Information seeking and integration are critical processes that underpin analysis and decision-making across various fields. Traditional search engines have reshaped how to seek information but often fall short when aligning with complex human intentions.
The main issue with current information-seeing methods is their inability to handle complex queries effectively. Traditional search engines frequently provide fragmented and noisy search results, making it difficult to find the necessary information.
Researchers from the University of Science and Technology of China and the Shanghai AI Laboratory have introduced MindSearch, a novel framework designed to mimic human cognitive processes in web information-seeking and integration. MindSearch is a multi-agent framework consisting of a WebPlanner and multiple WebSearchers. This innovative system leverages the strengths of both large language models (LLMs) and search engines, providing a more effective solution for complex information-seeing tasks.
MindSearch operates by decomposing complex user queries into smaller, manageable sub-questions. The WebPlanner orchestrates this process by modeling the query as a dynamic graph. This graph construction process involves breaking down the user query into atomic sub-questions, represented as nodes in the graph. The WebSearcher then performs hierarchical information retrieval, addressing each sub-question and collecting valuable data for the WebPlanner.
The framework has demonstrated significant improvements in response quality. Experimental evaluations have shown substantial enhancements in the depth and breadth of responses. In comparative analyses, human evaluators preferred responses from MindSearch over those from existing applications like ChatGPT-Web and Perplexity.ai.
MindSearch offers a simple multi-agent solution to complex information-seeking and integration tasks. Its explicit role distribution among specialized agents improves long-context management, facilitating more robust handling of complex and lengthy contexts. This design reduces the cognitive load on each agent and ensures that the information retrieval and integration processes are performed more efficiently.
In conclusion, MindSearch addresses the fundamental issues of traditional information-seeing methods by introducing a robust, multi-agent framework that combines the cognitive abilities of LLMs with the extensive data access of search engines.
Source: https://www.marktechpost.com/2024/08/01/mindsearch-a-multi-agent-ai-framework-processing-300-web-pages-in-under-3-minutes-to-enhance-information-retrieval-and-integration/