Data analysis method, device and equipment

Abstract

The invention provides a data analysis method, device and equipment, and the method comprises the steps: obtaining question information of a user for a target service type, and determining at least one drill-down dimension corresponding to the target service type; according to the question information and the at least one drill-down dimension, performing drill-down analysis on data in a preset database to obtain a plurality of data sets and an analysis result of each data set; analyzing the question information, the plurality of data sets and the analysis result through a preset question and answer model to obtain reply information corresponding to the question information; and outputting the reply information.

Description

Technical Field

[0001] This specification relates to one or more embodiments in the field of artificial intelligence, and more particularly to a data analysis method, apparatus, and device. Background Technology

[0002] Data analysis typically relies on the personal experience of individuals within the relevant field for data collection, processing, interpretation, and report writing. While this human-driven model can partially analyze data, it suffers from inefficiency, a limited analytical perspective, and a limited scale of data processing. Especially when data is growing rapidly, manual analysis exhibits significant lag and incompleteness.

[0003] Currently, general-purpose large models demonstrate powerful potential for natural language processing and data interpretation, offering new possibilities for automated analysis. However, directly applying general-purpose large models to data analysis in specialized and rigorous fields can lead to inaccurate analysis results. Summary of the Invention

[0004] In view of the above, one or more embodiments of this specification provide the following technical solutions: According to a first aspect of one or more embodiments of this specification, a data analysis method is proposed, the method comprising: Obtain user query information for the target service type and determine at least one drill-down dimension corresponding to the target service type; Based on the query information and the at least one drill-down dimension, drill-down analysis is performed on the data in the preset database to obtain multiple data sets and the analysis results of each data set; The question information, the multiple data sets, and the analysis results are analyzed using a preset question-and-answer model to obtain the response information corresponding to the question information. Output the response information.

[0005] According to a second aspect of one or more embodiments of this specification, a data analysis apparatus is provided, the apparatus comprising: The determination module is used to obtain user query information for a target service type and determine at least one drill-down dimension corresponding to the target service type; The first analysis module is used to perform drill-down analysis on the data in the preset database based on the query information and the at least one drill-down dimension, to obtain multiple data sets and the analysis results of each data set; The second analysis module is used to analyze the question information, the multiple data sets, and the analysis results through a preset question-and-answer model to obtain the response information corresponding to the question information. The output module is used to output the response information.

[0006] According to a third aspect of one or more embodiments of this specification, an electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor executes executable instructions to implement the steps of the method as described in any of the above embodiments.

[0007] According to a fourth aspect of one or more embodiments of this specification, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, implement the steps of the method as described in any of the above embodiments.

[0008] According to a fifth aspect of one or more embodiments of this specification, a computer program product is provided, comprising a computer program / instructions that, when executed by a processor, implement the steps of the method as described in any of the above embodiments.

[0009] As can be seen from the above embodiments, the data analysis method provided in this embodiment includes: obtaining user question information for a target service type, and determining at least one drill-down dimension corresponding to the target service type; performing drill-down analysis on data in a preset database based on the question information and the at least one drill-down dimension to obtain multiple data sets and analysis results for each data set; analyzing the question information, the multiple data sets, and the analysis results through a preset question-and-answer model to obtain response information corresponding to the question information; and outputting the response information.

[0010] By employing the above method, and by acquiring user questions regarding the target service type, combined with preset drill-down dimensions, accurate extraction and analysis of data from a preset database are achieved, improving the targeting and efficiency of data analysis. Furthermore, this embodiment inputs the question information, multiple datasets, and their analysis results into a preset question-and-answer model. This allows the model to generate response information not only based on the question information but also by integrating the analysis results of multiple datasets, thus improving the accuracy and interpretability of the response information. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a data analysis system provided in an exemplary embodiment.

[0012] Figure 2 This is a flowchart of a data analysis method provided in an exemplary embodiment.

[0013] Figure 3 This is a schematic diagram of a data analysis method workflow provided in an exemplary embodiment.

[0014] Figure 4This is a schematic diagram of the structure of a device provided in an exemplary embodiment.

[0015] Figure 5 This is a block diagram of a data analysis apparatus provided in an exemplary embodiment. Detailed Implementation

[0016] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification 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.

[0017] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data shall comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals shall be provided for users to choose to authorize or refuse.

[0018] First, the following terms appearing in this embodiment will be explained: Large Language Models (LLMs): These models typically refer to artificial intelligence models with a large number of parameters and massive training data, especially language models in deep learning. Examples include GPT (Generative Pre-trained Transformer) and PaLM (Pathways Language Model). These large models, pre-trained on massive amounts of text, possess powerful language understanding and generation capabilities, enabling them to perform various tasks such as question answering, writing, translation, and programming. The number of parameters in these models (reaching hundreds of billions or even trillions), the amount of training data, and the computational resource consumption are all very high.

[0019] Attribution analysis: Attribution analysis is a method used to identify and quantify the contribution of different factors to a final outcome. For example, in e-commerce, it is used to determine which channel (such as advertising, social media, email, etc.) facilitated a user conversion (such as a purchase). The goal of attribution analysis is to answer "the reasons for the corresponding result" and "the degree of influence of each factor on the corresponding result."

[0020] Prompt: A prompt is an instruction or question directed to the input of a large model, guiding it to generate the desired output. It can be a sentence, a question, or a piece of context, and may include examples (i.e., few-shot prompts). The design of the prompt directly affects the quality and relevance of the large model's output.

[0021] Intelligent Agent: In artificial intelligence, an intelligent agent is a software entity capable of perceiving its environment, making autonomous decisions, and taking actions to achieve specific goals. An intelligent agent typically possesses the following capabilities: perception (receiving input), reasoning (processing information), planning (formulating strategies), and action (output or execution). Among these, intelligent agents based on large models can invoke tools, remember history, and perform multi-step reasoning to complete complex tasks (such as booking airline tickets or writing reports), and are not limited to simple question-and-answer sessions.

[0022] Multi-agent: A multi-agent system consists of multiple agents that cooperate or compete with each other, working together to complete complex tasks through communication, negotiation, and division of labor. Each agent may have different roles, knowledge, or goals.

[0023] Workflow refers to the process of organizing, coordinating, and automating a series of ordered activities, steps, or operations according to predefined rules to accomplish a specific task or service goal. Its core lies in the structuring, standardization, and automation of processes to improve efficiency, reduce errors, and achieve traceability.

[0024] Currently, general-purpose large models demonstrate powerful natural language processing and data interpretation capabilities, offering new possibilities for automated data analysis. However, directly applying general-purpose large models to data analysis in specialized and rigorous fields can lead to inaccurate analysis results.

[0025] For example, as the digital transformation of the financial industry continues to deepen, the requirements for the depth, accuracy and timeliness of data-driven analysis in service decision-making are increasing. Currently, users directly input the financial data to be analyzed and the constructed prompts into a general large model for analysis. This method has the following problems: (1) It lacks a deep understanding of professional terminology, indicator system and service logic in the financial field, which can easily lead to the analysis results deviating from the actual service needs; (2) It relies on users to manually input data, which makes it difficult to support dynamic and multi-dimensional attribution drill-down analysis; (3) The inherent problem of "fictional information" and limited computing power of the large model can also easily output incorrect analysis results.

[0026] To address this, this embodiment proposes a data analysis method. By acquiring user questions regarding a target service type and combining them with preset drill-down dimensions, it achieves accurate extraction and analysis of data from a preset database, improving the targeting and efficiency of data analysis. Furthermore, this embodiment inputs the question information, multiple datasets, and their analysis results into a preset question-and-answer model. This allows the model to generate response information not only based on the question information but also integrating the analysis results from multiple datasets, thus improving the accuracy and interpretability of the response information.

[0027] The technical solutions provided by one or more embodiments of this specification will be described below.

[0028] refer to Figure 1 , Figure 1 This is a schematic diagram of a data analysis system 10 provided in an exemplary embodiment. For example... Figure 1 As shown, the data analysis system 10 may include a client 11 and a server 12.

[0029] The aforementioned server 12 can correspond to a server containing a single physical host, or a server cluster consisting of multiple independent physical hosts; or, it can correspond to a virtual server, cloud server, etc., hosted by a host cluster.

[0030] The aforementioned client can correspond to terminal devices such as smartphones, tablets, laptops, desktop computers, PCs (Personal Computers), PDAs (Personal Digital Assistants), wearable devices (e.g., smart glasses, smartwatches), smart in-vehicle devices, or game consoles.

[0031] Users can access the intelligent dialogue-style data analysis services provided by the aforementioned data analysis system through the aforementioned client; the aforementioned client and the aforementioned server can achieve user-oriented intelligent dialogue-style data analysis services through data interaction with each other.

[0032] Specifically, the aforementioned server can be equipped with a preset question-and-answer model and a preset database. The server can generate response information corresponding to the question information based on the preset database and the question-and-answer model.

[0033] For example, the aforementioned client can output a corresponding user interface, allowing the user to input natural language question text. The client can then send the user's input natural language question text to the server, which will rewrite it into corresponding question information. The server can then generate corresponding response information based on data in a pre-set database using a question-and-answer model, and output the response information to the user, or return the response information to the client. The client can then display the response information to the user through the user interface, thereby realizing a user-oriented intelligent dialogue-based data analysis service.

[0034] refer to Figure 2 , Figure 2 This is a flowchart of a data analysis method provided in an exemplary embodiment.

[0035] In this embodiment, the above-described data analysis method can be applied to, for example... Figure 1 The server shown is 12. (Example) Figure 2 As shown, the above data analysis method may include the following steps: Step 202: Obtain the user's question information regarding the target service type, and determine at least one drill-down dimension corresponding to the target service type.

[0036] In this embodiment, the query information can be structured SQL (Structured Query Language), which is used to query the corresponding data set from a preset database. Specifically, the corresponding data set can be queried from the preset database using RAG (Retrieval-Augmented Generation) technology.

[0037] In some embodiments, obtaining the user's question information includes: receiving the natural language question text input by the user; and performing semantic parsing and structured rewriting of the natural language question text using a preset language analysis model to obtain the structured question information.

[0038] The language analysis model can be pre-trained or a general-purpose large model. User-input natural language questions may contain unclear, insufficient, or ambiguous descriptions. In this embodiment, for user-input natural language questions with unclear semantics, the language analysis model first performs semantic parsing to obtain an intermediate question text, which is then returned to the user for confirmation. After user confirmation, the intermediate question text is rewritten in a structured manner to obtain the question information. If the semantics of the user-input natural language question text are clear, confirmation from the user is not required; semantic parsing and structured rewriting can be performed directly to obtain structured question information.

[0039] For example, a user can input a natural language question text on the client's user interface, such as "What is the default rate for 2022-2023?" The client sends this natural language text to the server. The server first performs semantic parsing on the natural language question text using a preset language analysis model to obtain an intermediate question text, such as "Analyze the loan default rate for 2022-2023." This intermediate question text can then be returned to the client for user confirmation, such as "Do you want to analyze the loan default rate for 2022-2023?" If the user replies "confirm" through the client, the server rewrites the intermediate question text in a structured manner to obtain structured question information. For example, the structured question information might look like this: "SELECT year,AVG(is_default) AS default_rate FROM loans WHERE year IN (2022, 2023) GROUP BY year" The structured question above indicates "analyze the loan default rate for 2022-2023".

[0040] In this embodiment, an NLP (Natural Language Processing) engine can be deployed on the server. This NLP engine can process natural language question text based on a preset language analysis model to obtain structured question information.

[0041] In some embodiments, determining at least one drill-down dimension corresponding to the target service type includes: analyzing the query information to obtain the target service type to which the query information is directed; and determining at least one drill-down dimension associated with the target service type based on the target service type, wherein the target service type and the at least one drill-down dimension have a preset association relationship.

[0042] After obtaining the query information, it is first analyzed to determine the target service type corresponding to the query. This analysis can be based on a pre-defined service type analysis model. For example, "analyzing the loan default rate in 2022-2023" corresponds to the target service type "loan service." "Analyzing deposit volume in 2024-2025" corresponds to the target service type "deposit service." In this embodiment, other methods can also be used to determine the target service type corresponding to the query information; this is not limited.

[0043] In some embodiments, the association between service types and at least one drill-down dimension can be preset. For example, in the financial field, service types include: loan services, deposit services, wealth management services, etc. Specifically, the at least one drill-down dimension associated with loan services includes: time dimension, region dimension, and loan product dimension. The at least one drill-down dimension associated with deposit services includes: time dimension, region dimension, customer dimension, and deposit product dimension. The at least one drill-down dimension associated with wealth management services includes: time dimension, wealth management product dimension, and wealth management channel dimension.

[0044] In some embodiments, the query information can also be directly analyzed using a pre-defined dimensional analysis model to obtain at least one drill-down dimension. This dimensional analysis model can also be a large model. For example, a prompt word can be pre-constructed, and the query information and the prompt word "Please analyze the drill-down dimension of the query information" can be input into the dimensional analysis model for analysis to obtain at least one drill-down dimension corresponding to the query information.

[0045] In this embodiment, the drill-down dimension corresponding to the query information can also be determined by other methods, and this embodiment does not limit this method.

[0046] In summary, this embodiment determines the drill-down dimension to facilitate accurate drill-down analysis of data in the preset database based on the drill-down dimension.

[0047] Step 204: Based on the query information and the at least one drill-down dimension, perform drill-down analysis on the data in the preset database to obtain multiple data sets and the analysis results of each data set.

[0048] In this embodiment, multiple data sets can be queried in a preset database based on the query information and drill-down dimensions. Then, each data set can be analyzed to obtain the corresponding analysis results. The data sets and analysis results can have a one-to-one correspondence.

[0049] Drill-down dimension allows you to query data at a finer level within the same dimension, such as from the national level, province to city, and then to county / district.

[0050] The preset databases include MaxCompute (a large-scale distributed data warehouse), MySQL (a relational database management system), and ClickHouse (a columnar online analytical processing database). In this embodiment, the preset database can also be other types of databases, which are not limited here.

[0051] For example, if the data in the preset database is financial data, this financial data includes a loan record table, which includes: loan ID, customer ID, loan amount, loan disbursement date, repayment status (normal or default), customer's region (province, city or county), and loan product type such as (personal consumer loan, personal mortgage or business loan).

[0052] In some embodiments, the query information and at least one drill-down dimension can be input into a preset query model, which processes the query information and at least one drill-down dimension based on data in a preset database to obtain multiple datasets and analysis results for each dataset.

[0053] In some embodiments, the step of performing drill-down analysis on data in a preset database based on the query information and the at least one drill-down dimension to obtain multiple data sets and analysis results for each data set includes: obtaining a drill-down analysis path corresponding to each of the drill-down dimensions, wherein the drill-down analysis path includes multiple levels; performing drill-down analysis on data in the preset database based on the query information and the drill-down analysis path to obtain multiple data sets and analysis results for each data set; wherein, one data set corresponds to one level and / or multiple levels, and if one data set corresponds to multiple levels, then the multiple levels belong to different drill-down analysis paths.

[0054] In this embodiment, one drill-down dimension corresponds to one drill-down analysis path. For example, for the aforementioned time dimension, the corresponding drill-down analysis path is from year to quarter to month, meaning the multiple levels included in this drill-down analysis path are year, quarter, and month, respectively. For the aforementioned region dimension, the corresponding drill-down analysis path is from province to city to county / district, meaning the multiple levels included in this drill-down analysis path are province, city, and county / district, respectively. For the aforementioned loan product dimension, the corresponding drill-down analysis path is from loan type to specific sub-product, meaning the multiple levels included in this drill-down analysis path are: loan type, specific sub-product, etc. For the aforementioned customer dimension, the corresponding drill-down analysis path is from customer category to customer level, meaning the multiple levels included in this drill-down analysis path are customer category, customer level, etc. For the deposit product dimension, the corresponding drill-down analysis path is from deposit category to deposit type to specific product code, meaning the multiple levels included in this drill-down analysis path are deposit category, deposit type, specific product code, etc. For the wealth management product dimension, the corresponding drill-down analysis path is from wealth management product category to risk level to product name, meaning that the multiple levels included in this drill-down analysis path are wealth management product category, risk level, and product name, respectively. For the wealth management channel dimension, the corresponding drill-down analysis path is from wealth management channel category to specific channel type to branch, meaning that the multiple levels included in this drill-down analysis path are wealth management channel category, specific channel type, and branch, respectively.

[0055] Furthermore, a preset query model can be used to perform drill-down analysis on the data in the preset database based on the query information and the drill-down analysis path, to obtain multiple data sets and the analysis results of each data set.

[0056] For example, at least one drill-down dimension includes: time dimension, region dimension, and loan product dimension. Referring to Table 1, since the query information includes time information (2022-2023), drill-down analysis can be performed on the data in the preset database at each level of the time dimension to obtain the corresponding data sets (e.g., data sets A1 to A3) and the corresponding analysis results (e.g., analysis results a1 to a3). Then, based on the time dimension, the region dimension is further considered, and drill-down analysis is performed on the data in the preset database to obtain the corresponding data sets (e.g., data sets B1 to B8) and the corresponding analysis results (e.g., analysis results b1 to b8). Based on the time dimension, the loan product dimension is further considered, and drill-down analysis is performed on the data in the preset database to obtain the corresponding data sets (e.g., data sets C1 to CN) and the corresponding analysis results (e.g., analysis results c1 to cN), where N is an integer greater than 1. Building upon the time dimension, further drill-down analysis is performed on the data in the preset database, considering the regional and loan product dimensions, to obtain corresponding datasets (e.g., datasets D1 to DM) and corresponding analysis results (e.g., analysis results d1 to dM), where M is an integer greater than 1. Table 1 provides examples of datasets and analysis results obtained based on (time dimension, loan product dimension) and (time dimension, regional dimension, loan product dimension), which can be referenced from (time dimension, regional dimension) and will not be listed individually here.

[0057] Table 1

[0058] In summary, in this embodiment, the multiple data sets obtained are derived from drill-down analysis of the data in the preset database based on the query information and at least one drill-down analysis path. Furthermore, the data sets and corresponding analysis results for (time dimension, loan product dimension), (region dimension, loan product dimension), and (time dimension, region dimension, loan product dimension) can be found in Table 1, and will not be elaborated further here.

[0059] For example, the query is "Analyze the loan default rate for 2022-2023," and at least one drill-down dimension includes: time dimension, region dimension, and loan product dimension. Based on this query and the corresponding drill-down analysis path, drill-down analysis is performed on the data in a preset database (such as a financial database), and the resulting dataset and analysis results are shown in Table 1. The dataset A1 and the corresponding analysis result a1 are also discussed.

[0060] Referring to Table 1, the level corresponding to data set A1 is "year". This data set includes data subset A11 (corresponding to the year 2022) and data subset A12 (corresponding to the year 2023), as shown in Table 2. The corresponding analysis result a1 includes the sub-analysis result a11 of data subset A11 and the sub-analysis result a12 of data subset A12.

[0061] Table 2

[0062] In some embodiments, multiple structured query texts can be determined based on the query information and at least one drill-down analysis path. Then, data in a preset database is queried based on these multiple query texts to obtain the corresponding data set. Finally, the data set is analyzed to obtain the corresponding analysis results. For example, one query text W1 is: SELECT YEAR(loan date) AS year, QUARTER(loan date) AS quarter, SUM(CASE WHEN repayment status = 'default' THEN loan amount ELSE 0 END) / SUM(loan amount) FROM Loan_Records GROUP BY year, quarter. Using this query text, multiple datasets can be obtained. For example, a subset of data within one dataset might contain the following sub-analysis results: The loan default rate in the second quarter of 2023 was 5%, a year-on-year increase of 2.3%.

[0063] In the query text above: SELECT (query) specifies which columns or calculation results to query. "YEAR (loan date) AS year" indicates the year the loan was issued. "QUARTER (loan date) AS quarter" indicates the quarter the loan was issued. "SUM (CASEWHEN repayment status = 'default' THEN loan amount ELSE 0 END)" means that if a loan is in default, it is included in the loan amount; otherwise, it is counted as 0, i.e., calculating the total amount of all "defaulted" loans in the corresponding year and quarter. "SUM (loan amount) FROM Loan_Records GROUP BY year, quarter" determines the total amount of all loans in the corresponding year and quarter. In summary, this query text indicates: From the loan record table, grouped by the year and quarter of the loan date, calculate the default rate for each quarter: Default rate = (Total amount of all "defaulted" loans in that quarter) / (Total amount of all loans in that quarter).

[0064] One of the query texts, W2, is: SELECT Province, SUM(CASE WHEN Repayment Status = 'Default' THEN Loan Amount ELSE 0 END) / SUM(Loan Amount) FROM Loan_Records WHERE Loan Date BETWEEN '2023-04-01' AND '2023-06-30' GROUP BY Province ORDER BY Default Rate DESC LIMIT 5. This query text indicates that the query should find the top 5 provinces with the highest default rates among loans disbursed in the second quarter of 2023 (April 1st to June 30th), and sort them by default rate from highest to lowest.

[0065] One of the query texts, W3, is: SELECT City, Loan Product Type, SUM(CASE WHEN Repayment Status = 'Default' THEN Loan Amount ELSE 0 END) / SUM(Loan Amount) FROM Loan_Records WHERE Loan Date BETWEEN '2023-04-01' AND '2023-06-30' AND Province = 'Jiangsu Province' GROUP BY City, Loan Product Type. This query text W3 indicates that the default rate for each type of loan product in Jiangsu Province during the second quarter of 2023 is calculated; specifically, it represents the percentage of defaulted loans in each city and for each type of loan product relative to the total loan amount for that product category.

[0066] It is understood that in this embodiment, multiple structured query texts can be determined based on the query information and at least one drill-down analysis path. Each query text corresponds to at least one level, and a corresponding data set can be retrieved for each query text. This embodiment does not list each query text, data set, and analysis result of the data set individually.

[0067] In summary, this embodiment combines the query information and at least one drill-down dimension to obtain different data sets and analysis results from multiple dimensions and levels, thereby achieving the accuracy and comprehensiveness of the obtained data sets and analysis results.

[0068] Step 206: Analyze the question information, the multiple data sets, and the analysis results using a preset question-and-answer model to obtain the response information corresponding to the question information.

[0069] The question-answering model mentioned above can be a pre-trained or general large model. This question-answering model summarizes the question information, multiple datasets and analysis results, and performs cross-analysis and summarization on these contents to extract the corresponding response information.

[0070] In some embodiments, the method further includes: inputting the question information and the analysis results into a preset attribution model, wherein the attribution model performs attribution analysis on the analysis results based on data in the preset database to obtain the attribution cause of the analysis results, wherein the preset attribution model is a pre-trained or general large model.

[0071] Attribution analysis, as we understand it, involves breaking down the components of a target metric in the analysis results to trace the driving factors and / or components behind that metric. Target metrics in the analysis results include, for example, sales revenue, conversion rate, user activity, loan default rate, and the rate at which loan default rates increase or decrease. The purpose of attribution analysis is to dissect the sources of the metric and identify the key factors influencing it.

[0072] Specifically, in the attribution analysis process, the indicator value in the analysis results can be broken down into multiple factors affecting that indicator value. Then, the changes in these factors are analyzed to determine the attributable causes of the indicator value. For example, if the target indicator in the analysis results is the loan default rate, with a corresponding value of 2%, this loan default rate is determined based on the total default rate (one element) and the total loan amount (another element). Therefore, the reasons for the target indicator value can be analyzed based on changes in the total default rate and the total loan amount.

[0073] In this embodiment, a pre-defined prompt word, such as "Please analyze the reason for obtaining this analysis result," can be preset. The prompt word, analysis result, and question information are then input into the attribution model. The attribution model can refer to data in a preset database and then analyze the analysis result based on the prompt word and question information to obtain the attribution reason for the analysis result.

[0074] Furthermore, the step of analyzing the question information, the multiple data sets, and the analysis results through a preset question-and-answer model to obtain the response information corresponding to the question information includes: analyzing the question information, the multiple data sets, the analysis results, and the attribution reasons through the question-and-answer model to obtain the response information.

[0075] In this embodiment, the question information, dataset, analysis results, and attribution reasons can be input into the question-answering model. After analyzing the question information, dataset, analysis results, and attribution reasons, the question-answering model obtains the response information. This approach allows the question-answering model to consider attribution reasons and data in the dataset during the analysis process, thereby improving the quality of the obtained response information.

[0076] In some embodiments, instead of inputting a dataset, only the question information, analysis results, and attribution reasons can be input into the question-answering model. The question-answering model can then summarize the analysis results and attribution reasons to obtain the corresponding response information. This approach avoids the question-answering model from analyzing a large dataset, thus improving the efficiency of the analysis.

[0077] In some embodiments, a level includes multiple values, a data set includes multiple data subsets, the values ​​corresponding to the multiple data subsets are not completely the same, the analysis result of a data set includes the sub-analysis results of each data subset included in the data set, and the method further includes: based on the query information and the analysis results, selecting at least one target data subset from the multiple data subsets included in each data set.

[0078] For example, if the question is "Analyze the loan default rate for 2022-2023", then for the "year" level, the corresponding values ​​are "2022" and "2023". If the level is "quarter", the corresponding values ​​are "First Quarter of 2022", "Second Quarter of 2022", "Third Quarter of 2022", "Fourth Quarter of 2022", "First Quarter of 2023", "Second Quarter of 2023", "Third Quarter of 2023", and "Fourth Quarter of 2023". If the level is "month", the corresponding values ​​are from "January 2022" to "December 2023". If the level is "province", the corresponding values ​​are such as "Jiangsu Province", "Zhejiang Province", etc. If the level is "city", the corresponding values ​​are such as "Suzhou City, Jiangsu Province", "Nanjing City, Jiangsu Province", etc. If the level is "county / district", the corresponding values ​​are such as "Taicang District, Suzhou City, Jiangsu Province", "Wujiang District, Suzhou City, Jiangsu Province", etc. When the tier is "Loan Type," the corresponding values ​​are personal loans, business loans, etc. When the tier is "Specific Sub-product," the corresponding values ​​are personal consumer loans, personal auto loans, personal travel loans, personal mortgages, business operating loans, business fixed asset loans, etc.

[0079] Furthermore, referring to Table 1, the level corresponding to data set A1 is "year". This data set includes data subset A11 (corresponding to the value of 2022) and data subset A12 (corresponding to the value of 2023), as shown in Table 2. The corresponding analysis result a1 includes the sub-analysis result a11 of data subset A11 and the sub-analysis result a12 of data subset A12.

[0080] If the hierarchy corresponding to data set B1 is "Year, Province", then data set B1 can include: data set B11, data set B12, data set B13, ..., where data set B11 corresponds to the value "2022, Jiangsu Province", data set B12 corresponds to the value "2022, Zhejiang Province", and data set B13 corresponds to the value "2023, Jiangsu Province". If the hierarchy corresponding to data set D1 is "Year, Province, Loan Type", then data set D1 can include: data set D11, data set D12, data set D13, ..., where data set D11 corresponds to the value "2022, Jiangsu Province, Personal Loan", data set D12 corresponds to the value "2022, Zhejiang Province, Personal Loan", and data set B13 corresponds to the value "2022, Zhejiang Province, Corporate Loan".

[0081] The above are some examples. For a dataset, if the dataset has P levels, and the i-th level among the P levels includes G... i There are several possible values, where P and G are... i All are integers greater than or equal to 1, with i ranging from 1 to P. Therefore, the total number of data subsets included in this dataset is S = G1 × G2 × ... × G p Therefore, the analysis results corresponding to this dataset also include S sub-analysis results.

[0082] Furthermore, for each dataset, at least one target data subset is selected from the multiple data subsets included in the dataset. For example, if the question asks about the loan default rate, then the data subset corresponding to the loan default rate being greater than 0 can be selected as the target data subset.

[0083] In this embodiment, the step of analyzing the question information, the multiple data sets, and the analysis results using a preset question-and-answer model to obtain the response information corresponding to the question information includes: analyzing the question information, at least one target data subset included in each data set, and the sub-analysis results corresponding to the target data subset using a preset question-and-answer model to obtain the response information corresponding to the question information.

[0084] It is understandable that by inputting only the target data subset and the corresponding sub-analysis results during the question-answering model processing, or by inputting only the corresponding sub-analysis results, unimportant data subsets can be removed, reducing the complexity of data analysis and improving analysis efficiency. In addition, it can make the obtained response information more targeted and avoid the influence of irrelevant data on the accuracy of the response information.

[0085] In some embodiments, the step of selecting at least one target data subset from multiple data subsets included in each data set based on the question information and the analysis results includes: analyzing the target indicator asked in the question information; for each data set, extracting the indicator value corresponding to the target indicator from each corresponding sub-analysis result; determining the data subset corresponding to the sub-analysis result whose indicator value is greater than the indicator value threshold as the target data subset, or, after sorting by indicator value size, determining the data subset corresponding to the first preset number of indicator values ​​as the target data subset.

[0086] For example, if the question is "Analyze the loan default rate in 2022-2023", then the target indicator for this question is "loan default rate". Then, the subset of data whose target indicator value is greater than the threshold value is determined as the target data subset.

[0087] In this embodiment, the threshold value of the indicator can be preset, such as 0. If the sub-analysis result of one of the data subsets is "the loan default rate in Jiangsu Province in 2022 is 0", then that data subset is not the target data subset. If the sub-analysis result of one of the data subsets is "the loan default rate in Zhejiang Province in 2022 is 1.2%", then that data subset is the target data subset.

[0088] In this embodiment, with a preset quantity of 5, if the sub-analysis result of a data subset is "the loan default rate in Suzhou, Jiangsu Province in 2022 was 5%", and it ranks first, then this data subset is the target data subset. If the sub-analysis result of a data subset is "the loan default rate in Nanjing, Jiangsu Province in 2022 was 4%", and it ranks second, then this data subset is the target data subset. If the sub-analysis result of a data subset is "the loan default rate in Huzhou, Zhejiang Province in 2022 was 3%", and it ranks third, then this data subset is the target data subset. If the sub-analysis result of a data subset is "the loan default rate in Hangzhou, Zhejiang Province in 2022 was 2%", and it ranks fourth, then this data subset is the target data subset. If the sub-analysis result of a data subset is "the loan default rate in Ningbo, Zhejiang Province in 2022 was 1%", and it ranks fifth, then this data subset is the target data subset. All other data subsets are not considered as target data subsets.

[0089] In this embodiment, when performing drill-down analysis at the "province" level, there may be 30 provinces. This embodiment only needs the top W provinces with the highest loan default rates. In some embodiments, the 30 provinces and the sub-analysis results (including loan default rates) of each province obtained from the drill-down analysis at the province level can also be input into the screening model, so that the screening model can automatically determine the value of W based on experience, and then determine the top W data subsets as the target data subset.

[0090] In this embodiment, the target data subset can also be determined in other ways, and there is no limitation on this.

[0091] In some embodiments, the step of selecting at least one target data subset from multiple data subsets included in each data set based on the question information and the analysis results includes: inputting the question information, the analysis results, and the multiple data sets into a preset filtering model, wherein the filtering model extracts the target indicator from the question information, and selects at least one target data subset from the multiple data subsets included in each data set whose corresponding sub-analysis results satisfy the target indicator.

[0092] In this embodiment, the filtering model can be a pre-trained model or a general-purpose large model. After the query model outputs multiple datasets and multiple analysis results, the query information, multiple datasets, and multiple analysis results are input into the filtering model, which automatically performs filtering to obtain at least one target data subset that meets the target indicators.

[0093] Meeting the target metric can be understood as either having a metric value greater than a threshold value (such as 0), or having a metric value that ranks among the top 100 values ​​in the dataset. Alternatively, the target metric can be determined in other ways, which are not limited here.

[0094] Furthermore, during the drill-down analysis of the above query model, the target data subset and the corresponding sub-analysis results can be automatically filtered. The output data set includes the corresponding target data subset, and the output analysis results include the sub-analysis results of the target data subset.

[0095] In this embodiment, the target data subset can also be obtained by other methods of filtering, which are not limited here.

[0096] Step 208: Output the response information.

[0097] In some embodiments, the method further includes: obtaining preset constraints before outputting the response information; inputting the constraints and the response information into a preset evaluation model for evaluation to obtain an evaluation result, wherein the evaluation result is used to indicate whether the response information satisfies the constraints; and outputting the response information includes: outputting the response information when the evaluation result indicates that the response information satisfies the constraints.

[0098] The preset evaluation model is a pre-trained or general-purpose large model. The constraints are the conditions that the response information must meet, such as correctness, reasonableness, and accuracy. The evaluation result can be "yes" or "no". "Yes" indicates that the response information meets the constraint and can be output. "No" indicates that the response information does not meet the constraint and will not be output.

[0099] In this embodiment, since the question-answering model may generate fictitious response information, it is necessary to evaluate the response information before the model outputs it, which can improve the stability and reliability of the output response information. Furthermore, the output response information can be sent to the client and displayed there for the user to view.

[0100] In this embodiment, the query model used can be a general large model. However, the query module may not cover specific scenarios, specific fields, or content services of different financial institutions. In this embodiment, drill-down dimensions can be considered. The corresponding data set and corresponding analysis results are first queried from the preset database and then input into the query model for processing to obtain accurate response information.

[0101] In some embodiments, reference is made to Figure 3 This illustrates a workflow diagram for a data analysis method. Figure 3 The system comprises multiple agents based on a large model: agent z1 (based on a language analysis model), agent z2 (based on a dimensionality analysis model), agent z3 (based on a query model), agent z4 (based on an attribution model), agent z5 (based on a question-answering model), and agent z6 (based on an evaluation model). Agent z1 processes the input natural language question text using the language analysis model to obtain the question information. Agent z2 processes the question information using the dimensionality analysis model to obtain multiple drill-down dimensions. Agent z3 performs drill-down analysis on data in a pre-defined database using the query model, based on the question information and the drill-down dimensions, to obtain a dataset and analysis results. Agent z4 performs attribution analysis on the analysis results using the attribution model, obtaining attribution results. Agent z5 then summarizes the question information, dataset, analysis results, and attribution results to obtain response information. The response information is evaluated by agent z6 and then output.

[0102] In this embodiment, a preset database can be provided to each large model (such as an attribution model or a query model), and multiple intelligent agents can be provided to provide relevant data in the preset database to the large model, thereby enhancing the large model's ability to acquire and analyze data.

[0103] Furthermore, the question-answering model in this embodiment only summarizes and analyzes the input data, without directly drilling down into the data in the preset database. The acquisition and drill-down analysis of other data are handled and fed back by other specialized intelligent agents, preventing the question-answering model from generating fictitious responses. In addition, an evaluation is performed before outputting the response information, further reducing the problem of inaccurate data analysis.

[0104] refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of a device provided in an exemplary embodiment. For example... Figure 4 As shown, device 400 mainly consists of a communication interface 402, a user interface 404, a processor 406, and a data storage 408. These components are interconnected and communicate with each other via a system bus, network, or other connection mechanism 410. The communication interface 402 enables device 400 to communicate with other devices, access networks, and transmission networks via analog or digital modulation. For example, the communication interface 402 may include a chipset and antenna for wireless communication with a radio access network or access point. Furthermore, the communication interface 402 can be a wired interface such as Ethernet, Token Ring, or a USB port, or a wireless interface such as Wi-Fi, Bluetooth, Global Positioning System (GPS), or a wide-area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 402 can also support other forms of physical layer interfaces and standard or proprietary communication protocols. The communication interface 402 may also include multiple physical communication interfaces, such as Wi-Fi, Bluetooth, and wide-area wireless interfaces.

[0105] User interface 404 includes receiving user input and providing output to the user. Therefore, user interface 404 may include input components such as a keypad, keyboard, touch-sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, and output components such as a display screen (which may be combined with a touch-sensitive panel), CRT, LCD, LED, display using DLP technology, printer, and other similar devices known or developed in the future. User interface 404 may also generate auditory output via speakers, speaker jacks, audio output ports, audio output devices, headphones, and other similar devices known or developed in the future. In some embodiments, user interface 404 may include software, circuitry, or other forms of logic capable of transmitting and receiving data from external user input / output devices. Additionally or alternatively, device 400 may support remote access from other devices via communication interface 402 or another physical interface (not shown). User interface 404 may be configured to receive user input, the position and movement of which may be indicated by an indicator or cursor described herein. User interface 404 may also be configured as a display device for rendering or displaying text fragments.

[0106] Processor 406 may contain one or more general-purpose processors and / or special-purpose processors.

[0107] Data storage 408 may include one or more volatile and / or non-volatile storage components and may be integrated wholly or partially with processor 406. Data storage 408 may include removable and non-removable components.

[0108] Processor 406 is capable of executing program instructions 418 (e.g., compiled or uncompiled program logic and / or machine code) stored in data storage 408 to perform the various functions described herein. Data storage 408 may contain a non-transitory computer-readable medium on which program instructions are stored, which, when executed by device 400, enable device 400 to perform any methods, processes, or functions disclosed in this specification and / or the accompanying drawings. Processor 406 executing program instructions 418 may result in processor 406 using data 412.

[0109] For example, program instructions 418 may include an operating system 422 (e.g., an operating system kernel, device drivers, and / or other modules) installed on device 400 and one or more applications 420 (e.g., a browser, social application, or game application). Similarly, data 412 may include operating system data 416 and application data 414. Operating system data 416 is primarily accessible to the operating system 422, while application data 414 is primarily accessible to one or more applications 420. Application data 414 may reside in a file system visible or hidden from the user of device 400.

[0110] Application 420 can communicate with operating system 422 through one or more application programming interfaces (APIs). These APIs help application 420 read and / or write application data 414, transmit or receive information via communication interface 402, receive or display information on user interface 404, etc.

[0111] In some terminology, application 420 may be simply referred to as "app". Furthermore, application 420 can be downloaded to device 400 through one or more online app stores or app markets. However, applications can also be installed on device 400 in other ways, such as through a web browser or a physical interface on device 400 (e.g., a USB port).

[0112] Corresponding to the embodiments of the aforementioned data analysis methods, this specification also provides embodiments of the data analysis apparatus.

[0113] refer to Figure 5 , Figure 5This is a block diagram of a data analysis apparatus provided in an exemplary embodiment. This data analysis apparatus can be applied to, for example... Figure 4 The device shown is used to implement the technical solution described in this specification.

[0114] The aforementioned data analysis device may include: The determination module 502 is used to obtain the user's question information regarding the target service type and determine at least one drill-down dimension corresponding to the target service type; The first analysis module 504 is used to perform drill-down analysis on the data in the preset database based on the question information and the at least one drill-down dimension, to obtain multiple data sets and the analysis results of each data set; The second analysis module 506 is used to analyze the question information, the multiple data sets and the analysis results through a preset question-and-answer model to obtain the response information corresponding to the question information; Output module 508 is used to output the response information.

[0115] In some embodiments, the first analysis module 504 is specifically used for: Obtain the drilling analysis path corresponding to each of the drilling dimensions, wherein the drilling analysis path includes multiple levels; Based on the question information and the drill-down analysis path, drill-down analysis is performed on the data in the preset database to obtain multiple data sets and the analysis results of each data set; In this context, a dataset corresponds to one level and / or multiple levels. If a dataset corresponds to multiple levels, then these multiple levels belong to different drill-down analysis paths.

[0116] In some embodiments, it also includes: The attribution analysis module (not shown in the figure) is used to input the question information and the analysis results into a preset attribution model. The attribution model performs attribution analysis on the analysis results based on the data in the preset database to obtain the attribution causes of the analysis results. The second analysis module 506 is specifically used for: The question-and-answer model is used to analyze the question information, the multiple data sets, the analysis results, and the attribution reasons to obtain the response information.

[0117] In some embodiments, a hierarchy includes multiple values, a data set includes multiple data subsets, the values ​​corresponding to the multiple data subsets are not completely identical, the analysis result of a data set includes the sub-analysis results of each data subset included in the data set, and the apparatus further includes: The filtering module (not shown in the figure) is used to filter out at least one target data subset from multiple data subsets included in each data set based on the question information and the analysis results. The second analysis module 506 is specifically used for: The question information, at least one target data subset included in each data set, and the sub-analysis results corresponding to the target data subset are analyzed by a preset question-and-answer model to obtain the response information corresponding to the question information.

[0118] In some embodiments, the filtering module is specifically used for: Analyze the target metrics asked in the question information; For each dataset, extract the corresponding indicator value of the target indicator from the respective sub-analysis results; The data subset corresponding to the sub-analysis results where the indicator value is greater than the indicator value threshold is determined as the target data subset. Alternatively, after sorting by indicator value size, the data subset corresponding to the first preset number of indicator values ​​is determined as the target data subset.

[0119] In some embodiments, the filtering module is specifically used for: The question information, the analysis results, and the multiple data sets are input into a preset filtering model. The filtering model extracts the target indicator from the question information and filters out at least one target data subset whose corresponding sub-analysis results satisfy the target indicator from the multiple data subsets included in each data set.

[0120] In some embodiments, it also includes: An evaluation module (not shown in the figure) is used to obtain preset constraints before outputting the response information; The constraints and the response information are input into a preset evaluation model for evaluation to obtain an evaluation result, which is used to indicate whether the response information satisfies the constraints. The output module 508 is specifically used to output the response information when the evaluation result indicates that the response information meets the constraint conditions.

[0121] In some embodiments, when determining at least one drill-down dimension corresponding to the target service type, the determining module 502 is specifically used for: Analyze the question information to obtain the target service type to which the question information is directed; Based on the target service type, at least one drill-down dimension associated with the target service type is determined, and the target service type and the at least one drill-down dimension have a preset association relationship.

[0122] In some embodiments, when obtaining the user's question information, the determining module 502 is specifically used for: Receive the natural language question text input by the user; The natural language question text is semantically parsed and structurally rewritten using a pre-defined language analysis model to obtain structured question information.

[0123] For ease of description, the above devices are described by dividing them into various modules or units based on their functions. Of course, when implementing one or more of these specifications, the functions of each module or unit can be implemented in one or more software and / or hardware components, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0124] Based on the same concept as the methods described above, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor executes the executable instructions to implement the steps of the methods as described in any of the above embodiments.

[0125] Based on the same concept as the methods described above, this specification also provides a computer-readable storage medium having computer instructions stored thereon that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0126] Based on the same concept as the methods described above, this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the methods as described in any of the embodiments above.

[0127] What those skilled in the art will understand is: In this specification, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, product, or apparatus. Without further limitation, the presence of additional identical or equivalent elements in the process, method, product, or apparatus that includes elements is not excluded.

[0128] In this specification, “a,” “an,” and “the” do not specifically refer to the singular, but may also include the plural.

[0129] In this specification, ordinal numbers such as "first," "second," etc., do not necessarily indicate order; they are often used to distinguish between objects. For example, "first server" and "second server" usually refer to two servers. To differentiate between these two servers, they are described as "first server" and "second server." Of course, sometimes these two servers may be the same server.

[0130] In this specification, unless explicitly stated otherwise, "receiving and sending data" does not necessarily mean direct receiving and sending; it can also mean indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving the data sent by B, or it can be understood as A indirectly receiving the data sent by B through other entities such as C. Similarly, B sending data to A can be understood as B sending the data directly to A, or it can be understood as B indirectly sending the data to A through other entities such as C. Here, C can be one entity, or it can be two or more entities.

[0131] In this specification, unless explicitly stated otherwise, the relationships between structures can be direct or indirect. For example, when describing "A is connected to B," unless it is explicitly stated that A and B are directly connected, it should be understood that A can be directly connected to B or indirectly connected to B. Similarly, when describing "A is on top of B," unless it is explicitly stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B or indirectly above B (AB is separated by other elements, and A is above B). And so on.

[0132] This specification uses specific terms to describe embodiments thereof. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those different embodiments or examples, without contradiction.

[0133] Although one or more embodiments of this specification provide method steps as shown in the embodiments or flowcharts, it is understood that the order of steps listed in the embodiments or flowcharts is only one way of executing many steps and does not represent the only execution order. Therefore, when the claims involve method steps, the changes and adjustments to the order of such steps, or the parallelism between steps, are also within the scope of protection of the claims.

Claims

1. A data analysis method, characterized in that, The method includes: Obtain user query information for the target service type and determine at least one drill-down dimension corresponding to the target service type; Obtain the drilling analysis path corresponding to each of the drilling dimensions, wherein the drilling analysis path includes multiple levels; Based on the query information and the drill-down analysis path, drill-down analysis is performed on the data in the preset database to obtain multiple data sets and the analysis results of each data set. Here, one data set corresponds to one level and / or multiple levels. If one data set corresponds to multiple levels, then the multiple levels belong to different drill-down analysis paths. The question information, the multiple data sets, and the analysis results are analyzed using a preset question-answering model to obtain the response information corresponding to the question information. The preset question-answering model is a pre-trained model or a general large model. Output the response information.

2. The method according to claim 1, characterized in that, Also includes: The question information and the analysis results are input into a preset attribution model. The attribution model performs attribution analysis on the analysis results based on the data in the preset database to obtain the attribution causes of the analysis results. The preset attribution model is a pre-trained model or a general large model. The step of analyzing the question information, the multiple data sets, and the analysis results using a preset question-and-answer model to obtain the response information corresponding to the question information includes: The question-and-answer model is used to analyze the question information, the multiple data sets, the analysis results, and the attribution reasons to obtain the response information.

3. The method according to claim 2, characterized in that, A hierarchy includes multiple values; a dataset includes multiple data subsets, where the values ​​corresponding to these subsets are not entirely the same; the analysis result of a dataset includes the sub-analysis results of each data subset included in the dataset; the method further includes: Based on the question information and the analysis results, at least one target data subset is selected from the multiple data subsets included in each data set; The step of analyzing the question information, the multiple data sets, and the analysis results using a preset question-and-answer model to obtain the response information corresponding to the question information includes: The preset question-and-answer model is used to analyze the question information, at least one target data subset included in each data set, and the sub-analysis results corresponding to the target data subset to obtain the response information corresponding to the question information.

4. The method according to claim 3, characterized in that, The step of selecting at least one target data subset from multiple data subsets included in each data set based on the query information and the analysis results includes: Analyze the target metrics asked in the question information; For each dataset, extract the corresponding indicator value of the target indicator from the respective sub-analysis results; The data subset corresponding to the sub-analysis results where the indicator value is greater than the indicator value threshold is determined as the target data subset. Alternatively, after sorting by indicator value size, the data subset corresponding to the first preset number of indicator values ​​is determined as the target data subset.

5. The method according to claim 3, characterized in that, The step of selecting at least one target data subset from multiple data subsets included in each data set based on the query information and the analysis results includes: The question information, the analysis results, and the multiple data sets are input into a preset filtering model. The filtering model extracts the target indicator from the question information and filters out at least one target data subset whose corresponding sub-analysis results satisfy the target indicator from the multiple data subsets included in each data set. The preset filtering model is a pre-trained or general large model.

6. The method according to any one of claims 1 to 5, characterized in that, Also includes: Before outputting the response information, obtain preset constraints; The constraints and the response information are input into a preset evaluation model for evaluation to obtain an evaluation result. The evaluation result is used to indicate whether the response information satisfies the constraints. The preset evaluation model is a pre-trained model or a general large model. The output of the response information includes: outputting the response information when the evaluation result indicates that the response information meets the constraint conditions.

7. The method according to any one of claims 1 to 5, characterized in that, Determining at least one drill-down dimension corresponding to the target service type includes: Analyze the question information to obtain the target service type to which the question information is directed; Based on the target service type, at least one drill-down dimension associated with the target service type is determined, and the target service type and the at least one drill-down dimension have a preset association relationship.

8. The method according to any one of claims 1 to 5, characterized in that, The process of obtaining user question information includes: Receive the natural language question text input by the user; The natural language question text is semantically parsed and structurally rewritten using a preset language analysis model to obtain structured question information. The preset language analysis model is either pre-trained or a general-purpose large model.

9. A data analysis device, characterized in that, The device includes: The determination module is used to obtain user query information for a target service type and determine at least one drill-down dimension corresponding to the target service type; The first analysis module is used to obtain the drill-down analysis path corresponding to each of the drill-down dimensions, the drill-down analysis path including multiple levels; according to the query information and the drill-down analysis path, drill-down analysis is performed on the data in the preset database to obtain multiple data sets and the analysis results of each data set, wherein one data set corresponds to one level and / or multiple levels, and if one data set corresponds to multiple levels, then the multiple levels belong to different drill-down analysis paths; The second analysis module is used to analyze the question information, the multiple data sets, and the analysis results through a preset question-answering model to obtain the response information corresponding to the question information. The preset question-answering model is a pre-trained model or a general large model. The output module is used to output the response information.

10. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in any one of claims 1 to 8 by executing the executable instructions.

11. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 8.

Images