Long short-term memory processing system based on large model and vector database

CN121681786BActive Publication Date: 2026-04-28SHANGHAI JIDOU TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHANGHAI JIDOU TECH CO LTD
Filing Date
2026-02-09
Publication Date
2026-04-28

AI Technical Summary

Technical Problem

Existing technologies struggle to maintain continuity and consistency in output results in multi-turn interactions and cross-session scenarios. Disorderly stacking of historical interaction information, mixing of short-term context and long-term preferences, and simultaneous introduction of semantically similar but applicable information lead to inconsistencies in generated results, conflicting conditions, and uncontrollable output.

Method used

By using a long short-term memory processing system based on a large model and vector database, semantic parsing and structured modeling are performed. Topic segmentation and intent stack mechanisms are introduced to achieve orderly organization and hierarchical alignment of intent states in multi-turn dialogues. Based on the query semantic representation, matching memory entries are obtained from the cache database and the long-term database to construct controllable reasoning prompts and input them into the large model to generate the target output.

Benefits of technology

It improves the consistency of generated results, the continuity of tasks, and personalized responses, and enhances the controllability and reliability of the system in complex interactive scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121681786B_ABST
    Figure CN121681786B_ABST
Patent Text Reader

Abstract

The application relates to the technical field of large model interaction, in particular to a long short-term memory processing system based on a large model and a vector database, and the following scheme is provided: short-term context memory and long-term stable memory are hierarchically managed and jointly searched by performing semantic analysis and structured modeling on user interaction content. In the presence of historical interaction content, a topic segment division and intent stack mechanism is introduced to orderly organize and sequentially align the intent state in multi-round dialogue, and matching memory entries are obtained from the cache database and the long-term database based on query semantic representation. Further, by normalizing, situating and rewriting, and conflict inhibiting the memory entries, controllable reasoning prompt information is constructed and input into the large model to generate target output, thereby improving the consistency, stability and controllability of the generation result in multi-round interaction and cross-session scenarios.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • SQL (Structured Query Language) statement generation method and system based on long and short term memory of large language model

    CN119884151A

  • Voice-driven intelligent agent interaction control method and device for water saving field

    CN120564702A