Natural Language Processing Course - University of Tehran

This repository contains continuous assignments from the Natural Language Processing course at the University of Tehran. The assignments cover various NLP concepts, from tokenization to knowledge-based QA systems. Below is an overview of each assignment.

Assignments Overview

Assignment 1: Tokenization and N-Grams

• Topics: Tokenization, Custom Tokenizers, BERT, GPT, N-Gram Language Models.
• Key Tasks:
  • Implement a custom tokenizer using regular expressions.
  • Compare tokenization methods in BERT and GPT.
  • Build N-gram models for text completion.

Assignment 2: Sentiment and Sarcasm Detection

• Topics: Sentiment Analysis, Sarcasm Detection, Word Embeddings (GloVe), Logistic Regression.
• Key Tasks:
  • Perform sentiment analysis using Naive Bayes.
  • Detect sarcasm using Logistic Regression and GloVe.
  • Explore word similarities with skip-gram.

Assignment 3: Semantic Role Labeling

• Topics: SRL, LSTM and GRU Encoders, Encoder-Decoder Models.
• Key Tasks:
  • Label semantic roles using SRL.
  • Implement LSTM and GRU for SRL.
  • Convert SRL tasks into question-answer pairs.

Assignment 4: Fine-Tuning Large Language Models

• Topics: Fine-Tuning, LoRA, QLoRA, In-Context Learning (ICL).
• Key Tasks:
  • Fine-tune large models like Roberta and LLaMA.
  • Use zero-shot and one-shot learning.
  • Analyze model performance with LoRA and P-Tuning.

Assignment 5: Machine Translation

• Topics: Machine Translation, BPE, LSTM, Transformer Models.
• Key Tasks:
  • Build an English-to-Farsi translation system using Fairseq.
  • Train LSTM and Transformer models with BPE.
  • Evaluate using BLEU and COMET scores.

Assignment 6: Langchain Knowledge-Based QA System

• Topics: Knowledge-based QA, LangChain, Chain of Thought Reasoning.
• Key Tasks:
  • Build a multi-step QA system using LangChain.
  • Implement relevancy checks and context-based answers.

Each assignment is organized into folders, containing code, data, and reports. Implementations are in Python, utilizing libraries like PyTorch, Fairseq, and Huggingface.