Speaker
Description
In 2019, alongside the sudden emergency by COVID-19 outbreak, an "infodemic" of misinformation emerged, causing public confusion and leading to mistrust in health authorities. In this paper, we propose a method to mitigate misinformation by enhancing query rewriting in retrieval-augmented generation (RAG) using Large Language Models (LLMs). LLMs such as ChatGPT and Llama have shown remarkable performance in natural language processing (NLP) tasks, especially when provided with external knowledge sources. Thus, existing works have used LLMs to verify the reliability of news articles. However, using news title as search query can limit the performance in evidence retrieval. Based on this, we generate improved retrieval queries by prompting LLMs with the news content to be able to retrieve more relevant evidence. This experiment was conducted using COVID-19 related news dataset. By providing better evidence, our method complements LLMs' internal knowledge and increases accuracy in assessing the article's reliability. Additionally, we fine-tune a smaller language model for the query rewriting task using reinforcement learning, allowing it to better adapt queries to the evidence database.
