AB
AB
AiBoss

All Categories

project

MindSearch - An AI search framework launched by the Shanghai Artificial Intelligence Laboratory

MindSearch is an open-source AI search framework developed by a joint research team at the Shanghai Artificial Intelligence Laboratory. It combines large-scale information gathering and processing capabilities. Utilizing the InternLM2.5 7B dialogue model, MindSearch can retrieve data from...

What is MindSearch?

MindSearch is an open-source AI search framework developed by a joint research team at the Shanghai Artificial Intelligence Laboratory. It combines large-scale information gathering and processing capabilities. Utilizing the InternLM2.5 7B dialogue model, MindSearch can collect relevant information from over 300 web pages within 3 minutes, completing a task that typically takes humans 3 hours. Employing a multi-agent framework to simulate human thinking, it plans before searching, improving the accuracy and completeness of the information. The project is fully open-source, and users can experience and deploy it locally for free.

MindSearch's main functions

  • Complex query processingBreaking down complex user queries into smaller, manageable sub-questions allows for more precise searching of relevant information.
  • Dynamic graph constructionBy constructing a directed acyclic graph (DAG), we can simulate the human thought process when solving problems, gradually refine the problem, and explore solutions.
  • Parallel Information RetrievalBased on a multi-agent architecture, parallel search of multiple sub-problems is achieved, improving the speed and efficiency of information retrieval.
  • Hierarchical retrieval strategyWebSearcher employs a coarse-to-fine search strategy, first collecting information broadly, and then selecting the most valuable pages for in-depth reading and information extraction.
  • Context ManagementEffectively manage context information in multi-agent systems to ensure the continuity and integrity of information during information retrieval and integration.
  • Response generationIt synthesizes the retrieved information to generate an accurate, comprehensive, and in-depth response to answer the original, complex query.
  • Performance improvementIn closed-set and open-set question-answering tasks, the above functions significantly improve the quality of answers, including both depth and breadth.
  • Human preferencesThe generated responses are more in line with human preferences; compared to other AI search engines, the solutions provided by MindSearch are more favored by human evaluators.

MindSearch's technical principles

  • WebPlannerAs a high-level planner, it is responsible for breaking down user queries into multiple sub-problems and simulating a mental model of multi-step information seeking through a dynamic graph construction process (directed acyclic graph, DAG).
  • WebSearcherPerform hierarchical information retrieval, retrieving and aggregating valuable information from the Internet based on the sub-questions assigned by WebPlanner.
  • Multi-agent collaborationWebPlanner and WebSearcher act as independent agents, handling problem decomposition and information retrieval tasks respectively, thus achieving parallel processing of tasks and effective integration of information.
  • Dynamic graph constructionBy generating and executing code, a logical graph for problem-solving is dynamically constructed, enabling LLM to progressively refine the problem and retrieve relevant information.
  • Context ManagementEffective context state transfer between multiple agents ensures that critical information is not lost during information retrieval and integration.

MindSearch's project address

Application scenarios of MindSearch

  • academic researchResearchers can use MindSearch to quickly collect and organize large amounts of literature to support their research.
  • Market AnalysisBusinesses can use MindSearch to collect market data, analyze competitor information, and monitor industry trends.
  • News reportJournalists can use MindSearch to gather background information on news events and quickly write reports.
  • Legal ResearchLegal professionals can use MindSearch to collect relevant legal provisions, cases, and precedents to aid in legal analysis and case preparation.
  • Technical SupportTechnical support teams can use MindSearch to quickly find methods and steps to solve technical problems.
Back