JitKnow 4.8 Adds Knowledge-Graph-Enhanced Retrieval
JitKnow 4.8 upgrades the AI knowledge base's retrieval architecture from pure vector search to ontology-driven, knowledge-graph-enhanced hybrid retrieval. The JitKnow team says the release adds entity-alias normalization, editable extraction, source linking, feedback workflows, and routing between knowledge-graph and existing text-based retrieval.
The release targets a common enterprise RAG problem—retrieving fragmented information and producing unsupported relationships—by combining structured graph data with conventional search, although the reported accuracy benefits have not been independently verified.
我们团队最近一直在迭代 JitWord 协同 AI 文档和 JitKnow AI 知识库。在大量用户反馈的驱动下,我们把知识库的底层检索体系做了一次重要升级:从"纯向量检索"升级为"本体驱动 + 知识图谱增强"。这篇文章记录一下技术思路和踩过的坑,给做企业知识库、RAG 落地的同行做个参考。
一、先说说痛点:传统 RAG 为什么容易"张冠李戴"
很多人在用 AI 知识库时会遇到一个很头疼的问题:明明文档里写得清清楚楚,AI 回答却经常出错、张冠李戴;同一个东西,文档里一会儿叫 A、一会儿叫 B,AI 就当成了两个完全不同的内容;问到两件事之间有什么关联,AI 要么答不上来,要么凭空编造关系。