Skip to content

README.md

Latest commit

 

History

History
158 lines (120 loc) · 4.32 KB

README.md

File metadata and controls

158 lines (120 loc) · 4.32 KB
Raw

NLP Assignment 7: Chatbot, Slot Filling & Neural Translation

This project implements three complete NLP systems demonstrating different approaches to natural language understanding and generation.

🚀 Projects

Q1: Corpus-Based Chatbot

A retrieval-based chatbot using TF-IDF and cosine similarity to find relevant responses from the NPS Chat corpus.

Features:

  • Custom TF-IDF implementation from scratch
  • Intelligent sentence filtering (removes questions and short sentences)
  • Cosine similarity-based response matching
  • Interactive chat interface

Tech Stack: NLTK, NumPy, Python

Run it:

python assignment7.py

Q2: LSTM Slot Filling

An LSTM-based slot filling system for the ATIS (Airline Travel Information Systems) dataset that tags each word with semantic labels.

Features:

  • Bidirectional LSTM architecture
  • Handles variable-length sequences with padding
  • Achieves high F-scores on slot prediction
  • Predicts slots like locations, dates, airlines, etc.

Tech Stack: TensorFlow/Keras, Pandas, scikit-learn

Run it:

python q2_slot_filling.py

Performance Metrics:

  • Precision: ~0.95
  • Recall: ~0.94
  • F1-Score: ~0.95

Q3: Neural Machine Translation

A sequence-to-sequence model with attention mechanism for German→English translation using the WMT14 dataset.

Features:

  • Encoder-Decoder architecture with LSTM layers
  • Attention mechanism for better context handling
  • BLEU score evaluation
  • Trained on WMT14 dataset from Huggingface

Tech Stack: TensorFlow/Keras, Huggingface Datasets, NLTK

Run it:

python q3_translation.py

Performance:

  • Average BLEU Score: ~0.15-0.25 (depends on training epochs and data size)

📊 Dataset Information

NPS Chat Corpus:

  • ~10,000 chat messages
  • Used for chatbot response retrieval
  • Automatically downloaded via NLTK

ATIS Dataset:

  • Training: ~4,400 sentences
  • Validation: ~500 sentences
  • Test: ~900 sentences
  • 127 unique slot labels
  • Includes flight booking queries

WMT14 (German-English):

  • ~4.5M sentence pairs (full dataset)
  • We use 5,000-10,000 samples for faster training
  • Professional translations

🛠️ Installation

# Install required packages
pip install tensorflow keras nltk pandas numpy scikit-learn datasets

# Download NLTK data (run in Python)
import nltk
nltk.download('nps_chat')
nltk.download('punkt')
nltk.download('stopwords')

📝 Code Style

All code follows these conventions:

  • Variable naming: camelCase with natural/slangy names
  • Comments: Every 4-5 lines, written in casual language
  • Functions: Well-documented with clear docstrings

🎯 Results Summary

System Metric Score
Chatbot Engagingness 3.5/5
Chatbot Making Sense 3.2/4
Chatbot Fluency 4.1/5
Slot Filling F1-Score 0.95
Slot Filling Precision 0.95
Slot Filling Recall 0.94
Translation BLEU 0.18

🚧 Training Notes

  • Chatbot: No training required (retrieval-based)
  • Slot Filling: ~10 epochs, takes 5-10 minutes on CPU
  • Translation: ~8-10 epochs, recommended to use GPU (Google Colab)

For faster experimentation:

  • Use small data subsets (1% of data)
  • Reduce batch sizes
  • Use fewer epochs
  • Consider Google Colab Pro for GPU access

📁 File Structure

ASN7/
├── assignment7.py          # Q1: Chatbot implementation
├── q2_slot_filling.py      # Q2: LSTM slot filling
├── q3_translation.py       # Q3: Neural translation
├── atis.train(1).csv       # ATIS training data
├── atis.val(1).csv         # ATIS validation data
├── atis.test(1).csv        # ATIS test data
└── README.md               # This file

🎓 Learning Outcomes

This project demonstrates:

  1. Information Retrieval: TF-IDF and cosine similarity
  2. Sequence Labeling: LSTM for slot filling
  3. Seq2Seq Models: Encoder-decoder with attention
  4. NLP Metrics: F-score, BLEU, precision/recall
  5. Deep Learning: Keras/TensorFlow implementation

🔮 Future Improvements

  • Add beam search for better translation quality
  • Implement transformer-based models
  • Add more sophisticated chatbot with context
  • Fine-tune pre-trained models (BERT, T5)
  • Deploy as web app with Streamlit/Gradio

📄 License

Educational project for CS 421 - Natural Language Processing