
A diagram of the multi-agent framework and three-layer cascaded filtering. [Photo/news.hzau.edu.cn]
A research team from the College of Informatics at Huazhong Agricultural University (HZAU) has made a significant advancement in Web intelligence. Their paper, titled "A Unified and Time-Efficient Multi-Agent Framework for Data Discovery", was recently published at the Web Conference 2026, addressing the ongoing challenge of balancing efficiency and accuracy in massive Web table discovery.
To navigate the complexities of large-scale data lakes, the team developed a sophisticated system consisting of six specialized agents: Configurator, Planner, Analyzer, Searcher, Matcher, and Aggregator Agent.
With the LangGraph workflow engine, these agents work together to streamline the data integration process. The core innovation is the Three-Layer Cascaded Filtering (TLCF) algorithm. This mechanism starts with metadata-based coarse screening, proceeds with vector-search for semantic recall, and ends with high-precision context reasoning powered by Large Language Models (LLMs).
This multi-agent approach addresses the efficiency-accuracy trade-off that has long impeded Web data integration. Beyond its technical performance, the research establishes a new paradigm for utilizing multi-agent systems to tackle large-scale information retrieval tasks.
By providing a unified and time-efficient solution, the HZAU team has paved the way for more intelligent and scalable data discovery in the era of big data.