EICopilot - Baidu launches an AI-powered enterprise information search and exploration tool.

What is EICopilot?

EICopilot is an AI-powered enterprise information search and exploration tool developed by Baidu Research Institute. Driven by a large language model (LLM), it helps enterprise users efficiently retrieve and interpret information within large-scale knowledge graphs. EICopilot's core advantage lies in its powerful natural language understanding capabilities, transforming user natural language queries into precise graph database query scripts and quickly generating result summaries. It features automatic script generation and execution, effectively improving query accuracy and execution efficiency through innovative query masking strategies and context learning technology.

EICopilot's main functions

• Natural Language Query UnderstandingEICopilot is based on LLM to understand natural language queries and can automatically transform users' natural language questions into queries on knowledge graphs.
• Automatic script generation and executionThe system can automatically generate and execute Gremlin scripts to retrieve complex enterprise relationships in the knowledge graph and provide efficient information summarization.
• Data Preprocessing and Vector DatabasesEICopilot collects real-world enterprise-related queries to build a seed dataset, writes search scripts using the Gremlin language, forms a vector database, and optimizes database queries.
• reasoning abilityIt employs CoT (Cooperation of Thought) and ICL (Internal Context Learning) technologies to provide more accurate query responses.
• Query intent identificationBy employing a novel query masking strategy, entity names in queries are hidden, allowing for better identification of query intent and improved script accuracy.

EICopilot's technical principles

• Data preprocessingEICopilot constructs a seed dataset by collecting real-world queries related to enterprise information and writes search scripts using the Gremlin language. After annotation and augmentation, the data forms a vector database for In-Context Learning (ICL), providing accurate support for graph database queries.
• Online query processingWhen processing user queries, EICopilot achieves efficient response through the following steps:
  • Intent understanding and decision makingIt uses LLM to understand user intent and handle non-standard query statements.
  • Disambiguation processingIt resolves ambiguities in queries through Named Entity Recognition (NER) and entity retrieval models based on Natural Language Processing Customization (NLPC).
  • Query language generationTransform user queries into graph database query languages (such as Gremlin) and improve the accuracy of intent matching through query masking strategies.
  • Reflection ModuleThe generated query statements are validated and corrected to ensure the accuracy of syntax and logic.
  • Result generationTransform query results into text analysis to provide concise and comprehensive insights.
• Innovative query masking strategyBy hiding entity names in the query, EICopilot can more accurately identify query intent and improve the accuracy of query script generation.
• Performance optimizationEICopilot significantly outperforms traditional methods in terms of data retrieval speed and accuracy, reducing the syntax error rate to 10.00% and improving execution correctness to 82.14%.

EICopilot project address

• arXiv technical paper:https://arxiv.org/pdf/2501.13746

Application scenarios of EICopilot

• Financial AnalysisFinancial analysts can use EICopilot to quickly check the investment information of corporate shareholders. EICopilot provides efficient and accurate solutions through natural language understanding, complex query generation, and information summarization.
• Corporate SurveyWhen conducting due diligence on a company, users can use EICopilot to quickly obtain key information such as the company's registration information, major shareholders, and affiliated companies.
• Market ResearchMarket researchers can use EICopilot to explore the relationship networks between companies. EICopilot can quickly provide detailed information and relationship graphs of relevant companies through automated query script generation and execution.
• Business intelligence analysisEICopilot can help users quickly obtain and analyze enterprise information, such as querying basic information about a company, such as its legal representative and registered capital.