A question answering method based on memory retrieval

By searching for associated target images in the agent's historical interaction information and inputting them into a large model to generate response information, the problem of the agent being unable to handle too many recall text blocks is solved, and historical interaction experience is effectively reused and the accuracy of question answering is improved.

CN122240676BActive Publication Date: 2026-07-21DIGITAL QINGDAO CONSTRUCTION CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
DIGITAL QINGDAO CONSTRUCTION CO LTD
Filing Date
2026-05-20
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

In existing technologies, when the number of recalled text blocks is too large, the intelligent agent cannot effectively process the recalled information, resulting in the inability to effectively reuse historical interaction experience.

Method used

By searching for target historical interaction information that is semantically related to the question to be answered in the stored historical interaction information, and obtaining the saved target image, the target image records the analysis log of the agent's historical response information, including the question to be answered, historical response information, agent's thinking content, etc., and inputting it into the agent's target big model to generate response information.

Benefits of technology

This ensures that the agent can effectively reuse historical interaction experience, avoids exceeding the context limit of the large model, and improves the accuracy and efficiency of question answering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122240676B_ABST
    Figure CN122240676B_ABST
Patent Text Reader

Abstract

The application relates to the technical field of artificial intelligence, in particular to a question and answer method based on memory retrieval. Target picture of target historical interaction information associated with semantics of a question to be replied is acquired. Since the target picture records analysis logs of an intelligent agent when generating historical reply information based on historical questions to be replied, the analysis logs include at least one of the question to be replied, the historical reply information, thinking content of the intelligent agent, tool calling information, and core information of the historical question to be replied and the historical reply information, therefore, the question to be replied and the target picture are input into a target large model of the intelligent agent, so that the target large model generates target reply information of the question to be replied based on information in the target picture. Since the token length of a single picture is determined by the picture resolution, even if the information recorded in the target picture is more, it will not exceed the context upper limit of the target large model, so that the intelligent agent can effectively reuse historical interaction experience.
Need to check novelty before this filing date? Find Prior Art