· Zehui Chen · Tutorials · 2 min read
An open-source AI Search Engine Framework with Perplexity.ai Pro performance
Technical Report: https://arxiv.org/abs/2407.20183
- LLMs fail to decompose complex requests into atomic queries, which adds difficulty in accurately and completely retrieving relevant information.
- search results are relatively massive compared to other tasks, requiring dedicatedly pre-selection.
- iterative web search content may quickly exceed the maximum capacity of LLM input length.
To address these issues, we introduce MindSearch, a simple yet effective LLM-based multi-agent framework for web search, consisting of a Web Planner and Web Searcher. WebPlanner models the complex problem-solving minds as a dynamic graph construction process: it decomposes the question into sub-queries as graph nodes and progressively extends the graph based on the search result from WebSearcher. Tasked with each sub-query, WebSearcher performs hierarchical information retrieval with search engines and collects valuable information for WebPlanner.
The multi-agent design of MindSearch dispatches a load of processing massive information to different agents, enabling the whole framework to process a much longer context i.e., of more than 300 web pages). To validate the effectiveness of our approach, we extensively evaluate MindSearch on both closed-set and open-set QA problems with GPT-4o and InternLM2.5-7B models. Experimental results demonstrate that our approach significantly improves the response quality in terms of depth and breadth. Besides, we also show that responses from MindSearch based on InternLM2.5-7B are preferable by humans to ChatGPT-Web (by GPT-4o) and Perplexity.ai applications, which implies that MindSearch delivers a competitive solution to the AI search engine. Code is available at https://github.com/InternLM/MindSearch.