Part-of-Speech Tagging using Hidden Markov Models

This repository contains the implementation of a Part-of-Speech (POS) tagging system using Hidden Markov Models (HMMs) along with various decoding techniques and adversarial training strategies for sequence labeling tasks.

Overview

The project focuses on:

• HMM-based POS Tagging: Implementing HMMs for sequence labeling, associating hidden states (POS tags) with observed words.
• Decoding Techniques Comparison: Comparing and evaluating Greedy Decoding and Viterbi Decoding methods for sequence decoding.
• Adversarial Training: Exploring adversarial techniques to enhance the robustness and accuracy of the HMM-based POS tagging system.

Key Components

• Datasets: Wall Street Journal Dataset is utilized for training, validation, and testing purposes.
• Algorithms: Implementation of HMMs, Greedy Decoding, Viterbi Decoding, and Adversarial HMMs.
• Evaluation: Calculating accuracy metrics for decoding techniques and adversarial training strategies.

Installation and Usage

1. Dependencies: Ensure Python (version 3.10) and necessary libraries are installed.
2. Dataset: Sample of Wall Street Journal Dataset (provided in /data).
3. Training: Execute train_and_evaluate_hmm() and train_and_evaluate_advs_hmm() functions in main.py for respective models.
4. Decoding: Access decoding techniques via the decoder.py module.
5. Evaluation: Evaluate performance metrics using the provided evaluation functions.

Note

For every use case comment out the part not required in the hmm.py script and run the main.py script.

Course Information

• Course: DSCI 599 - Optimization Techniques for Data Science
• Term: Fall 2023
• Project Title: Part-of-Speech Tagging using Hidden Markov Models: Comparing decoding techniques and exploring adversarial training strategies

This project was developed as the final project for the Optimization Techniques for Data Science course offered in Fall 2023.