we can use directly, but most of them only have individual reports that we have to fetch information manually. I want to use linguistic model and text processing technique to extract information efficiently from individual police report, including incident, date, victims, information officer, information, and so on. This would largely shorten the time we need on processing the reports.

Hakam, Brian

AI Debate Simulator-Using persona based LLMs(Similar to Character AI) This project would fine tune LLMs to answer like specific people, using scraped transcripts and a tone analysis of the person(funny, serious). Then attach the person’s history and debate relevant events to the context window. For example, to do Abraham Lincoln, an LLM would first be fine tuned on speech of that time, then specifically tuning on his transcripts and text. Then another LLM would be trained in this fashion on another person. The 2 or more LLMs would then debate a chosen topic with an intermediately and neutral LLM asking questions, then ultimately deciding the winner. This process would be fully automated, where you type in a name and it would web scrape and train all on its own.

Henry, Gao

I have two ideas:

Create an AI model that will identify text written by AI. It would be difficult because AIs are becoming more and more human-like. Some ideas would be to look for some “errors” that AIs won't make, or check the cosine similarities between sentences. The “errors” should be something that is grammatically correct but not often used. By checking cosine similarities, we can see if the sentences use similar words, or have similar structures.
Use human-written text to train an AGI agent that writes like a real human. Instead of always choosing the token with the highest probability, choose a random token based on the probability assigned. In this way AI will have a larger vocabulary set. Another possible approach is to analyze the sentence structures and let each sentence structure appear in a probability similar to that in reality.

Henry Gao

Hou, Carol

I'm interested in a project focusing on sentiment analysis or an email spam detector. I am interested in researching machine learning models that can be used for the classification tasks (classifying a social media post as positive, negative, or neutral, or for emails as spam or not spam.) I haven't narrowed down the main social media website I want to use as a data source for sentiment analysis, but I'm currently thinking of X (twitter), Yelp, etc. The goal of the project would be to determine what customers are thinking about certain products, restaurants, etc. Or, to detect if an email is spam or not.

Hu, Yutong

I would like to build a natural language-based course recommendation system that helps students find relevant courses based on their preferences and academic history. The system will process natural language inputs from users to understand their course requirements and generate personalized recommendations. The system will accept free-form text descriptions of desired courses, parsing key information such as: subject area/discipline, course topics and content, instructor preferences, credit hour requirements, and etc. Besides, we would like to add more advanced filtering based on students' personal information: their majors and the course they've already taken. We might need to use named entity recognition and some filtering engine.

Jarman, Robert

I have a few ideas.

Online Multiplayer Game: Create a simple multiplayer game (e.g., trivia like charades, word games like Wordle, or strategy-based optimization such as collecting coins on a map within a limited number of steps or time). The game would feature real-time interaction and leaderboards to engage players.

Pocket Teaching Assistant (Pocket-TA): Develop an app that uses natural language processing to summarize lecture notes, create flashcards, organize notes collaboratively in real-time (with features like version history, tags, and search), and answer questions from uploaded documents. This tool aims to make studying for exams more efficient.

Gym Companion App: Build an app to personalize fitness journeys with tailored exercise, diet, and sleep plans based on a user’s current build and desired physique. The app also fosters community by helping users find nearby sports partners for games like basketball or tennis, promoting health and social connection.

Jennings, Caleb

Create a model using machine learning combined with computational linguistics that learns the writing style of a list of classic authors from a database of many classic literature books, and the model tries to predict an author based on a sample text that was left out of the database for testing. Finding the database should be easy since much of classic literature is public domain (though it would take some time to collect it all). A number of machine learning techniques could be applied such as Stochastic Gradient Descent, Random Forests, Neural Networks, and Clustering. The challenge would be to determine how to parse the data, what learning technique to use, and how to use the data in learning the model.

Jiang, Jenny

An idea I had using NLP is a tool that processes hundreds of news articles that are released daily regarding markets, finance, and the economy (from Wall Street Journal, New York Times, CNBC, MarketWatch, etc.) that then provides a summary of what happened. As someone that likes to keep up with the news and markets myself, it can be exhausting reading multiple articles a day from different platforms to get an accurate grasp of current events. An NLP tool could process all these news articles using big data to then generate daily summaries of the news. This can help make keeping up with the economy more accessible for those that don’t follow the news.

Jitngamplang, Varakorn

I like the idea of a tool that helps find music that precisely matches a desired mood or theme. I want to develop a tool that uses natural language processing techniques to analyze song lyrics and facilitate music discovery based on a specified lyrical characteristics. Ideally, it would searches for songs based on themes (e.g., romance, travel), sentiment (e.g., melancholic, upbeat), and other lyrical features. By extracting keywords, analyzing sentiment, and employing other NLP processes, I hope it will recommend songs that resonate with given preferences. This goes beyond a simple sentiment analysis tools or aggregator as this would require analyzing music rhetoric.

Jung, Seungwon

I am open to any topic to research. I am open to topics like sentiment analysis, news fraud detection, text-to-speech tools that add emotional expressiveness, and even poetry-generating writing editor. Also, research related to bias detection and ethical language generation can be done. My idea at the moment is to research the effectiveness of zero-shot vs. few-shot learning on dynamic gaming(or just a normal life if sufficient datasets are not found) NPCs. Although it is well-known few-shot learning will be more effective than zero-shot, I would like to research its limitations using BLEU, ROUGE, or BERTScore to evaluate the quality of a generated text using few-shot learning.

Kansal, Rhea

I am open to any project ideas, but one that I think could be interesting is using NLP to detect fake news. In this project, we would develop a system that can classify news articles or social media posts as either fake or real. The system could analyze linguistic patterns, stylistic features, and content to identify misinformation and provide users with a confidence score for the classification. The prevalence of misinformation in today’s society makes this a highly relevant topic.

Another idea I have is a system that identifies and extracts key skills, qualifications, and experiences from resumes. This tool can be used to match resumes with specific job descriptions, streamlining the recruitment process. It could also rank resumes based on how closely they align with the job requirements, helping recruiters quickly find the best candidates.

Kim, Gyul

I would like to analyze patient reviews on healthcare systems and doctors to identify key topics such as wait time, doctor empathy, quality of care, misdiagnosis, and facility conditions. Through topic modeling, the project can identify recurring themes to provide a summarized review on a facility, service, doctor/ any other patient feedback. The project will involve collecting and preprocessing patient reviews from platforms such as Healthgrades, Zocdoc, or Reddit, followed by implementing unsupervised learning models to extract meaningful topics. Sentiment analysis can also be integrated to assess whether patient sentiments about specific topics are positive, negative, or neutral.

Kim, Olivia

Embedding hand gestures with NLP. As LLMs expand out to voice recognition, there has been significant more researches about tone, but to my short knowledge hand gesture research hasn't been done as extensively. By recording the hand movements made when speaking a certain text and analyzing it, nuances will be able to be captured more effectively. This could also be implemented to turn a written text into ai-generated videos of sign language. However, I strongly believe this is too good of an idea to not have been made already.

King, Marisa

I would enjoy developing a project that examines the different words and terms used by the media to describe events (both global and internal), and how their specific word choices differ based on factors such as race, gender, sexuality, politics, religion, etc.

Lin, Ryan

My team project concept is a some kind of text-message/DM chatbot that can mimic slang or like styles of texting. People speak differently over text whether in content, style, or length of messages. I think its pretty interesting how language over text can be so different than using proper grammar, and in a sense it is its own kind of dialect of language. I'm open to doing other stuff cuz I'm also interested in using different types of data including video, audio, and picture data but I'm not really certain how yet.

Mi, Renzhi

I'm thinking about developing a program that classifies books and novels into their respective genres based on word frequency patterns and word types, which can potentially help to build the page suggestions for readers. This might involve processing volumes of text, extracting frequently occurring terms and utilizing them as key indicators to infer the overarching theme and genre. By leveraging NLP techniques, the program will identify genre-specific linguistic patterns and compare them against predefined datasets corresponding to various literary genres such as mystery, fantasy, science fiction, and romance. This automated classification framework enhances genre identification by analyzing textual content directly, so it can provide a more objective and data-driven approach to literary categorization.

Okumura, Harutoshi

Political Science Research Paper on using LLMs to detect explicit contradictions from presidential debate speech candidates.

Can LLM rely solely on Natural Language Structure to find explicit contradictions on candidates' past speeches (public, tweets, websties, e.t.c), and use it against them to solidify an argument of contradiction, and thereby lack of credibility in certain topics?

Are there certain semantical features and law of contradiction, that we can adapt reliably for a consistent system that detects contradictions (no matter how fundamental it may be), and apply to large corpus of past speeches.

Pham, Chloe

I'm potentially interested in the idea of "translating" texts written in a highly elevated/academic/obscure tone to language that may be more accessible to the average person. While there is an argument for "flowery" and "smart" writing, I think there is an elitist undertone to conveying ideas in a way that is inaccessible to many, especially if the text's purpose is education. There is a large set of existing source texts to draw from, along with some existing SparkNotes-type "translations" of classics and other foundational texts.

Reicin, Noah

Develop a Natural Language Processing (NLP) system that analyzes financial documents to perform two main tasks:

Sentiment Analysis: Identify and categorize the document's tone—positive, neutral, or negative—based on relevant keywords' frequency and contextual usage.
Key Focus Extraction: Detect the frequency and distribution of specific domain-relevant terms to determine the company’s primary areas of emphasis (e.g., sustainability initiatives, R&D, and shareholder returns).

Sezgin, Alas

I would like to work on a dogwhistle detector, specifically for Twitter (or, reluctantly, now named "X") as with the new administration of the website, far right rhetoric (like fascism) has become less and less censored and perhaps even more common, albeit hidden behind a veil of dogwhistles and plausible deniability which make it hard for the average person to determine if what they are consuming is a joke or far right rhetoric. This has the effect of familiarizing users to unkowingly become familiarized with harmful rhetoric through humor, the selective lack of censorship towards which can sway user opinions on a supposedly neutral website. This is why it's important to detect dogwhistles or general far right rhetoric, which may be difficult for the average person. It would be nice to implement this as a browser add-on but that might be outside the scope of this class. This would likely involve a lot of hard-coding or feeding labeled data into a model, so I am open to new ideas.

Shah, Jiya

This idea was made in collaboration with Yutong Hu. Although I don’t have any concrete ideas on implementation currently, we were thinking of something that helps students choose courses based on information inputted in natural language. It would use their description of the course (i.e., which academic discipline, professors they’d like, how many hours they want/need, which major/general education requirements they still need to fulfill, general subjects they would like to explore, etc.) in addition to information about what courses they have already taken and their major(s) and/or minor(s). This would all be to provide suggestions of what courses they could take based on the course atlas of that semester.

Shah, Riyan

My idea is to process pieces of code and using an NLP algorithm, generate meaningful comments for it so other people looking at the code can better understand it.

create a list of variables used with their purpose
comment long lines of code to explain what they are trying to do (eg. long mathematical expressions, or lines involving lots of variables and packages)
create comments for blocks of code that are trying to perform a task (eg. this for loop is meant to do blah blah blah)

Sixing, Wu

My idea is to make a Brilliant Reply model. I hope the model can work to assist people in replying in various scenarios. 1. email writing assistant: helps to formulate emails for different occasions, including networking, job applications, announcements, etc.); 2. text message reply: helps to generate an appropriate and engaging response for personal and business usage, being able to handle slangs and emojis.); 3. scenarios based replies: eg: helps to generate a hook-up message or flirting replies in ins, a coffee chat invitation in LinkedIn, appropriate complaints message to customers services, etc; 4. tries to tailor the users' tone and learn from the previous text.

Suh, EunGyul

Project Concept 1: Accounting Fraud Detector. I assume that the SEC filings of companies involved in accounting fraud may contain exaggerated or magnified language and tone. I would like to develop a classification or anomaly detection model that identifies accounting fraud by analyzing companies' SEC filings. However, I anticipate that datasets containing SEC filings from fraudulent companies would not be available enough to train the model effectively.

Project Concept 2: Fake News Detector. Fake news on social media has become a significant concern in recent times. Similar to the first concept, I aim to develop a detection model that identifies potential fake news by analyzing social media posts, such as those on platforms like X. I would collect dataset from X with API or use existing dataset on fake news on social media. Additionally, I am interested in exploring which components of a social media post—such as whether the account is verified, whether the post contains images, the post's length, etc.—serve as indicators of potential fake news via extracting meta features from posts

Tolmochow, Gregory

Natural Language Query Interface for Structured Data This project aims to develop an NLP-powered interface that allows users to query structured databases (CSV or SQL) using natural language. Instead of writing complex SQL queries or manually filtering data, users can ask questions conversationally, and the system will break down their queries into structured filters. For example, given a housing dataset, a user might ask, "How many homes built in 2023 have 2 or more residents within 5 miles of a school?" The model would extract relevant columns such as year_built = 2023, # of ppl > 2, and dist_to_school < 5. These filters would be put into blocks that users can then refine these filters before and after applying them, or even pick which ones to remove. This project combines natural language understanding with data retrieval, making database interaction more intuitive and user-friendly.

Ukpong, Imeikan

One idea that hopefully will call on knowledge learned in this course is a tool that is able to take in prompts like conversations or simple sentences and returns whether that input is either positive or negative, (depending on the difficulty maybe neutral or more nuanced statements can be considered and categorized as well). To figure out what to produce as an output, the tool could observe certain keywords based on a certain heuristic (like words or symbols that tend to generally be important when people use them, like “I just got a new job!” has “new job” - positive and exclamation point further helps model be sure that the statement is positive).The sign of the output (Either positive or negative) is solely based on whether the input elicits positive emotions like happiness, laughter, etc. or negative emotions like sadness, pain, etc.

Wen, Yuanhuizi

This project aims to develop an AI-powered resume-to-job matching system using a dataset containing candidate resumes, the job positions they are applying for, and a numerically labeled match score. We will train a neural network-based regression model to predict the match score between a given resume and job position (already found the dataset from Kaggle). The model's performance will be evaluated using appropriate metrics such as Mean Squared Error (MSE) or R² score (We will do more research on this). This system has two key applications: (1) Automating resume screening to help recruiters efficiently rank candidates, reducing manual effort, and (2) Assisting job seekers in evaluating how well their resume aligns with different job positions before applying. By streamlining the hiring process, this project saves time for both candidates and recruiters, increasing overall efficiency in job matching.

Wu, Junting

Different poets have different styles. People who read enough poems can know the author of a poem through its style. But what is the style? Is it the use of words (some poets would use certain words repeatedly, some poets have characteristic ways of arranging the sequence of words, etc.)? Is it the meaning? Is it the way of starting or ending a poem? Or is it just a feeling? I wonder whether AI can classify poems according to their authors just based on their style. The result may shed light on the understanding of a poet's style. Only English poets would be chosen to avoid issues with translation.

Xiaotong, Liu

My project aims to develop a program for analyzing text in books and novels to determine their genre based on word frequency patterns. The program will skim through a large amount of text, identify the most frequently occurring words, and use them as key indicators to make inferences about the overall theme or genre. Using Natural Language Processing (NLP) techniques, the program will identify genre-specific words and compare them to predefined datasets of literary genres such as suspense, fantasy, science fiction, or romance. This automated classification system helps readers to effectively categorize novels based on textual content rather than metadata.

Xing, Jerry

My team project concept is a program that evaluates media and determines the level of bias. It would specifically evaluate news articles and transcripts of news videos. While all media has the same goal of informing the viewer or reader, often there is a certain level of bias from the person who wrote the article or script. I believe that natural language processing could be very applicable for analyzing media and gauging the amount of bias. In terms of deliverables, the end product would aggregate various sources of media on a topic, and compare the degrees of bias between the content that is written on a singular subject.

Xinyuan, Hu

I'm planning to build a tool that uses natural language processing to analyze and evaluate research paper credibility on platforms like arXiv and Google Scholar. Given the massive volume of papers being published today, having an automated way to assess paper quality would be incredibly helpful. The project will use well-established papers from top journals as training data to identify what makes research credible. The analysis will focus on some key aspects such as: methodology robustness (how well research methods are described and justified), citation patterns (how the work connects with existing research), and overall writing quality, etc. The process might be retrieving papers from websites, extracting the paper contents, and hen developing a scoring method.

Xu, Jack

I'm interested in making a sentiment analysis tool for studies/papers that can classify the overall quality, or to see if there are patterns that can be found in papers that are considered low-quality in meta-research. Given the text of a paper, it should predict the quality of the evidence and overall methodology. Also, maybe it could highlight some common patterns that are indicative of high quality, etc.

Yang, Junhyeok

Idea: Making web-scraped data readable for LLM

Description: Web-scraped data from various websites often contain unstructured and irrelevant content, making it difficult for LLMs to process effectively. Issues include raw HTML tags, boilerplate text, headers, footers, advertisements, and redundant information. NLP techniques can be leveraged to clean, preprocess, and structure this data into a format optimized for LLMs. Methods such as text normalization, entity recognition, summarization, and noise filtering help remove unwanted elements while preserving meaningful content. By applying NLP-driven parsing and formatting, web-scraped data can be transformed into a structured, high-quality dataset for better comprehension and usability for LLMs.

Yeruva, Sujith

I would like to do something involving sentiment analysis on stocks. There are sites like SeekingAlpha, ValueInvestorsClub, Yahoo Finance, and many others that analyze a stock and sometimes pitch it. There is also discussion on social media sites like Twitter and Reddit from individual retail investors. I would like to use this to potentially generate a "sentiment score" that shows what the public perception is on certain stocks (and how it might vary between different sources).

Yirdaw, Elshaday

I mainly have two project ideas that I would like to work on throughout the semester. My first (and main) project idea involves creating a tool that can estimate the readability level of a text provided by a user. In addition, I would like for this tool to offer some kind of suggestions that can make the text more accessible. These suggestions could range from simple modification, like replacing difficult words with more commonly used words, to advanced modifications, such as restructuring sentences or paragraphs that may otherwise be unclear and hard to understand. Essentially, the goal of the project would be to estimate the current readability level of the given text and provide modification suggestions to improve the readability of the text (perhaps to some level of complexity that a user might want). The secondary idea I am considering is developing some kind of a vocabulary builder. This tool would take a text provided by a user and identify words (and phrases) that may be challenging to understand. Then, using these words (and phrases), it would generate a vocabulary list along with definitions so that users can use it to expand their understanding.

Yoon, James

With the Information Age’s abundance of information, such an overwhelming sea of viewpoints and novel ideas may embolden more polarizing instances of media. Consequently, there remains an ever-urgent necessity to identify bias within journalism which may be obviated through utilizing NLP and subsequent sentiment analysis techniques. By leveraging sentiment analysis, named entity recognition, and topic modeling, one could assess the emotional tone, political leaning, and framing of articles from multiple sources. Users could input URLs or text to receive a breakdown of sentiment polarity, lexical bias, and comparative analysis against a diverse dataset of news sources. The tool could further employ machine learning models trained on labeled datasets to enhance accuracy, offering readers an objective lens through which to evaluate media narratives.

Yuxuan, Shi

Text Simplification for Patient Education

The complexity of medical language has been a major barrier in consumer health informatics. For individuals untrained in medical field, the healthcare processes become a black box. Here are some potential directions that I aim to optimize through this project: lexical and contextual complexity; health literacy; language and cultural barriers; patient-centric communication tools.

Solution: a multi-agent environment that can do:

lexical simplification: replace complex medical jargon with simpler terms; syntactic simplification to shorten long, complex sentences.
context specific explanations. ("benign" as "not cancerous" in a pathology report)
dynamic summarization: highlight and prioritize most critical information, such as diagnoses, treatment plans, next steps.

Zahid, Zeshan

The project Im thinking about creating is a webapp or program that can make edits to your resume weather it be format or edits to your words in the way you describe certain stuff in your resumes. Im familiar with backend webapp creation with Python and think using some of the libraries along with possibly Ai like chatgpt 4 to be able to make those edits. This tool could be useful for students as well as job seekers by helping tailor their resumes. Using Ai would make it alot more robust to where once inputed we could use Ai to tailor the resume for certain jobs or purposes and make effective edits.

Zhang, Jingzhi

Interactive Life Story Generator: Turn user-provided life details into a realistic, interactive, fictional story, where real-time user choices can adjust the plot (i.e., exploring future "what-if" scenarios). The project will allow users to specify the genre and ending type and use pre-trained models to dynamically generate realistic narratives based on the user's past experiences. The focus will likely be on prompt engineering to ensure coherence and immersion with a web-based UI for user interaction.

NLP Essentials

Overview

hashtagPrerequisites

hashtagSections

Syllabus

hashtagGeneral

hashtagInstructors

hashtagGrading

hashtagHomework

hashtagConcept Quizzes

hashtagProgramming Assignments

hashtagTeam Project

hashtagProject Grading

hashtagContribution

hashtagConsensus

Schedule

Development Environment

hashtagPython

Homework

hashtagTask 1: Getting Started

hashtagPackage Installation

hashtagTest Program

hashtagCommit & Push

hashtagSubmission

hashtagTask 2: Project Ideas

hashtagRubric

Text Processing

hashtagSections

Frequency Analysis

hashtagWord Counting

hashtagWord Frequency

hashtagSave Output

hashtagReferences

Regular Expressions

hashtag

Homework

hashtagTask 1: Chronicles of Narnia

hashtag

Language Models

hashtagContents

N-gram Models

hashtagUnigram Estimation

hashtagBigram Estimation

hashtagReferences

Maximum Likelihood Estimation

hashtagSequence Probability

hashtagChain Rule

hashtagMarkov Assumption

hashtagInitial Word Probability

hashtagReferences

Homework

hashtagTask 1: Bigram Modeling

hashtagImplementation

hashtagNotes

hashtagTask 2: Sequence Generation

hashtagImplementation

hashtagExtra Credit

hashtagSubmission

hashtagRubric

Projects

Speed Dating

hashtagProcess

Proposal Pitch

hashtagRequirements

hashtagSubmission

hashtagRubric

Project Ideas (2024)

hashtagMara Adams

hashtagNate Adams

hashtagCalvin Brauer

hashtagWenxuan Cai

hashtagMichael Cao

hashtagAlec Chapman

hashtagMarcus Cheema

hashtagMurphy Chen

hashtagAndrew Chung

hashtagHenry Dierkes

hashtagTung Dinh

hashtagBenjamin Dixon

hashtagLindsay Esterman

Prerequisites

Sections

General

Instructors

Grading

Homework

Concept Quizzes

Programming Assignments

Team Project

Project Grading

Contribution

Consensus

Python

Task 1: Getting Started

Package Installation

Test Program

Commit & Push

Submission

Task 2: Project Ideas

Rubric

Sections

Word Counting

Word Frequency

Save Output

References

Task 1: Chronicles of Narnia

Contents

Unigram Estimation

Bigram Estimation

References

Sequence Probability

Chain Rule

Markov Assumption

Initial Word Probability

References

Task 1: Bigram Modeling

Implementation

Notes

Task 2: Sequence Generation

Implementation

Extra Credit

Submission

Rubric

Process

Requirements

Submission

Rubric

Mara Adams

Nate Adams

Calvin Brauer

Wenxuan Cai

Michael Cao

Alec Chapman

Marcus Cheema

Murphy Chen

Andrew Chung

Henry Dierkes

Tung Dinh

Benjamin Dixon

Lindsay Esterman

Calla Gong

Hunter Grimes

Frederic Dean Michae Guintu

Molly Han

Paige Hendricks

Jerry Hong

Peter Jeong

Helen Jin

Carl Kassabian

Michelle Kim

Will Kohn

Andrew Lee

Sam Liu

Andrew Lu

Louis Lu

Wenzhuo Ma

Ja'Zmin McKeel

Izana Melese

Ruichen Ni

Ellie Paek