EvoOntology：让Agent通过自演化本体层提升数据查询能力
Banger paper on self-evolving ontologies for agents.
Agent处理多源数据的痛点解决方案，提供了可复用的本体演化框架与MCP集成思路，值得做Data Agent的同学参考。
Banger paper on self-evolving ontologies for agents.
关于智能体自演化本体论的优秀论文。
You just can't go wrong with implementing an ontology layer for your agents.
为你的智能体实现一个本体层绝不会出错。
This paper shows exactly why.
这篇论文清楚地展示了原因。
The show that GPT-5.5 gains 26.7 points on DDR-Bench when the data agent can query an ontology of the data it works with.
研究表明，当数据智能体能够查询其工作数据的本体时，GPT-5.5 在 DDR-Bench 上的得分提升了 26.7 分。
Why is this useful?
为什么这很有用？
Data agents normally see tables, files and databases through generic tools, reading column names and file paths one call at a time.
数据智能体通常通过通用工具查看表格、文件和数据库，每次调用只能读取列名和文件路径。
The alternative is a hand-written semantic layer pasted into the prompt, which does not scale to many sources.
另一种方法是将手写语义层粘贴到提示词中，但这无法扩展到多个数据源。
EvoOntology builds the ontology with a dedicated agent and serves it as an MCP server with schema, content and tool layers.
EvoOntology 通过专用智能体构建本体，并以 MCP 服务器的形式提供模式、内容和工具层。
The data agent queries it at runtime.
数据智能体在运行时对其进行查询。
The ontology is then edited in small typed steps, and each edit is kept only if a paired evaluation on the same backbone shows it helps.
随后以小型类型化步骤编辑本体，并且仅当配对评估显示该编辑有助于提升基于相同主干模型的性能时，才保留该编辑。
Across six backbones on DDR-Bench, accuracy rises 17.8 points on average, from 4.8 on Qwen3.5-Flash to 26.7 on GPT-5.5. On BIRD, execution accuracy rises 7.4 points. Edits to the tool layer account for 57% of the gain from evolution.
在 DDR-Bench 的六个主干模型上，准确率平均提升 17.8 分，从 Qwen3.5-Flash 的 4.8 分提升至 GPT-5.5 的 26.7 分。在 BIRD 数据集上，执行准确率提升 7.4 分。工具层的编辑贡献了演化带来的 57% 的提升。
Paper: https://arxiv.org/abs/2609.15779
论文：https://arxiv.org/abs/2609.15779
Chat with Paper: https://academy.dair.ai/papers/evoontology-a-self-evolving-ontology-layer-for-data-agents-2609.15779
与论文对话：https://academy.dair.ai/papers/evoontology-a-self-evolving-ontology-layer-for-data-agents-2609.15779
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力