LLM Tutorial for Social Science Research

This repository contains materials for a workshop on using Large Language Models (LLMs) in social science research. The tutorial focuses on practical implementations using the langchain package and OpenAI's GPT-3.5-turbo model.

Overview

The tutorial covers:

1. Introduction to generative LLMs and their applications in social science
2. Implementation of two key methods:
   • Chat completion for text annotation
   • Retrieval-Augmented Generation (RAG)

Contents

• LLM_Tutorial.ipynb: Main Jupyter notebook containing the tutorial
• LLM_Tutorial.html: An html file rendered from the Jupyter notebook.
• data/ (gitignored but is required to run the file locally): Data for this tutorial is downloaded from the Global Populism Dataset.

Key Topics

Text Annotation Implementation

• Setting up OpenAI API and LangChain
• Handling text encoding and chunking
• Prompt engineering
• Chain creation and execution
• Validation using Krippendorff's alpha

RAG Implementation

• Word embeddings
• Vector stores
• Document retrieval
• Response generation
• Performance evaluation

Requirements

• Python packages:
  • langchain
  • openai
  • pandas
  • scikit-learn
  • dotenv
  • tiktoken
  • krippendorff