HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems
Jiejun Tan, Zhicheng Dou, Wen Wang, Mang Wang, Weipeng Chen, Ji-Rong Wen
2024-11-06
Summary
This paper discusses HtmlRAG, a new approach that uses HTML instead of plain text to improve how systems retrieve and generate knowledge from the web, especially for large language models (LLMs).
What's the problem?
Retrieval-Augmented Generation (RAG) systems often extract plain text from HTML sources when retrieving information from the web. This process loses important structural details and meanings found in HTML, such as headings and tables, which can lead to less accurate responses from LLMs.
What's the solution?
The authors propose HtmlRAG, which keeps the original HTML format for retrieved knowledge. To manage the extra complexity that comes with HTML, they developed strategies to clean and shorten the HTML content while preserving important information. This includes a two-step method that removes unnecessary parts of the HTML, making it easier for RAG systems to process effectively.
Why it matters?
Using HTML directly allows RAG systems to retain more useful information, leading to better performance in understanding and generating responses. This advancement is significant because it enhances the capabilities of LLMs, making them more reliable for tasks like answering questions and providing information.
Abstract
Retrieval-Augmented Generation (RAG) has been shown to improve knowledge capabilities and alleviate the hallucination problem of LLMs. The Web is a major source of external knowledge used in RAG systems, and many commercial systems such as ChatGPT and Perplexity have used Web search engines as their major retrieval systems. Typically, such RAG systems retrieve search results, download HTML sources of the results, and then extract plain texts from the HTML sources. Plain text documents or chunks are fed into the LLMs to augment the generation. However, much of the structural and semantic information inherent in HTML, such as headings and table structures, is lost during this plain-text-based RAG process. To alleviate this problem, we propose HtmlRAG, which uses HTML instead of plain text as the format of retrieved knowledge in RAG. We believe HTML is better than plain text in modeling knowledge in external documents, and most LLMs possess robust capacities to understand HTML. However, utilizing HTML presents new challenges. HTML contains additional content such as tags, JavaScript, and CSS specifications, which bring extra input tokens and noise to the RAG system. To address this issue, we propose HTML cleaning, compression, and pruning strategies, to shorten the HTML while minimizing the loss of information. Specifically, we design a two-step block-tree-based pruning method that prunes useless HTML blocks and keeps only the relevant part of the HTML. Experiments on six QA datasets confirm the superiority of using HTML in RAG systems.