Context-Aware Movie Chatbot: Multi-Turn Conversations

Project Overview

This project is part of the AAI-520: Natural Language Processing course in the Applied Artificial Intelligence Program at the University of San Diego under the guidance of Professor Kahila Mokhtari, Ph.D. Our goal is to design and implement a generative-based chatbot that not only engages in multi-turn conversations but also incorporates sentiment analysis to adapt its responses based on the emotional tone of user input.

The chatbot is trained using the Cornell Movie Dialogs Corpus, enabling it to handle diverse conversations with coherence, context-awareness, and emotional sensitivity.

Visit the Cornell Movie-Dialogs Corpus

Repository Structure (edit when project complete)

File/Folder Name	Description
Final Project Report-Team 6.ipynb	The Jupyter Notebook containing all the code for building and evaluating the chatbot.
Final Project Deliveries/	Final Project Code, PowerPoint Presentation, and Brief Report
requirements.txt	Dependencies required for the project (e.g., TensorFlow, PyTorch, Hugging Face).
README.md	The project overview and structure (this file).
data/	Raw dataset files, including the Cornell Movie-Dialog Corpus.
models/	Trained model checkpoints.
.gitignore	Lists files/directories ignored by Git.
LICENSE	Licensing information for the project.

---

Final Project Report

The final deliverables for this project include a comprehensive report in PDF format that contains:

• The full project report
• The Jupyter Notebook code
• Powerpoint presentation
• References and additional materials

Final Deliverables:

Final Project Deliveries/

• Final_Project_Notebook_Team_6.pdf: The full notebook converted to PDF, containing all code, analysis, and chatbot implementation.
• Final_Project_Report_Team_6.pdf: The final project report, including methodology, results, and references.
• Final_PowerPoint_Presentation_Team_6.pptx: The PowerPoint presentation summarizing key points and project progress.

Please refer to these files for all the details regarding the project, methodology, and evaluation.

Project Status: ✅ Completed

Team Members:

• Outhai Xayavongsa - Team Leader
• Saad Saeed - Lead Assistant
• Anand Fernandes - Team Member

Installation

To install and run the project on your machine, follow these steps:

1. Clone the repository:

   git clone https://github.com/oxayavongsa/NLP-Chatbot.git
   cd NLP-Chatbot
   

2. Create and activate a virtual environment:

   python3 -m venv chatbot-env
   source chatbot-env/bin/activate
   

3. Install the required packages:

   pip install -r requirements.txt
   

4. Download the Cornell Movie-Dialog Corpus:

   kaggle datasets download -d rajathmc/cornell-moviedialog-corpus
   unzip cornell-moviedialog-corpus.zip
   

5. Run the Chatbot:

   python NLP_Chatbot.py
   

Run the Chatbot

• Option 1: Open Chatbot.ipynb in Jupyter Notebook or Google Colab to run the chatbot interactively.
• Option 2: Use the command line interface to interact with the chatbot (see step 5 above).

Dataset Information

• We used the Cornell Movie-Dialogs Corpus which contains:

• 220,579 conversational exchanges between 10,292 pairs of movie characters.

• 9,035 characters from 617 movies.

• In total, 304,713 utterances with metadata such as genres, release year, IMDB rating, and character gender.

Methods Used

• Python for development.
• PyTorch for model training.
• Hugging Face Transformers for leveraging pre-trained models (T5).
• Jupyter Notebook or Google Colab for experimentation and testing.

Technologies

• Preprocessing: Data was cleaned by removing punctuation, stopwords, lemmatization, and rare words.
• Model Architecture: T5 (Text-To-Text Transfer Transformer) is used for multi-turn conversations and context-aware responses.
• Sentiment Analysis: Incorporated into the chatbot to adjust responses based on the user's emotional tone.

How the Chatbot Works

The chatbot uses a Transformer-based model (T5) to maintain multi-turn conversations. The sentiment analysis layer allows the chatbot to detect and adapt to emotional cues in the user's input, generating appropriate responses. It has been trained using the Cornell Movie-Dialogs Corpus, giving it the ability to handle movie-like dialogues with contextual coherence.

Example Usage

(Replace the above link with an actual gif demo or video)

Future Improvements

• Enhanced context retention over longer conversations.
• Fine-tuning the model for specific conversational styles or tones.
• Improving the user interface for better interaction.

License

This project is licensed under the MIT License – see the LICENSE file for details.

Acknowledgements

• Thanks to Cristian Danescu-Niculescu-Mizil and Lillian Lee for providing the Cornell Movie-Dialogs Corpus.
• Special thanks to Professor Kahila Mokhtari, Ph.D., for guidance throughout the course.
• Collaboration tools: GitHub, Slack, and Jupyter Notebook/Google Colab.