Data processing method and system, electronic equipment and storage medium

Abstract

The invention provides a data processing method and system, electronic equipment and a storage medium. The method comprises the following steps: performing semantic intention prediction on a target problem of a target user to obtain a semantic intention of the target problem; selecting a target reference problem matched with the semantic intention of the target problem from a plurality of reference problems of which the semantic intentions are determined; wherein each reference question is associated with at least one database table for querying the corresponding reply content. And querying a database table associated with the target reference question for target reply content matched with the target question, and feeding back to the target user based on the target reply content. According to the method, the data query difficulty for the database table can be simplified, so that the threshold requirement on user use is reduced.

Description

Technical Field

[0001] This application relates to the field of communications, and more particularly to a data processing method, system, electronic device, and storage medium. Background Technology

[0002] The operation and maintenance of communication networks is inseparable from monitoring network status. Currently, information on determining network status is recorded in database tables. As communication networks grow larger, the number of database tables also increases. When checking a specific network status, maintenance personnel often only focus on some data in a subset of database tables. However, a thorough understanding of all database tables is required to configure query statements purposefully, which is not only difficult to operate but also demands a high level of expertise. Summary of the Invention

[0003] The purpose of this application is to provide a data processing method, electronic device, and storage medium that can simplify the difficulty of data querying for database tables, thereby lowering the barrier to entry for users.

[0004] To achieve the above objectives, the embodiments of this application are implemented as follows: Firstly, a data processing method is provided, including: Semantic intent prediction is performed on the target user's target question to obtain the semantic intent of the target question; Among multiple baseline questions with determined semantic intent, a target baseline question that matches the semantic intent of the target question is selected; wherein each baseline question is associated with at least one database table for querying the corresponding response content; Query the database table associated with the target benchmark question to find the target response content that matches the target question, and provide feedback to the target user based on the target response content.

[0005] In a second aspect, a data processing system is provided, comprising multiple intelligent agents, each of which is configured with a standard operating procedure, the standard operating procedure executing the method described in the first aspect, and each of the intelligent agents' standard operating procedures corresponding to its own algorithm for semantic intent prediction and / or a database table for query response content.

[0006] Thirdly, embodiments of this application provide an electronic device, including: a processor; and a memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method described in the first aspect.

[0007] Fourthly, a computer-readable storage medium is provided for storing computer-executable instructions that, when executed by a processor, implement the method described in the first aspect.

[0008] In this embodiment, the user describes the information they expect to retrieve from a database table by asking a question. Based on this query pattern, pre-defined possible user questions are used as baseline questions. Semantic intent prediction is performed on each baseline question to determine its semantic intent and the database tables that can provide corresponding answers. This establishes a relationship between the baseline questions and the database tables that can provide corresponding answers. In subsequent applications, when a target user asks a target question, the semantic intent of the target question is determined, and a target baseline question matching the semantic intent is selected. The database tables associated with the target baseline question are then used as the query scope to find the target answer content that matches the target question. Especially in scenarios involving querying the network status of a communication network, the solution in this embodiment only requires the user to describe the query in natural language. The program then intelligently understands the semantic intent of the user's question to match the corresponding baseline question, thereby quickly locating the database tables that can provide answers. For the user, there is no need for a high level of understanding of the information in each database table of the communication network, thus greatly simplifying the data query and lowering the barrier to entry for using database tables. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the first type of data processing method according to an embodiment of this application.

[0011] Figure 2 This is a schematic diagram illustrating the semantic prediction of a target problem using the data processing method of this application embodiment.

[0012] Figure 3 This is a schematic diagram of the application architecture of the data processing method according to an embodiment of this application.

[0013] Figure 4 This is a schematic diagram of a second flow of the data processing method according to an embodiment of this application.

[0014] Figure 5 This is a schematic diagram of the structure of a data processing device according to an embodiment of this application.

[0015] Figure 6 This is a schematic diagram of the structure of the data processing system according to an embodiment of this application.

[0016] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0017] Currently, when communication network maintenance personnel check network status, they need to first export all database tables of the communication network, and then find the corresponding Sheet page (a type of spreadsheet display) for each database table to configure data query methods. As communication networks grow larger and the number of database tables increases, maintenance personnel often only focus on some data in a subset of database tables when checking a specific network status item. However, they need a deep understanding of all database tables to configure targeted query statements, which is not only difficult to operate but also requires a high level of expertise. Therefore, this application aims to propose a data processing solution that simplifies the data querying process for database tables.

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] One embodiment of this application proposes a data processing method. Figure 1 This is a flowchart illustrating the data processing method, which includes the following steps: S102, perform semantic intent prediction on the target user's target question to obtain the semantic intent of the target question.

[0020] In this embodiment, users describe the information they wish to query by asking questions. For example, if a target user wants to query the number of university stations deployed in Luoyang, the target question can be described as "How many university stations are there in Luoyang?"

[0021] In one implementation, an embedding layer model can be used to predict the semantic intent of the target question. For example... Figure 2 As shown, the main structure of the embedding layer model includes an embedding layer and a Long Short-Term Memory (LSTM) network. The embedding layer converts each character in the text content of the target problem into a character embedding vector, resulting in a character embedding vector sequence E1, E2, ..., E... n , where n represents the number of characters; the Long Short-Term Memory network is used to process the word embedding vector sequence E1, E2, ..., E from the embedding layer output. nSemantic encoding is performed to obtain the semantic encoding sequence H1, H2, ..., H corresponding to the target question. n This semantic encoding sequence, or semantic vector, serves as the semantic intent of the target question as recognized by the machine in this embodiment.

[0022] S104, among multiple baseline questions whose semantic intent has been determined, select a target baseline question that matches the semantic intent of the target question; wherein each baseline question is associated with at least one database table that queries the corresponding response content.

[0023] This embodiment pre-defines potential user questions as baseline questions. For each baseline question, a database table providing corresponding answers is further defined. For example, if the baseline question is "How many university base stations are there in Luoyang?", and database tables 13 and 24 contain base station information for universities in Luoyang, then the following relationships can be determined: The baseline question is "How many university stations are there in Luoyang?" - Database tables 13 and 24.

[0024] Based on the above, the semantic intent of each benchmark question is further determined (the method for determining semantic intent can refer to the semantic intent prediction of the target question). By comparing the semantic intent of the target user with the semantic intent of each benchmark question (calculating the data distance between semantic vectors), the target basic question that matches the target question can be determined, and the database table associated with the target benchmark question can be determined as the query scope of the target question.

[0025] In practical applications, considering that the target user's query may contain multiple pieces of information, this embodiment can decouple the pre-defined complex sample user questions to more accurately locate the corresponding query range, and determine the resulting decoupled questions as the baseline questions. For example, if the sample user question is "How many stations does Luoyang Unicom have in each district?", it can be solved into the following baseline questions: 1. Provide the number of 5G base stations for each "administrative region" in Luoyang City; 2. Provide the number of 5G cells in each "administrative region" of Luoyang City; 3. Provide the number of 4G base stations in each "administrative region" of Luoyang City; 4. Give the number of 4G cells in each "administrative region" of Luoyang City.

[0026] It should be understood that problem decoupling allows complex sample user questions to be broken down into multiple simplified baseline questions. Similarly, after similarity matching with the user's target question, the matched target baseline question serves as a reference result for the decoupled target question. Therefore, the database table associated with the target baseline question can be considered a more precise query range. It should be noted that the baseline question in this embodiment can also directly adopt the aforementioned complex sample user question. Compared to decoupling the problem, setting the complex sample user question as the baseline question results in a target baseline question that is more closely related to the actual questioning scenario.

[0027] S106, query the database table associated with the target baseline question for the target response content that matches the target question, and provide feedback to the target user based on the target response content.

[0028] This embodiment is based on a large language model. For each target question, it generates query statements to query the associated database tables and executes each query statement to obtain the matching target response content.

[0029] Specifically, before generating query statements, prompts need to be set to guide the large language model. Prompts are instructive texts provided when interacting with the large language model. For example, a prompt guiding the large language model to translate could be, "You play the role of a translator expert, translating the user-input text into the user-specified language." In other words, the prompts define the role the large language model should play based on the desired function. Besides guiding the large language model to play its role, the prompts must also indicate at least the database table associated with each target question and the target baseline question. This enables the large language model to generate queries that retrieve target answers matching the target question from the associated database table and execute each query to obtain the matching target answer.

[0030] To this end, a query method can be configured for each baseline question, using the large language model to retrieve the corresponding answer content based on prompt words. This query method includes rules on how to guide the large language model to configure prompt words automatically. For target questions raised by target users, the original query method for each target baseline question is further customized based on the large language model and the database tables associated with the target question and the target baseline question. Based on the configured query method, a query statement is generated to query the database tables associated with the target baseline question, and the query statement is executed to obtain the target answer content that matches the target question.

[0031] In practical applications, in addition to the database tables associated with the target problem and the target benchmark problem, the prompt words can further include any of the following extended information: 1) Data requirements proposed by the target users.

[0032] Data requirements may include, but are not limited to, at least one of the following: content of interest to the target user, content of interest not interest, data precision, and data format. Correspondingly, the large language model can customize the query method associated with each target question according to the database table associated with the target question, the target baseline question, and the data requirements provided by the target user, and execute each query to obtain the matching target response content. Here, we take an example where the data requirements include content of interest not interest to the target user. In addition to asking "How many stations does Luoyang Unicom have in each district?", the user further sets the query method to include the data requirement "I don't need to check 4G" in the prompt. Correspondingly, the large language model, guided by the prompt, only generates queries for the number of 5G stations and 5G cells, and no longer generates queries for the number of 4G stations and 4G cells.

[0033] 2) Data usage rules for target fields associated with the target problem.

[0034] Data usage rules may include, but are not limited to, at least one of the following: data operation rules, data aggregation rules, data value retrieval rules, and data sorting rules. These are settings that guide the large language model on how to use the queried data.

[0035] This embodiment, based on the semantic intent of the target question, determines multiple matching target fields from the database table associated with the target baseline question (the matching method can be the semantic vector-based similarity comparison described earlier, which will not be repeated here). Correspondingly, the large language model can customize the query method associated with each target question according to the data usage rules of the database table associated with the target question, the target baseline question, and the target fields, and execute each query statement to obtain the matching target response content. For example, after a target user asks "How many cells does Luoyang Unicom have?", the target fields in the database table that can provide the response content can be determined, which provide information on Luoyang's 5G base stations and 4G base stations. These target fields are configured with the cell configuration for each 5G base station and each 4G base station, and the data operation rules for calculating the number of cells are based on the number of 5G base stations and 4G base stations and the cell configuration. Then, by setting the query method to include the question "How many cells does Luoyang Unicom have?" and the data operation rules in the prompt, the large language model can query the number of 5G base stations and 4G base stations in Luoyang from the target fields, calculate the number of cells according to the data operation rules, and use the number of cells as the query result.

[0036] The above is an exemplary introduction to setting prompt words. In practical applications, after obtaining the target answer content for each target question, this embodiment can also summarize the target answer content and feed the summary results back to the target user. As a feasible implementation method, the target answer content can be summarized using a large language model. That is, based on the large language model, data aggregation calculation is performed on all target answer content, and the data aggregation calculation results are summarized and refined using natural language expression to obtain the summary result.

[0037] Furthermore, to facilitate a more intuitive viewing of the summary results by the target users, this embodiment can also display the results in chart form. As a feasible implementation method, the target users can proactively provide chart configurations (chart dimensions, chart metrics, chart statistical methods, chart types, etc.) and / or chart examples to the large language model, so that the large language model can display the summary results to the target users according to the chart configurations and / or chart examples.

[0038] In this embodiment, the user describes the information they wish to retrieve from a database table by asking a question. Based on this query pattern, pre-defined possible user-assigned questions are used as baseline questions. Semantic intent prediction is performed on each baseline question to determine its semantic intent and the database tables that can provide corresponding answers. This establishes a relationship between the baseline questions and the database tables that can provide corresponding answers. In subsequent applications, when a target user queries and asks a target question, the semantic intent of the target question is determined. A target baseline question matching the target question's semantic intent is selected, and the database tables associated with the target baseline question are used as the query scope to find the target answer matching the target question. Particularly in scenarios involving querying the network status of a communication network, based on this embodiment, the user only needs to describe the query in natural language. The program then intelligently understands the semantic intent of the user's question to match the corresponding baseline question, thereby quickly locating the database tables that can provide answers. For the user, there is no need for a high level of understanding of the information in each database table of the communication network, thus significantly simplifying data querying and lowering the barrier to entry for using database tables.

[0039] The application of the method in this embodiment will be described by way of example below.

[0040] refer to Figure 3 As shown, the application architecture of the method in this embodiment includes: Platform as a Service (PaaS), database table server, and database table service interface front-end.

[0041] The platform as a service includes the models mentioned earlier, such as the embedding layer model for performing semantic prediction and the DaDa language model for performing queries. In addition, the platform as a service provides a semantic vector library of benchmark questions for matching against the target questions posed by users.

[0042] The database table server includes Agent nodes and a toolkit. The Agent node, essentially a platform-as-a-service (PAS) agent, can invoke the models, semantic vector libraries, and services configured within the PAS to further implement functions such as question decomposition, query statement generation, response content summarization, and chart generation. The Agent's main workflow is: "Identify user-submitted questions -> Determine and configure multiple query methods based on the user-submitted questions -> Generate query statements (SQL) based on multiple query methods -> Execute the query statements (SQL) -> Summarize the query results (response content) from multiple query methods -> Generate charts." Specifically, the execution of the query statements (SQL) utilizes the SQL executor within the toolkit, and the summarization calculator within the toolkit is used to summarize the query results for multiple sub-questions.

[0043] The front-end interface of the database table service is used to display summary results in a chart format.

[0044] based on Figure 3 The application architecture shown, and the corresponding process of the method in this embodiment are as follows: Figure 4 As shown, it includes: 1) Semantic intent prediction: This step imports the user's query into the embedding layer model for overall semantic intent prediction, generating the corresponding semantic vector.

[0045] 2) Problem breakdown: The Agent pre-labels each baseline question with a database table label (used to label at least one database table) that can provide answers. Next, the Agent matches the user's query with each baseline question based on similarity. If no target baseline question meets the similarity threshold, it is determined that the user's query cannot be answered, and the agent can directly refuse to answer the user. If a target baseline question meets the similarity threshold, the database table indicated by the database table label of the target baseline question is determined as the query scope.

[0046] As can be seen from the foregoing, the baseline problem in this embodiment is obtained through problem decoupling. After similarity matching with the target problem proposed by the user, the matched target baseline problem is equivalent to a reference result for problem decoupling of the target problem.

[0047] In addition, each baseline question is also configured with a query method that allows users to use a large language model to find the corresponding answer content based on the prompt words.

[0048] Based on the user's query input, the matched target benchmark question, and the database table indicated by the target benchmark question's database table label, the agent further customizes the query method for the target benchmark question and constructs prompts based on the configured query method. Specifically, this step involves task splitting strategies and Prompt construction.

[0049] 2.1-Task splitting strategy: That is, based on the question entered by the user, determine and configure the query method for the relevant target benchmark question.

[0050] For example, if a user's query is: "How many stations does Luoyang Unicom have in each district?", the target benchmark question obtained after similarity matching is: "How many stations does Luoyang Unicom have in each district?". The query method for the target benchmark question can be: 1. Provide the number of 5G base stations for each "administrative region" in Luoyang City; 2. Provide the number of 5G cells in each "administrative region" of Luoyang City; 3. Give the number of 4G base stations in each "administrative region" of Luoyang City; 4. Give the number of 4G cells in each "administrative region" of Luoyang City.

[0051] In addition, the query method has the following requirements: ①The scope is limited to a single table; ② The description needs to explicitly include the table name, field names, and query conditions.

[0052] 2.2- Prompt Construction: ① The number of target benchmark problems is required to be 3; ② Output format requirements: The query method for each target benchmark problem should be on a new line to distinguish between the serial numbers 1, 2, and 3; ③ Includes the actual questions entered by the user in the query; ④ Includes the user's data request (empty if the user does not submit any data request during the first preliminary query).

[0053] 3) Subtask n query processing: The subtask is defined as generating and executing a query statement based on each query method. For each subtask, the Agent first finds a matching database table and constructs a prompt based on the database table description and query method, so that the large language model returns the most suitable database table for the subtask.

[0054] The agent constructs a prompt for generating SQL query statements, which includes table creation statements, data usage rules for each relevant target field, SQL specifications, query methods, etc., requiring the large language model to generate a pure SQL statement.

[0055] Afterwards, the Agent will judge the return results of the large language model, extract the SQL statement between the Select and the semicolon (or the end of the sentence), and call the SQL executor to execute the SQL statement to obtain the query results.

[0056] In addition, the Agent can use the query method as the key and the corresponding query result as the value. After each subtask is completed, the key-value pair is written into the subtask's memory for storage.

[0057] 4) Subtask result aggregation: The Agent retrieves the query results of each subtask from all key-value pairs in the subtask Memory, and passes the query results as JSON Strings (in the format used to generate charts) to the large language model for natural language summarization. The prompt for natural language summarization is: if calculation is required, output calc(...). After output, ReAct iteration is performed through the large language model, calling the calculator, inputting the calculation expression, and replacing calc(...) with the result of the calculation, until the large language model outputs the final result.

[0058] 5) Generation of statistical charts 5.1 Input the summarized string, construct a prompt, and output the chart type based on the large language model. Options include: bar chart, line chart, and pie chart.

[0059] 5.2 For the selected chart type, construct a prompt containing a few examples of typical Echars options for that type, such as option = {"title": {"text": "Number of different 5G station types", "subtext": "Unit: Number"},"legend":{"show":true,"orient":"vertical","icon":"inherit", "left": "right","top": "top"},"xAxis": {"type": "category","data": ["Macro-site", "Indoor-distributed", "Micro-site"],"axisLabel":{"rotate":30}},"yAxis": {"type":"value"},"series": [{"data": [10, 20, 30],"type": "bar"}]} 5.3 The large language model generates opinion options for the summary.

[0060] 5.3 Extract the string after "option =" using regular expressions, up to "}".

[0061] 5.4 The data is sent to the front-end interface, where the front-end renders and displays the chart.

[0062] 6) User adjustment results Users can submit data requests for adjustments based on the chart results. The Agent adds the user's data requests to the summary prompt and then re-executes the "subtask result aggregation" and "statistical chart generation" processes.

[0063] In addition, the Agent can also adjust the type of statistical chart, add user requirements to the statistical chart generation prompt, and then re-execute the "statistical chart generation" process.

[0064] Based on the application architecture of the data processing method in this embodiment, relevant application examples are as follows: Application Example 1: 1) A user wanted to query the number of university sites of China Unicom in Luoyang, so he asked the system: "How many university sites does China Unicom have in Luoyang?"

[0065] 2) The system first calculates the similarity with the benchmark question set to 0.7, which is greater than the preset threshold of 0.6. Then, it prompts the user "@all Please assist in answering the user's question."

[0066] 3) After receiving a user's question, the Agent, acting as a planning expert through a large language model, determines and configures the relevant query methods and provides the following query methods: "1. Query the number of 5G base stations covered by Luoyang Unicom in universities, colleges, research institutes or laboratories; 2. Query the number of 4G base stations covered by Luoyang Unicom in universities, colleges, research institutes or laboratories to determine the total number of university base stations," and ask users to click confirm or type a description of their adjustment request.

[0067] 4) The user clicks "Confirm".

[0068] 5) The large language model acts as an agent for SQL experts to find the table separately for each query method and return the table names "mni_insight_same_station_identification_view_5G" and "mni_insight_same_station_identification_view_4G".

[0069] 6) After obtaining the table's schema through the large language model acting as an SQL expert agent, construct a prompt (taking the first query method as an example): ### Task Generate an SQL query to answer the question: [QUESTION] Query the number of 5G base stations covered by China Unicom in Luoyang.[ / QUESTION] ### Database Schema The query will run on a database with the following schema: CREATE TABLE mni_insight_same_station_identification_view_5G ( siteAddrProduct VARCHAR(256), -- Standard, matched with LIKE, '%NR%': 5G, '%LTE%': 4G siteName VARCHAR(256), -- Base station name siteId BIGINT, -- Base station gNodeB ID cellId INTEGER, -- Cell ID cellName VARCHAR(256), -- Cell Name morphology VARCHAR(256), -- Land cover type, urban area, suburbs, towns, rural areas, rural areas, main urban area, general urban area scenario VARCHAT(256), -- Scenario type: university, subway, shopping mall, laboratory city ​​VARCHAR(256) -- City, matched using LIKE, '%Changsha%': Changsha City ); ### Answer Given the database schema, here is the SQL query that [QUESTION] Query the number of 5G base stations in universities, colleges, research institutes or laboratories covered by China Unicom in Luoyang[ / QUESTION].

[0070] 7) The Agent sends the corresponding SQL1 "SELECT COUNT(*) FROM mni_insight_same_station_identification_view_5G WHERE city LIKE '%Luoyang%' AND (siteAddrProduct LIKE '%NR%' OR siteAddrProduct LIKE '%LTE+NR%') AND (scenario LIKE '%university%' OR scenario LIKE '%college%' OR scenario LIKE '%research institute%' OR scenario LIKE '%laboratory%')" and SQL2 "SELECT COUNT(*) FROM mni_insight_same_station_identification_view_4G WHERE city LIKE '%Luoyang%' AND (siteAddrProduct LIKE '%LTE%' OR siteAddrProduct LIKE '%LTE+NR%') AND (scenario LIKE '%university%' OR scenario LIKE '%college%' OR scenario LIKE '%research institute%' OR scenario LIKE '%laboratory%')" to the large language model for generation.

[0071] 8) The Agent, which acts as an SQL expert through the large language model, calls a tool to send an http request to the MNI microservice, obtaining the query results "The number of 5G base stations covering universities, colleges, research institutes or laboratories by Luoyang Unicom": [["count"],

[20] ]} and "The number of 4G base stations covering universities, colleges, research institutes or laboratories by Luoyang Unicom": [["count"],

[10] ]}, and stores them in Memory.

[0072] 9) The Agent, which acts as a summarization expert through the large language model, summarizes the results and replies with "The number of 5G base stations of Luoyang Unicom in universities, colleges, research institutes or laboratories is 20, and the number of 4G base stations is 10."

[0073] 10) The Agent, which acts as a chart expert through the large language model, determines to use a bar chart and generates the option for echarts as follows: {"title":{"text":"Number of 4G and 5G base stations operated by Luoyang Unicom in universities, colleges, research institutes or laboratories","subtext":"Unit: Number"},"legend":{"show":true,"orient":"vertical","icon":"inherit","left":"right","top":"top"},"xAxis":{"type":"category","data":["Number of base stations"],"axisLabel":{"rotate":-30}},"yAxis":{"type":"value"},"series":[{"name":"Number of 5G base stations","data":

[20] ,"type":"bar"},{"name":"Number of 4G base stations","data":

[10] ,"type":"bar"}]}.

[0074] 11) The Agent controls the rendering of the bar chart on the front-end interface.

[0075] Application Example 2: 1) Following the result of "Application Example 1", the user typed a follow-up question: "@Summary Expert, how many times larger is a 5G base station than a 4G base station?"

[0076] 2) The large language model acts as an agent to summarize the statement "The number of 5G base stations of Luoyang Unicom in universities, colleges, research institutes or laboratories is calc(20 / 10) times that of 4G." After being called by the "calculator" tool, the final output is "The number of 5G base stations of Luoyang Unicom in universities, colleges, research institutes or laboratories is twice that of 4G."

[0077] Application Example 3: 1) A user wanted to query the number of university sites of China Unicom in Luoyang, so he asked the system: "How many university sites does China Unicom have in Luoyang?"

[0078] 2) The system calculates a similarity of 0.7 with the benchmark question set, which is greater than the preset threshold of 0.6. Therefore, the system prompts the user "@all, please assist in answering the user's question."

[0079] 3) After receiving a user's question, the Agent, acting as a planning expert through a large language model, determines and configures the relevant query methods and provides the following query methods: "1. Query the number of 5G base stations covered by Luoyang Unicom in universities, colleges, research institutes or laboratories; 2. Query the number of 4G base stations covered by Luoyang Unicom in universities, colleges, research institutes or laboratories to determine the total number of university base stations," and ask users to click confirm or type a description of their adjustment request.

[0080] 4) Data requirement for the user's reply "I don't want to query 4G data".

[0081] 5) When the Agent that uses the large language model as a planning expert receives the user's words, it adjusts the query method and replies that the adjusted query method is: "1. Query the number of 5G base stations in universities, colleges, research institutes or laboratories covered by Luoyang Unicom" and asks the user to click to confirm or type a description of the adjustment request.

[0082] 6) The user clicks to confirm.

[0083] 7) The Agent that uses the large language model as a SQL expert finds the table for the sub-question and returns the table name "mni_insight_same_station_identification_view".

[0084] 8) The Agent that uses the large language model as a planning expert obtains the schema of this table and generates the corresponding SQL "SELECT COUNT(*) FROM mni_insight_same_station_identification_view WHERE city LIKE '%Luoyang%' AND (siteAddrProduct LIKE '%NR%' OR siteAddrProduct LIKE '%LTE+NR%') AND (scenario LIKE '%university%' OR scenario LIKE '%college%' OR scenario LIKE '%research institute%' OR scenario LIKE '%laboratory%')".

[0085] 9) The Agent that uses the large language model as a planning expert calls the tool to send an http request to the MNI microservice, obtains the query result "The number of 5G base stations in universities, colleges, research institutes or laboratories covered by Luoyang Unicom": [["count"],

[20] ]}", and stores it in Memory.

[0086] 10) The Agent that uses the large language model as a summarization expert summarizes the results and replies "The number of 5G base stations of Luoyang Unicom in universities, colleges, research institutes or laboratories is 20."

[0087] 11) The Agent, which acts as a chart expert through a large language model, determines the use of bar charts and generates echarts options as {"title":{"text":"Number of 5G base stations of Luoyang Unicom in universities, colleges, research institutes or laboratories","subtext":"Unit: number"},"legend":{"show":true,"orient":"vertical","icon":"inherit","left":"right","top":"top"},"xAxis":{"type":"category","data":["Number of base stations"],"axisLabel":{"rotate":-30}},"yAxis":{"type":"value"},"series":[{"name":"Number of 5G base stations","data":

[20] ,"type":"bar"}]}.

[0088] 12) The Agent controls the rendering of the bar chart on the front-end interface.

[0089] Application Example 4: 1) A user wants to ask about alarm-related issues, so he asks the system: "How many alarms are there on the network management system right now?"

[0090] 2) The system calculates a similarity of 0.1 with the benchmark question set, which is less than the preset threshold of 0.6. Therefore, the system prompts the user, "Sorry, I can only answer questions related to MNI. Please re-enter."

[0091] In addition, refer to Figure 5 As shown, another embodiment of this example also provides a data processing apparatus 500, including: The semantic intent prediction module 510 is used to predict the semantic intent of the target user's target question and obtain the semantic intent of the target question.

[0092] The benchmark question matching module 520 selects a target benchmark question that matches the semantic intent of the target question from a plurality of benchmark questions whose semantic intent has been determined; wherein each benchmark question is associated with at least one database table for querying the corresponding response content.

[0093] The data query execution module 530 queries the database table associated with the target benchmark question for target response content that matches the target question, and provides feedback to the target user based on the target response content.

[0094] Based on the device of this embodiment, the user describes the information they expect to retrieve from the database table by asking a question. Building upon this query pattern, pre-defined possible user-assigned questions are used as baseline questions. Semantic intent prediction is performed on each baseline question to determine its semantic intent and the database table that can provide corresponding answers for each baseline question. This establishes a relationship between the baseline questions and the database tables that can provide corresponding answers. In subsequent applications, when a target user queries and asks a target question, the semantic intent of the target question is determined. A target baseline question matching the semantic intent of the target question is selected, and the database table associated with the target baseline question is used as the query scope to find the target answer matching the target question. Particularly in scenarios involving querying the network status of a communication network, the device of this embodiment allows the user to simply describe the query in natural language. The program then intelligently understands the semantic intent of the user's question to match the corresponding baseline question, thereby quickly locating the database table that can provide the answer. For the user, there is no need for a high level of understanding of the information in each database table of the communication network, thus significantly simplifying data querying and lowering the barrier to entry for using database tables.

[0095] Optionally, the baseline question is pre-configured with a query method that uses a large language model to query corresponding answer content based on prompt words; the data query execution module 530 queries the database table associated with the target baseline question for target answer content that matches the target question, and provides feedback to the target user based on the target answer content, including: configuring the query method for each target baseline question according to the database table associated with the target question and the target baseline question based on the large language model; generating a query statement to query the database table associated with the target baseline question based on the configured query method; executing the query statement to obtain target answer content that matches the target question; summarizing all the target answer content obtained from the query; and feeding back the summary result to the target user.

[0096] Optionally, the data query execution module 530's step of feeding back the summary results to the target user includes: displaying the summary results to the target user according to the chart configuration and / or chart example provided by the target user based on the large language model; wherein, the chart configuration includes at least one of the following: chart dimension, chart indicator, chart statistical method, and chart type.

[0097] Optionally, the data query execution module 530 configures the query method for each target benchmark question based on the large language model and according to the database tables associated with the target question and the target benchmark question. This includes configuring the query method associated with each target question based on the large language model and according to the target question, the database tables associated with the target benchmark question, and the data requirements provided by the target user. The data requirements include at least one of the following: content that the target user is interested in, content that is not interested in, data precision, and data format.

[0098] Optionally, the data query execution module 530, based on the large language model, configures the query method for each target benchmark question according to the database tables associated with the target question and the target benchmark question. This includes: determining multiple matching target fields from the database tables associated with the target benchmark question based on the semantic intent of the target question; each target field is configured with data usage rules, which include at least one of data operation rules, data aggregation rules, data value rules, and data sorting rules corresponding to the target field; and configuring the query method associated with each target question based on the large language model, according to the database tables associated with the target question and the target benchmark question, and the data usage rules of the target fields.

[0099] Optionally, the data query execution module 530's summary of all the target response content includes: performing data aggregation calculation on all the target response content based on a large language model, and summarizing and refining the data aggregation calculation results in natural language expression.

[0100] Optionally, the quasi-basic problem includes: a complex sample user problem, or a decoupled problem obtained by decoupling the sample user problem.

[0101] It should be noted that the data processing device in this embodiment can be used as... Figure 1 The execution body of the method shown is therefore able to achieve... Figure 1 The functions and steps of the method will not be described in detail here.

[0102] In addition, refer to Figure 6 As shown, another embodiment of this example also provides a data processing system 600, including: a plurality of intelligent agents 610.

[0103] Each of the aforementioned intelligent agents is configured with a Standard Operating Procedure (SOP), which executes the above-described document. Figure 1 The method includes the following steps: Semantic intent prediction is performed on the target user's target question to obtain the semantic intent of the target question.

[0104] Among a number of baseline questions with determined semantic intent, a target baseline question that matches the semantic intent of the target question is selected; wherein each baseline question is associated with at least one database table that queries the corresponding response content.

[0105] Query the database table associated with the target benchmark question to find the target response content that matches the target question, and provide feedback to the target user based on the target response content.

[0106] In the data processing system 600 of this embodiment, each intelligent agent's standard operation procedure corresponds to its own algorithm for semantic intent prediction and / or database table for query and response content. That is, the target user's target question is published to each intelligent agent through the data processing system 600, and the intelligent agent, acting as an agent node, attempts to process the target user's query.

[0107] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Please refer to it. Figure 7 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0108] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0109] Memory is used to store computer programs. Specifically, a computer program may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides the computer program to the processor.

[0110] Specifically, the processor reads the corresponding computer program from non-volatile memory into memory and then runs it, forming the above-mentioned logical structure. Figure 5 The data processing device shown. Correspondingly, the processor executes the program stored in the memory, and specifically performs the following operations: Semantic intent prediction is performed on the target user's target question to obtain the semantic intent of the target question.

[0111] Among a number of baseline questions with determined semantic intent, a target baseline question that matches the semantic intent of the target question is selected; wherein each baseline question is associated with at least one database table that queries the corresponding response content.

[0112] Query the database table associated with the target benchmark question to find the target response content that matches the target question, and provide feedback to the target user based on the target response content.

[0113] The above is as described in this instruction manual. Figure 1 The communication encryption control method disclosed in the illustrated embodiments can be applied to a processor and implemented by the processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the processor or by instructions in the form of software. The processor mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0114] Of course, in addition to software implementation, the electronic device described in this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0115] Furthermore, this application also proposes a computer program that can operate a computer to perform the following operations: Semantic intent prediction is performed on the target user's target question to obtain the semantic intent of the target question.

[0116] Among a number of baseline questions with determined semantic intent, a target baseline question that matches the semantic intent of the target question is selected; wherein each baseline question is associated with at least one database table that queries the corresponding response content.

[0117] Query the database table associated with the target benchmark question to find the target response content that matches the target question, and provide feedback to the target user based on the target response content.

[0118] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer programs. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0120] The above are merely embodiments of this specification and are not intended to limit the scope of this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification. Furthermore, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of this document.

Claims

1. A data processing method, comprising: Semantic intent prediction is performed on the target user's target question to obtain the semantic intent of the target question; Among multiple baseline questions with determined semantic intent, a target baseline question that matches the semantic intent of the target question is selected; wherein each baseline question is associated with at least one database table for querying the corresponding response content; Query the database table associated with the target benchmark question to find the target response content that matches the target question, and provide feedback to the target user based on the target response content.

2. The method according to claim 1, The baseline question is pre-configured with a query method that uses a large language model to search for the corresponding answer content based on the prompt words; The step of querying the database table associated with the target baseline question for target answer content that matches the target question, and providing feedback to the target user based on the target answer content, includes: Based on the large language model, according to the database tables associated with the target question and the target benchmark question, the query method for each target benchmark question is configured, so as to generate a query statement to query the database table associated with the target benchmark question based on the configured query method, and execute the query statement to obtain the target answer content that matches the target question; All the target responses obtained from the query are summarized, and the summary results are fed back to the target user.

3. The method according to claim 2, The step of feeding back the summarized results to the target user includes: Based on the large language model, the summary results are displayed to the target user according to the chart configuration and / or chart example provided by the target user; wherein, the chart configuration includes at least one of the following: chart dimension, chart indicator, chart statistical method, and chart type.

4. The method according to claim 2, Based on the large language model, the query method for each target benchmark problem is configured according to the database tables associated with the target problem and the target benchmark problem, including: Based on the large language model, the query method associated with each target question is configured according to the target question, the database table associated with the target benchmark question, and the data requirements provided by the target user; wherein, the data requirements include at least one of the target user's content of interest, content of non-interest, data precision, and data format.

5. The method according to claim 2, Based on the large language model, the query method for each target benchmark problem is configured according to the database tables associated with the target problem and the target benchmark problem, including: Based on the semantic intent of the target problem, multiple matching target fields are determined from the database table associated with the target benchmark problem. The target fields are configured with data usage rules, which include at least one of the following: data operation rules, data aggregation rules, data value retrieval rules, and data sorting rules corresponding to the target fields. Based on the large language model, the query method associated with each target question is configured according to the target question, the database table associated with the target benchmark question, and the data usage rules of the target field.

6. The method according to claim 2, The process of summarizing all the target responses includes: Based on the large language model, data aggregation calculations are performed on all the target response contents, and the data aggregation calculation results are summarized and refined using natural language expressions.

7. The method according to any one of claims 1-6, The quasi-base problem includes: The complex sample user problem, or the decoupled problem obtained after decoupling the sample user problem.

8. A data processing system comprising a plurality of intelligent agents, each of the intelligent agents being configured with a standard operating procedure, the standard operating procedure executing the method as described in any one of claims 1-8, and each of the intelligent agents having its own standard operating procedure corresponding to an algorithm for semantic intent prediction and / or a database table for query response content.

9. An electronic device, comprising a processor; and a memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium for storing computer-executable instructions that, when executed by a processor, implement the method as claimed in any one of claims 1-7.

Images