Graduate Research Project (CS-421)

Guidelines

  • Explore different directions in NLP
  • Look at existing datasets and shared tasks in those areas
  • Get a sense of:
    • What is the task
    • What kind of input/output is being used
    • How systems are evaluated
    • What baseline models already exist

Team Formation

Team formation is an important step of the project. Students are encouraged to start looking for potential partners with similar research interests.

We will release a Research Group Form where you can:

  • Register your group if you already have one, or
  • Submit your research interests if you are looking for teammates

Based on the submitted interests, the instructional team will help form groups with similar research interests.

Research Group Form Link: Will be posted soon!

After reviewing a few directions and forming a team, you will meet with an instructor to:

  • Narrow down your topic
  • Decide on a concrete problem statement
  • Define the focus of your final project (e.g., modeling, analysis, comparison)

Note:
For your graduate project, you do not need to choose a problem from the shared tasks listed below or define a full research problem right away.

These topics are meant to help you explore areas of interest, not to lock you into a specific idea.

Research Topics

1. Emotion, Sentiment, and Stance

Tasks in this area examine how people express feelings, opinions, or positions in text and beyond.

You might explore:

  • Emotion recognition
  • Stance detection (support, oppose, neutral)
  • Multimodal emotion or stance detection (text + images, audio, or video)

Shared Tasks:

  • https://github.com/emotion-analysis-project/SemEval2025-task11
  • https://zeroqiaoba.github.io/MER2025-website/#introduction

2. Question Answering (Q&A)

These tasks focus on systems that answer questions based on internal or external knowledge.

You might explore:

  • Answering questions from documents
  • Open-domain question answering
  • Multi-step reasoning questions

Shared Tasks:

  • https://sites.google.com/view/mediqa2024
  • https://www.codabench.org/competitions/3360/

3. Retrieval-Augmented Generation (RAG)

RAG systems combine search/retrieval with text generation.

You might explore:

  • Comparing retrieval methods
  • Multi-document RAG
  • Evaluating faithfulness and correctness of RAG systems
  • Domain-specific RAG (e.g., finance, medicine)

Shared Tasks:

  • https://ibm.github.io/mt-rag-benchmark/MTRAGEval/
  • https://sites.google.com/view/semeval2026-task12/introduction

4. Text Generation, Reasoning, and Reliability

This area looks at how and why models generate text—and when they make mistakes.

You might explore:

  • Interpreting model outputs
  • Causal, logical, or mathematical reasoning
  • Controlled text generation
  • Hallucinations in generated text

Shared Tasks:

  • https://sites.google.com/view/semeval2026-task12/
  • https://helsinki-nlp.github.io/shroom/
  • https://llmunlearningsemeval2025.github.io/

5. Summarization

Summarization tasks require models to shorten text while preserving key information.

You might explore:

  • Long-document summarization
  • Factual consistency in summaries
  • Simplified or lay summaries
  • Domain-specific summarization (news, scientific, medical)
  • Evaluation metrics for summarization

Shared Tasks:

  • https://peranssumm.github.io/docs/
  • https://biolaysumm.org/

6. Multimodal NLP

Multimodal tasks combine language with other modalities.

You might explore:

  • Vision–language tasks such as image captioning
  • Image-based question answering
  • Multimodal emotion recognition
  • Cross-modal reasoning

Shared Tasks:

  • https://fever.ai/task.html
  • https://visualqa.org/
  • https://nocaps.org/

7. Fact Checking and Misinformation

These tasks focus on verifying claims and detecting false or misleading information.

You might explore:

  • Automated fact checking
  • Evidence retrieval for claims
  • Explainable verification
  • Misinformation in social media
  • Multimodal misinformation (text + images)

Shared Tasks:

  • https://checkthat.gitlab.io/clef2026/

8. Social Media and Mental Health

These tasks use text from social platforms to study behavior, opinions, and well-being.

You might explore:

  • Mental health states and symptoms in text or images (e.g., memes)
  • Hate speech or harmful content detection
  • Bias and fairness issues

Shared Tasks:

  • https://sites.google.com/site/offensevalsharedtask/
  • https://clpsych.org/shared-task/

9. Conversational Systems and Dialogue

These tasks focus on building and analyzing systems that interact with users via conversation.

You might explore:

  • Task-oriented dialogue systems (e.g., recommendation systems)
  • Open-domain conversational agents
  • Response generation and ranking
  • Consistency and coherence in dialogue
  • Hallucinations and factuality in responses
  • Safety and harmful behavior in conversational systems

Shared Tasks:

  • https://mcgill-nlp.github.io/FaithDial/
  • https://sites.google.com/view/persona-knowledge-workshop/shared-task
  • https://dstc12.dstc.community/

Why Shared Tasks?

Shared tasks and benchmarks are an excellent starting point because they:

  • Provide ready-to-use datasets
  • Clearly define tasks
  • Explain evaluation metrics
  • Often include starter code

You are encouraged to explore shared task websites to review datasets and task formulations before finalizing your research project topic.

Deadlines & Milestones

  • Research Team Formation Form Submission: January 30, 2026
    (Students must submit the form by this date.)

  • Midterm Research Report: TBD

  • Midterm Research Presentations: March 3 & March 5, 2026

  • Final Research Report: TBD

  • Final Research Presentations: April 28 & April 30, 2026