BotNetSense: Detecting Bots in Elections Using LLMs and Network Analysis

This project explores the use of large language models (LLMs) and retweet network structure to detect bot-like behavior in Hebrew Twitter data from Israeli elections.

Project Structure

• notebooks/ – Jupyter notebooks for data analysis, LLM prompting, and evaluation
• data/ – Labeled tweet data and model responses
• presentation/ – Final slides summarizing the project and findings

Key Methods

• Gemini LLM prompting (Minimal vs. Full-feature)
• Network analysis with igraph
• Manual evaluation of bot behavior in context

Results

Full-feature prompts combining text and network signals significantly improved bot detection performance (Recall ↑ from 55% to 72%).