Conflict-RAG: Understanding Evolving Conflicts Using Large Language Models

This paper proposes Conflict-RAG, a method of working around the limitations of large language models (LLMs) to improve their efficacy in understanding current events and global conflicts. This method includes several steps. First, we create a database of Arabic news sources through web scraping. Then, we use weak supervision to create labels for the data to ensure they are relevant. Next, we use retrieval augmented generation (RAG) to inform the LLM about regional perspectives and current events that it would not otherwise know. Finally, we use an LLM to generate a response to a user’s query in order to answer their question. Our method provides an interface that allows non-Arabic-speaking users to gain an understanding of Arabic news sources. We demonstrate how our method improves response generation from an LLM by investigating the Israel-Hamas conflict.