AI-based database composite query method

Abstract

The invention relates to an AI-based database composite query method. The method comprises an early-stage configuration step and a query use step, the use query step comprises the following steps: extracting a keyword group through an LLM large model; meanwhile, the Http request is verified, and a data theme list capable of being queried is obtained; intention recognition is conducted, and a query data theme is screened out from the obtained data theme list according to the recognized intention; obtaining detail data of the query data theme, and obtaining three pieces of data most similar to the natural language description in the http request in the query data theme; then inputting the SQL statement into an LLM large model to obtain an SQL statement; and combining the SQL statement with the keyword, performing composite query, obtaining an output result, and replying to the user. According to the method, from theme access, SQL execution to multi-level verification of a result field, sensitive information leakage is avoided by aiming at parameter-out permission filtering, and the data output security is enhanced.

Claims

1. A database compound query method based on AI, characterized in that: Includes preliminary configuration steps and usage query steps; The preliminary configuration steps include keyword query configuration, data subject query configuration and user authority configuration; The query step includes: Step 1: Get the natural language description of the current user input and generate an HTTP request; Step 2: Extract keyword phrases from the input natural language description using the LLM model. Simultaneously, verify the HTTP request to see if there are any data topics available for querying. If no such data topics exist, terminate the query. If so, obtain a list of such data topics. Step 3: Perform intent recognition on the natural language description in the HTTP request. If no intent is recognized, the query ends. If an intent is recognized, the most appropriate data topic is selected from the data topic list obtained in Step 2 based on the recognized intent and used as the query data topic. Step 4: Obtain detailed data of the query data subject and the three data items within the query data subject that are most similar to the natural language description in the HTTP request; then input the detailed data of the query data subject and the three most similar data items into the LLM model as prompt words for generating SQL, and the LLM model outputs the SQL statement; Step 5: Combine the SQL statement obtained in step 4 with the keyword combination obtained in step 1 to perform a compound query and obtain the output result; Step 6: Reply to the user based on the output result obtained in step 5 and the natural language description input by the user.

2. The AI-based database compound query method according to claim 1, characterized in that: The keyword query setting includes the input parameter open field setting and the output parameter field setting; the data subject query configuration includes the following configuration content: subject name, subject description, question example SQL, configuration association table within the subject, table permission setting, and setting output content restrictions for each table; user permission configuration is to set specific permission rules based on the user's identity information.

3. The AI-based database compound query method according to claim 1, characterized in that: In step 4, the query data details information includes the DDL of the accessible table, the dictionary value description, the problem example SQL, and the outer join table.

4. The AI-based database compound query method according to claim 1, characterized in that: The compound query in step 5 is specifically as follows: Match the keyword group to the input parameter field. If a match is found, obtain the queryable output data corresponding to the input parameter field. Then match and verify the queryable output data with the output parameter field, and return the verified output data as the keyword output data. Execute SQL statements, obtain execution results, filter according to user permission rules, and obtain SQL output data; Finally, the keyword output data is merged with the SQL output data to obtain the output result.

Images