低时延智能问答方法、装置、电子设备和存储介质

By performing legality verification and historical mapping set matching on the client query requests of the intelligent question answering system, and combining small sample example fragments to assist large model reasoning, the timeliness problem of low latency response and high frequency repeated queries in the intelligent question answering system is solved, and a low latency and efficient intelligent question answering experience is achieved.

CN121597799BActive Publication Date: 2026-07-17SHANGHAI JIUFANGYUN INTELLIGENT TECH CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHANGHAI JIUFANGYUN INTELLIGENT TECH CO LTD
Filing Date
2025-11-28
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing intelligent question-answering systems are inadequate in terms of low-latency response, especially in real-time sensitive interactive scenarios such as financial consultation and real-time customer service. Users are prone to experiencing system lag due to prolonged lack of response, and the system is inefficient when performing frequent and repetitive queries.

Method used

After validating the client's query request, the system immediately returns the first frame of data. It then uses a pre-built historical mapping set to match the intent category and tool call trajectory summary to directly generate the final result. For queries that do not match, it uses a few sample fragments as supplementary hints to assist the large model in the first stage in intent and tool planning. Combined with streaming push technology, it reduces time-consuming steps such as large model inference and tool scheduling.

Benefits of technology

It significantly improves the perceived timeliness for users, adapts to high-frequency and repetitive query scenarios, reduces the perceived system lag, and achieves low-latency response. It is especially suitable for scenarios such as financial consultation and real-time customer service, and allows users to obtain processing progress in real time through streaming push technology.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121597799B_ABST
    Figure CN121597799B_ABST
Patent Text Reader

Abstract

本发明公开一种低时延智能问答方法、装置、电子设备和存储介质。方法包括:对客户端发送的当前查询请求进行校验,并向客户端反馈首帧数据;确定历史映射集合中是否存在与当前查询请求匹配的历史意图类别和历史工具调用轨迹摘要;若存在,根据历史意图类别和历史工具调用轨迹摘要生成最终查询结果,并推送至客户端;若不存在,从历史映射集合选出有效历史意图类别与有效历史工具调用轨迹摘要,并将其转为少样本示例片段,并作为第一阶段大模型的补充提示;通过第一阶段大模型和补充提示,确定当前查询请求的工具规划结果;根据工具规划结果,进行工具的调度执行;利用第二阶段大模型,根据工具的调度执行结果,生成最终查询结果推送至客户端。
Need to check novelty before this filing date? Find Prior Art