A serial two-layer retrieval method applied to artificial intelligence private memory banks

Abstract

This invention discloses a serial two-layer retrieval method applied to private memory databases for artificial intelligence. The invention employs a primary retrieval mode for semantic decomposition and hierarchical addressing retrieval, and determines the validity of the results of the primary retrieval mode. When the primary retrieval mode fails to find matching data or cannot identify valid search keywords, a fallback intelligent retrieval mode is automatically triggered to complete semantic expansion and hierarchical floating retrieval, forming a complete automated retrieval process without requiring secondary user input. This invention effectively solves the technical problems of retrieval failure when precise searches yield no results or when user queries or vague questions lack valid keywords, improving the success rate and smoothness of AI memory retrieval. The solution can be widely applied to large-model private memory database scenarios.

Description

Technical Field This invention relates to the field of artificial intelligence private memory retrieval technology, specifically a serial two-layer retrieval method applied to artificial intelligence private memory. Background Technology Current AI-powered private memory retrieval methods mostly employ a single, precise search mode. This type of retrieval relies on valid search keywords in the user's input, which has significant technical drawbacks: First, when the precise search fails to find a matching data, the retrieval process terminates directly, requiring the user to re-edit their input, resulting in extremely low interaction efficiency. Second, when users use rhetorical questions or vague queries, valid search keywords cannot be identified, rendering the precise search mode completely ineffective and unable to provide search results. Existing technologies lack an automatically integrated two-layer retrieval execution mechanism, failing to cover complex input scenarios such as those without keywords or using rhetorical questions, and thus struggling to meet the full-scenario usage needs of AI memory retrieval. Summary of the Invention (a) The technical problem that the invention aims to solve This invention overcomes the technical problems of existing single precise retrieval modes, such as retrieval failure, cumbersome interaction, and incomplete scenario coverage. It provides a serial two-layer retrieval method for artificial intelligence private memory, realizing a complete automated retrieval process that automatically triggers fallback retrieval after the primary retrieval fails, and is adaptable to complex input scenarios such as rhetorical questions and the absence of effective keywords. (II) Technical Solution A serial two-layer retrieval method applied to private memory banks for artificial intelligence includes the following steps: 1. Primary search steps The primary retrieval mode is used to retrieve the memory retrieval command input by the user. The primary retrieval mode is a specified level addressing retrieval based on semantic decomposition.

[0001] 2. Validity Determination Steps The validity of the search results of the primary search mode is determined. The valid search keywords refer to core words known in the technical field that can be used to accurately match the data in the memory database. There are two determination scenarios: one is that no matching data is found in the primary search mode; the other is that valid search keywords cannot be identified from the search instructions entered by the user, including cases where the user enters a rhetorical question.

[0002] 3. Search Execution Steps When the primary search mode finds matching data, the corresponding search result is output directly without triggering the fallback intelligent search mode. When a failure condition that meets any validity criterion is determined, the fallback intelligent retrieval mode is automatically triggered to perform the retrieval. The fallback intelligent retrieval mode adopts a combination of semantic expansion matching and floating retrieval of level 0 to level 3 of a fixed-level tree directory, so that the retrieval can be completed without the user having to input instructions again. (III) Beneficial Effects 1. Construct a two-layer search process that is executed serially. If the primary search fails, a fallback search is automatically triggered, eliminating the need for secondary user intervention and significantly improving search interaction efficiency.

[0003] 2. It covers complex input scenarios such as rhetorical questions and the absence of effective keywords, solving the application blind spots of the single precise search mode and achieving full-scenario search adaptation.

[0004] 3. A fallback retrieval method combining semantic expansion and level 0 to 3 hierarchical floating is adopted to improve retrieval success rate while ensuring retrieval efficiency.

[0005] 4. It can be directly adapted to the hierarchical architecture of existing large-scale AI private memory libraries, with strong adaptability, low modification cost, and high practicality. Detailed Implementation 1. Example 1: No results found in a precise search User-input search command: Find patent amendment records dated March 20, 2026. (1) First retrieval step: Using semantic segmentation-based hierarchical addressing retrieval, no corresponding data was found; (2) Validity determination steps: It is determined that no matching data was found in the primary search mode; (3) Search execution steps: Automatically trigger the fallback intelligent search mode, and output all patent-related records in March 2026 through semantic expansion matching and hierarchical floating search.

[0006] 2. Example 2: User's question has no valid keywords User's search query: What do you think of me? (1) Primary retrieval step: Using semantic segmentation-based hierarchical addressing retrieval, effective search keywords cannot be identified; (2) Validity determination steps: Determine the invalidity determination situation; (3) Search execution steps: Automatically trigger the fallback intelligent search mode, match the user's historical interaction memory through semantic expansion, and output historical memory data related to user evaluation. V. Summary This invention achieves full-scenario adaptation for AI memory bank retrieval through a serially executed two-layer retrieval mechanism, solving the failure problem of single precise retrieval in situations with no results, no keywords, or rhetorical questions. It has outstanding substantive features and significant technological advancements, and can be widely applied to various large-model AI private memory bank retrieval systems.

Claims

1. A serial two-layer retrieval method applied to a private memory bank of artificial intelligence, characterized in that, include: The primary search mode is used to search the memory search command entered by the user; The validity of the search results for the primary search mode is determined. The validity determination includes two situations: one is that no matching data is found, and the other is that valid search keywords cannot be identified. When any of the above conditions are met, the fallback intelligent search mode is automatically triggered to perform the search; The primary search mode and the fallback intelligent search mode are executed sequentially, eliminating the need for the user to input commands a second time.

2. The method according to claim 1, characterized in that, The primary retrieval mode is a specified level addressing retrieval based on semantic segmentation.

3. The method according to claim 1, characterized in that, The situations in which valid search keywords cannot be identified include situations where the user enters a rhetorical question.

4. The method according to claim 1, characterized in that, The aforementioned fallback intelligent retrieval mode employs a combination of semantic expansion matching and hierarchical floating retrieval, both well-known in the technical field, to perform the retrieval.

5. The method according to claim 1, characterized in that, When the primary search mode finds matching data, the corresponding search result is output directly without triggering the fallback intelligent search mode.