Information processing method and device based on table data, storage medium and program product

Abstract

The embodiment of the invention provides an information processing method and equipment based on table data, a storage medium and a program product, a self-correction closed-loop processing mechanism of understanding, translation, execution and correction is constructed through a collaborative strategy generation model, a code generation model, a code interpreter and a strategy correction model, so that the data analysis process is automatic, and the efficiency is improved. Data analysis requirements of users can be automatically converted into executable code scripts, and answer results can be obtained. When an error occurs in execution, manual intervention is not needed, self error correction can be carried out through the strategy correction model, and a correct execution strategy is regenerated, so that efficient and continuous execution of the task is ensured, and the accuracy of an answer result is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a table data-based information processing method and device, a storage medium and a program product. BACKGROUND

[0002] With the improvement of the intelligent level of industry, enterprises increasingly rely on business systems to analyze and manage table information. At present, business systems mainly use three ways to analyze and manage table data: The first way is to rely on users to directly filter, calculate and summarize data on the interface of the business system. This way has the problem of errors due to human negligence. The second way is to export the data of the business system as a local file and then use desktop office software to process. This way breaks the real-time connection between data analysis and the business system, so that business data updates that occur after export cannot be counted, resulting in a time effectiveness deviation of the analysis results obtained, which cannot reflect the latest business situation. The third way is to customize the development of the business system to meet specific analysis needs in the form of fixed reports. Since the preset calculation logic often cannot accurately match the user's dynamic and specific request intention, the results produced do not match the real needs.

[0003] In summary, the existing scheme is difficult to guarantee the accuracy of the analysis results when dealing with dynamic and complex table data analysis needs due to its inherent operational complexity, data time effectiveness lag and logic solidification and other problems. Therefore, there is an urgent need for a table data processing method that can provide accurate and reliable analysis results for users. SUMMARY

[0004] Aspects of the present application provide a table data-based information processing method, device, storage medium and program product to improve the accuracy of table data processing.

[0005] The embodiment of the present application provides a table data-based information processing method, which comprises the following steps: S1, acquiring current multi-table data and a data analysis requirement of a user on the current multi-table data; S2, performing intention recognition and mapping inference processing by calling a strategy generation model according to the current multi-table data and the data analysis requirement, to obtain a target processing strategy, wherein the target processing strategy comprises an execution step required to complete the data analysis requirement; S3, performing translation processing by calling a code generation model according to the target processing strategy, to generate a target code script; S4, executing the target code script by calling a code interpreter, to perform data analysis on the current multi-table data according to the execution step, to obtain a data analysis result; S5, if the data analysis result comprises execution error information, modifying the target processing strategy by calling a strategy modification model according to the execution error information, to obtain a new target processing strategy; and S6, re-executing steps S3-S6 according to the new target processing strategy, until a data analysis result not comprising execution error information is obtained.

[0006] The embodiment of the present application also provides an electronic device, comprising a memory and a processor; the memory is used for storing a computer program; the processor is coupled with the memory and is used for executing the computer program, to implement the steps in any one of the table data-based information processing methods.

[0007] The embodiment of the present application also provides a computer readable storage medium storing a computer program / instruction, which enables the processor to implement the steps in any one of the table data-based information processing methods when the computer program / instruction is executed by the processor.

[0008] The embodiment of the present application also provides a computer program product, comprising: a computer program / instruction, which enables the processor to implement the steps in any one of the table data-based information processing methods when the computer program / instruction is executed by the processor.

[0009] In the embodiment of the present application, a table data-based information processing method is provided, a self-correcting closed-loop processing mechanism of understanding-translation-execution-correction is constructed by cooperating a strategy generation model, a code generation model, a code interpreter and a strategy modification model, so that the data analysis process is automated, the data analysis requirement of a user can be automatically converted into an executable code script and an answer result is obtained. When an error occurs, the strategy modification model can be used for self-correction without manual intervention, to re-generate a correct execution strategy, thereby ensuring efficient and continuous execution of the task and significantly improving the accuracy of the answer result. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings: Figure 1 A working principle diagram of the information processing method based on table data provided by an exemplary embodiment of the application; Figure 2 A flowchart of the information processing method based on table data provided by another exemplary embodiment of the application; Figure 3 An entry form diagram of the intelligent assistant in the user interface provided by another exemplary embodiment of the application; Figure 4 A diagram of automatically generating candidate recommended questions in the user interface by the intelligent assistant provided by another exemplary embodiment of the application; Figure 5 A diagram of answering user data query requirements in the user interface by the intelligent assistant provided by another exemplary embodiment of the application; Figure 6 A diagram of answering user data integration and analysis requirements in the user interface by the intelligent assistant provided by another exemplary embodiment of the application; Figure 7 An internal structure diagram of the processing device provided by another exemplary embodiment of the application. DETAILED DESCRIPTION

[0011] In order to make the objects, technical solutions and advantages of the application clearer, the technical solutions of the application will be described below in conjunction with the specific embodiments of the application and the corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the application.

[0012] In order to solve the problems in the background art, an embodiment of the application proposes a technical solution of integrating an artificial intelligence (AI) large model in a business system, aiming to realize automatic analysis and management of table data through natural language interaction. The user only needs to describe the analysis requirements through natural language, and the AI large model can automatically understand, execute and return accurate analysis results.

[0013] However, when the AI large model is directly applied to the table question and answer task in the real business scenario, it still faces serious technical challenges in ensuring the accuracy and reliability of the answer results, which are embodied in the following aspects: Lack of semantic planning capability: AI large models generally lack the ability to structure, multi-step decompose and plan user intent when faced with multi-step, cross-dimensional or implicit complex business logic queries. The lack of this capability leads to inherent bias in the model's overall understanding of the analysis task, which fundamentally affects the accuracy of the final results.

[0014] Logical unreliability of code generation: In the code generation phase, AI large models are prone to produce illusory code with correct syntax but incorrect business logic. The execution of such code often directly leads to program errors or incorrect results that are seriously inconsistent with the facts, posing a potential risk to the reliability of business decisions.

[0015] Lack of closed-loop iterative correction mechanism: AI large models usually output the first execution result of the code as the final answer. When the execution fails or the result is obviously abnormal, they lack the ability to analyze the cause of the error and iteratively correct the analysis strategy and generated code based on this feedback. This open-loop processing process makes it difficult to deal with complex and variable analysis scenarios in the real world.

[0016] In view of this, in another embodiment of the present application, a table data-based information processing method is provided, which builds a self-correcting closed-loop processing mechanism of understanding→translation→execution→correction through a cooperative strategy generation model, a code generation model, a code interpreter and a strategy correction model, so that the data analysis process is automated, and the user's data analysis requirements can be automatically converted into executable code scripts and the answer results are obtained. When the execution fails, no manual intervention is required, and the strategy correction model can be used for self-correction to regenerate the correct execution strategy, thereby ensuring efficient and continuous execution of the task and significantly improving the accuracy of the answer results.

[0017] By introducing a processing strategy as an intermediate state, the logical planning and code implementation are decoupled, separating the task of understanding user intent and logical planning from the task of translating logical planning into specific code. This allows for flexible adaptation or extension to support new programming languages without modifying the core intent understanding and strategy correction logic, greatly enhancing flexibility and expandability.

[0018] In addition, when the code script fails to execute, the embodiments of the present application do not simply modify the code script, but trace back to the root cause of the problem, i.e., the understanding of user intent or the possible bias in the processing strategy. By correcting the processing strategy, the accuracy of the subsequently generated code script is ensured, thereby improving the accuracy of the answer results.

[0019] As Figure 1As shown, an information processing system applying an information processing method based on table data can be provided, which can adopt a client / server architecture, the client 10 serving as an entry for user interaction and being responsible for receiving data analysis requirements of the user. The user can submit data analysis requirements through a text input box, voice input, etc. After receiving the data analysis requirements, the client 10 sends them to the server 11. The server 11 is responsible for responding to the requests of the client. An AI processing engine 110 driven by an AI large model can be deployed locally or on the cloud of the server, and can access current multi-table data as the basis for analysis, which can be obtained from the client and stored in a local database, a distributed file system or a cloud storage system. The AI large model can refer to an AI model with a relatively large parameter scale and capable of understanding natural description language. The large model can recognize various information forms such as text, voice, pictures, instructions, signals, etc., and can also understand the contextual association, deep semantics, sentiment tendency and implied intent of the data analysis requirements.

[0020] When the server receives the data analysis requirements from the client, it can utilize the AI processing engine 110 to perform a self-correcting closed-loop processing process to obtain the answer results. In the embodiments of the present application, the AI processing engine 110 can include a strategy generation model, a code generation model, a code interpreter and a strategy correction model. The strategy generation model is responsible for performing deep semantic analysis and intent recognition on the received data analysis requirements, and converting them into a structured, machine-readable processing strategy. The code generation model is responsible for receiving the initial processing strategy output by the strategy generation model, and translating the processing strategy into a code script in a specific programming language. The strategy correction model is responsible for receiving the execution error information returned when an error occurs during the execution of the code script, and based on this, correcting the error in the initial processing strategy, thereby generating a corrected processing strategy to provide higher quality input for a new round of code generation. The three models can be independently trained or jointly trained according to actual needs, with high flexibility, and through close cooperation, they jointly build a closed-loop data processing mechanism with self-correction and learning capabilities. After completing this process, the server returns the final answer results to the client for presentation.

[0021] The following describes in detail the process of the information processing method based on table data, as shown in the following figure: Figure 2 The method comprises: S1, obtaining current multi-table data and a data analysis requirement of the user on the current multi-table data.

[0022] S2, according to the current multi-table data and the data analysis requirement, calling a strategy generation model to perform intent recognition and mapping inference processing, obtaining a target processing strategy, the target processing strategy including execution steps required to complete the data analysis requirement.

[0023] S3, according to the target processing strategy, calling the code generation model for translation processing to generate a target code script.

[0024] S4, calling the code interpreter to execute the target code script to perform data analysis on the current multi-table data according to the execution steps, to obtain a data analysis result.

[0025] S5, if the data analysis result includes execution error information, according to the execution error information, calling the strategy correction model to correct the target processing strategy to obtain a new target processing strategy.

[0026] S6, according to the new target processing strategy, re-executing steps S3-S6 until a data analysis result that does not include execution error information is obtained.

[0027] The data analysis requirement can refer to the user's intention to analyze the current multi-table data in an unstructured and conversational manner. This data analysis requirement can cover various task types, such as data query requirements, data statistical calculation requirements, data integration analysis requirements, data prediction requirements, etc., which are not limited here.

[0028] The current multi-table data can refer to a collection of multiple data tables. Each data table can include metadata such as table name, column name, data type, etc. to describe its internal structure. Different data tables can be logically related through common fields. These data tables and their internal columns collectively constitute data entities that can be referenced. A data entity is the smallest semantic unit that can be independently identified and operated in the current multi-table data. It can be: a complete data table, or a specific data column in the data table. In addition, the current multi-table data can dynamically change as the data entity content and structure change. The same data analysis request initiated at different time points can result in different results due to different current multi-table data, ensuring that the analysis result can reflect the latest situation.

[0029] After obtaining the data analysis requirement, a strategy generation model can be called to process the data analysis requirement and the current multi-table data. The strategy generation model is responsible for receiving the data analysis requirement and converting it into a target processing strategy. The strategy generation model includes intent recognition and mapping inference processing.

[0030] The intent recognition can refer to a process of deeply understanding the semantic of the data analysis requirement to obtain the intent information, which can express the core elements of the analysis task and is mainly used to describe what the user wants to do. The intent information can be composed of multiple semantic units, for example, assuming that the data analysis requirement is to query the total sales of a certain region in the last month, the intent recognition can decompose it into a series of semantic units such as "perform a query and aggregation operation", "the object of the operation is sales", "the limited range of the operation is Beijing region and last month".

[0031] The mapping inference processing can refer to a process of converting the intent information obtained by the intent recognition from an abstract semantic level to a data level to generate a structured target processing strategy. The target processing strategy can be understood as a machine-readable specification of the data analysis requirement. The target processing strategy can include parameters required for task execution, which is used for subsequent translation into executable code scripts by the code generation model.

[0032] After obtaining the target processing strategy, the code generation model can be called to translate it to obtain the target code script. The code generation model receives the target processing strategy and translates it into a target code script in a specific programming language. For example, it can convert the target processing strategy into a Python code or an SQL query statement. The translation can be understood as a process of converting the abstract logical step sequence in the target processing strategy into the syntax and function call of the specific programming language. After the translation process is completed, the target code script can be obtained, which can refer to a source code file that can be directly executed by a computer.

[0033] After obtaining the target code script, the code interpreter can execute the target code script. The code interpreter can refer to an independent running environment capable of executing the code script, such as a Python interpreter or an SQL database engine, which is responsible for actual data calculation and processing.

[0034] After the code interpreter executes the target code script, the data analysis result can be obtained, which can refer to the output returned by the code interpreter after executing the code script. The data analysis result can include execution success information or execution error information. The execution error information can refer to an error report issued by the code interpreter when the target code script fails due to various reasons during execution.

[0035] If the data analysis result includes execution error information, the strategy correction model can be called to analyze the execution error information and the target processing strategy to infer a method for correcting the target processing strategy, and then adjust the target processing strategy to obtain a new target processing strategy. The new target processing strategy can refer to the processing strategy formed after adjustment, which can be sequentially called by the code generation model and the code interpreter for processing, starting a new round of code generation and execution code cycle to obtain the correct execution result, thereby generating the answer result corresponding to the data analysis requirement.

[0036] In the embodiments of the present application, a table data-based information processing method is provided. Through a collaborative strategy generation model, a code generation model, a code interpreter and a strategy correction model, a self-correcting closed-loop processing mechanism of understanding→translation→execution→correction is constructed, so that the data analysis process is automated, and the user's data analysis requirements can be automatically converted into executable code scripts and the answer results are obtained. When an error occurs, manual intervention is not required, and the strategy correction model can be used for self-correction to regenerate the correct execution strategy, thereby ensuring efficient and continuous execution of the task and significantly improving the accuracy of the answer results.

[0037] In an optional embodiment, according to the current multi-table data and the data analysis requirement, a strategy generation model is called to perform intent recognition and mapping inference processing to obtain a target processing strategy, including: performing intra-table mode extraction and inter-table relationship identification on the current multi-table data to obtain data structure information; inputting the data structure information and the data analysis requirement into the strategy generation model to perform the following operations in the strategy generation model: performing intent recognition on the data analysis requirement to obtain intent information, the intent information including operation intent, operation object and operation condition; performing mapping inference processing on the intent information based on the data structure information to generate a logical structure expressing the intent information, the logical structure including operation logic, data entity and constraint parameter; performing deconstruction and arrangement processing on the logical structure to obtain an ordered execution sequence, the ordered execution sequence including multiple execution steps; and parameterizing each execution step in the ordered execution sequence to generate a target processing strategy.

[0038] In the embodiments of the present application, the intra-table mode extraction can refer to the process of analyzing the internal structure of a single data table in the current multi-table data. It aims to identify and understand which table headers (fields) the data table is composed of, the data type of each table header (such as text, number, date), and other possible metadata (such as primary key, whether to allow null, association between different table headers, etc.). The inter-table relationship identification can refer to further identifying and determining how different data tables are connected to each other through common fields (foreign keys) after analyzing all single data tables. The purpose is to determine the logical association between data tables, for example, the “order table” is associated with the “customer table” through the “customer identifier”.

[0039] Through intra-table mode extraction and inter-table relationship identification on the current multi-table data, data structure information can be obtained. The data structure information can be understood as a comprehensive knowledge graph about the overall architecture of the current multi-table data, which contains both the internal details of each table and the association between all tables.

[0040] After obtaining the data structure information, the data structure information and the data analysis requirement can be input into a strategy generation model, and the following operations can be performed inside the strategy generation model: S21, performing intent recognition on the data analysis requirement to obtain intent information, the intent information including an operation intent, an operation object, and an operation condition.

[0041] Based on the above description, the intent information can be composed of multiple semantic units. In an embodiment of the present application, the semantic units can be further summarized into three core categories for subsequent mapping and processing. The three core categories are operation intent, operation object, and operation condition. The operation intent can refer to the data operation expected to be performed by the user, including but not limited to data operations such as aggregation, filtering, joining, sorting, and prediction. The operation object can refer to the object on which the operation intent acts, for example, "sales" in "sales of the Beijing region" is the operation object. The operation condition can be used to limit various constraints of the operation mode, including but not limited to time range, geographic location, numerical interval, sorting basis, and connection key.

[0042] By decomposing and summarizing the user's data analysis requirement into these three structured semantic units, the preliminary quantification and standardization of the user's intent information can be achieved, and the efficiency and accuracy of the subsequently generated processing strategy can be improved.

[0043] S22, performing mapping and reasoning processing on the intent information based on the data structure information to obtain a logic structure, the logic structure including operation logic, data entity, and constraint parameter.

[0044] Mapping and reasoning processing can refer to the process of mapping the intent information into a logic structure that can be understood by a machine. The logic structure can refer to the content dynamically generated or extracted by the strategy generation model based on the data structure information in order to realize the intent information. In an embodiment of the present application, the logic structure includes operation logic, data entity, and constraint parameter. For example, the "sales total" semantic unit is mapped to the amount column of the sales_table table; "Beijing region" is mapped to the filter condition "Beijing" of the city column in the region_table table; "summarize" is mapped to the SUM() aggregation function that can be understood by a machine; and "last month" is mapped to a specific date range, and the date range is associated with the sales data.

[0045] Operation logic can refer to the structured expression of the operation intent after mapping. For example, "sum" in the user's intent information can correspond to the SUM function as the operation logic; "find the maximum value" can correspond to the MAX function as the operation logic; and "group statistics" can correspond to a composite GroupBy-Aggregate operation logic.

[0046] Data entity can refer to the specific data entity that the operation object is precisely located in the data structure information. For example, the customer in the user intent information may specifically point to the customer information table or the customer name column in the table after mapping.

[0047] Constraint parameter can refer to the filtering condition after the operation condition is mapped. For example, "in Beijing" in the user intent information becomes a specific constraint parameter such as WHERE city = Beijing after mapping.

[0048] In the embodiments of the present application, the policy generation model can generate multiple independent logical structures according to the complexity of the user intent information, for example, one for association query and the other for aggregation statistics. These logical structures are not sorted and need to be logically arranged through the following steps.

[0049] S23, deconstructing and arranging the logical structure to obtain an ordered execution sequence, the ordered execution sequence comprising a plurality of execution steps.

[0050] Deconstruction can refer to the process of parsing and splitting the operation logic in the logical structure into non-divisible execution steps. One operation logic can correspond to one or more execution steps, for example, a complex operation logic containing GroupBy-Aggregate can obtain independent "grouping step" and "aggregation step" after deconstruction. Arrangement can refer to organizing and arranging a plurality of execution steps according to the relationship between different execution steps. For example, the JOIN step must be completed first before filtering the new table after joining.

[0051] The ordered execution sequence can refer to a logical flowchart with execution order, which can be composed of a plurality of execution steps to ensure that the data analysis task can be executed in the correct logical order. The execution step can refer to each independent, sequentially arranged execution unit in the ordered execution sequence. For example, "step one: filter the data meeting the conditions from table A", "step two: join the results of step one with table B", "step three: group the results after joining and sum up".

[0052] S24, parameterizing each execution step in the ordered execution sequence to generate a target processing strategy.

[0053] In the embodiments of the present application, parameterization can refer to replacing the abstract placeholder contained in each execution step in the ordered execution sequence with a real and specific value. This process ensures that the final generated processing strategy has no ambiguity and can be directly translated by the code generation model.

[0054] In an optional embodiment, the logical structure is deconstructed and arranged to obtain an ordered execution sequence, including: deconstructing operation logic of the logical structure into a plurality of execution steps, wherein each execution step corresponds to a logical operator; analyzing data dependency relationships between the plurality of execution steps to obtain an operator dependency graph; and performing topological sorting according to the operator dependency graph to obtain the ordered execution sequence.

[0055] In an embodiment of the present application, a logical operator can refer to a non-divisible minimum execution unit, and the logical operator can refer to a pre-defined standardized instruction oriented to a bottom-layer computing task. Each logical operator corresponds to a computing action. For example, a "filter" operator (FILTER) is only responsible for filtering data, and a "join" operator (JOIN) is only responsible for joining data.

[0056] Analyzing data dependency relationships between the plurality of execution steps can include analyzing inherent dependency relationships and data flow order between the plurality of execution steps. The inherent dependency relationships can refer to the fact that certain steps must be performed after other steps are completed. The data flow order can refer to the input-output relationship between adjacent steps in a processing flow. For example, a result set of a FILTER step serves as input data for a subsequent AGGREGATE step, and the direction of the data flow determines the order of operations.

[0057] The operator dependency graph can refer to a result of structuring and graphically representing data dependency relationships, and can be implemented as a directed acyclic graph including nodes and directed edges. A node can represent an independent execution step or a logical operator. A directed edge can represent a data dependency relationship. The direction of an arrow is from a parent step to a child step.

[0058] After obtaining the operator dependency graph, topological sorting can be performed thereon to obtain an ordered execution sequence. The ordered execution sequence can represent a clear instruction sequence, and a code generation model can translate the instruction sequence into executable code step by step.

[0059] The same logical operator can have different function names, parameter formats, or implementation manners in different programming language functions. For example, a logical operator representing "filtering" can correspond to the syntax df[df['column'] == value] in the Pandas library of Python, and correspond to the WHERE column = value clause in SQL. Based on this, embodiments of the present application can achieve complete decoupling of planning of data processing logic and specific code implementation by introducing a logical operator as an intermediate abstraction layer. The core task of a strategy generation model can be to output an ordered execution sequence composed of standardized logical operators and independent of a specific language, and a code generation model translates the ordered execution sequence into an executable code script in a specific target language.

[0060] In the embodiments of the present application, all data processing tasks can follow the same set of logical operators, ensuring that the underlying processing logic is consistent regardless of the language of the code generated, thereby ensuring the accuracy and reproducibility of the analysis results. When a new programming language needs to be supported, the core strategy generation model does not need to be changed. Only a new "translation module" needs to be added to the code generation model to enable it to map existing standard logical operators to the syntax of the new language. This greatly reduces the maintenance cost of the system and improves flexibility, making it easy to expand. In addition, the strategy generation model can focus on understanding user intent and making complex logical planning, without having to worry about the implementation details of the specific language. Similarly, the code generation model can focus on generating efficient and standardized target code. The two models each perform their own functions, allowing optimization of logical planning capabilities and code generation quality to be performed independently without interfering with each other.

[0061] In an optional embodiment, each logical operator includes a parameter slot to be filled, and parameterizing each execution step in the ordered execution sequence to generate a target processing strategy includes: traversing each execution step in the ordered execution sequence; identifying the parameter slot to be filled of the corresponding logical operator for the currently traversed execution step; obtaining the target parameter value corresponding to the parameter slot from the logical structure to fill the parameter slot, the target parameter value including the data entity and / or the constraint parameter, to generate a target processing strategy.

[0062] Each logical operator includes a parameter slot to be filled, which can refer to a variable name defined in advance inside the logical operator, waiting to be assigned a value, and can define the input data required by the logical operator. For example, a logical operator is FILTER(target_column, operator_symbol, value), and target_column, operator_symbol, and value in this operator are parameter slots to be filled. The specific filling values can be obtained from the logical structure, for example, for the target_column parameter slot, the corresponding data entity, such as salary, can be determined from the logical structure. For the operator_symbol parameter slot, the corresponding constraint parameter, such as ">", can be determined from the logical structure. For the value parameter slot, the corresponding constraint parameter, such as 5000, can be determined from the logical structure.

[0063] In the embodiments of the present application, if the parameter slot to be filled in the logic operator cannot find the corresponding target parameter value in the logic structure, a corresponding operation can be performed according to a preset rule. For example, if the missing is a core parameter necessary for executing the logic operator, the strategy generation can be interrupted and an error or a request for supplementary information can be returned to the user; if the missing is an optional parameter, the parameter slot can be filled with a predefined default value to ensure the integrity and executability of the target processing strategy. This differentiated processing mechanism based on the difference in parameter types can ensure the robustness when facing incomplete or ambiguous user intent information.

[0064] In an optional embodiment, the target processing strategy includes at least one parameterized execution step, each parameterized execution step corresponds to a logic operator and the filled target parameter value thereof, and according to the target processing strategy, a code generation model is called for translation processing to generate a target code script, including: for each parameterized execution step in the target processing strategy, the corresponding logic operator and the filled target parameter value are mapped to the corresponding code fragment in the target programming language; the code fragments are combined according to the arrangement order of the parameterized execution steps in the target processing strategy to generate the target code script.

[0065] In the embodiments of the present application, the target programming language can refer to the programming language used by the code expected to be generated, which is the language environment relied on when the data processing task is finally executed. The code fragment can refer to the partial source code of the target programming language generated according to a parameterized execution step in the target processing strategy.

[0066] In an optional embodiment, the target programming language is determined according to a preset execution environment configuration, and before the code generation model is called for translation processing to generate the target code script according to the target processing strategy, the method further includes: in the initialization stage, loading a preset configuration file, the configuration file including at least one execution environment and the corresponding programming language thereof; determining the currently activated execution environment according to the activation instruction in the configuration file; extracting the corresponding programming language from the currently activated execution environment to be used as the target programming language for this time of code generation.

[0067] In the embodiments of the present application, the initialization stage can refer to the period in which the information processing system performs preloading and environment preparation before generating the target processing strategy. Specifically, it can be the moment when the entire service or application is started. In this stage, the information processing system can perform a series of preparatory work to ensure that the subsequent code generation task can proceed smoothly. One of the key tasks is to load and parse the configuration file, obtain the defined environment information in the configuration file, and provide support for subsequent rapid decision-making. By completing the configuration loading in the initialization stage, repeated file reading operations can be avoided each time the user data analysis requirements are processed, thereby improving the response speed and running efficiency.

[0068] The configuration file can be used to store the static configuration information of the information processing system, to decouple the environment dependency of code generation from the main program logic, so that the supported programming languages of the information processing system can be flexibly switched or extended without modifying the core code. In a specific embodiment, the configuration file can adopt standard formats such as JSON, YAML or XML.

[0069] The execution environment can refer to an independent configuration defined in the configuration file. Each execution environment can represent a target platform to which the code script is to be deployed and run, and contains a set of related attributes. In the embodiments of the present application, each execution environment at least contains identification information and a programming language to be generated. Optionally, the execution environment can also contain description information, required third-party libraries, computing engine version, etc. When the code script needs to be executed, the corresponding code interpreter can be called according to the identification information of the selected execution environment, and the programming language, third-party libraries, etc. defined in the execution environment are used as the context of the interpreter runtime, so as to ensure that the code script is executed in the expected execution environment.

[0070] The activation instruction can refer to an instruction in the configuration file for specifying which execution environment is currently in effect. When the configuration file is parsed in the initialization stage, the activation instruction can be read to know which execution environment should be used as a reference for subsequent operations. The currently activated execution environment is the one selected from the defined execution environments according to the activation instruction.

[0071] After determining the currently activated execution environment, the corresponding programming language can be extracted therefrom and used as the target programming language for the current code generation task. In this way, high flexibility and automatic configuration of code generation can be achieved.

[0072] In an optional embodiment, before the code generation model is invoked for translation processing to generate the target code script according to the target processing strategy, the method further comprises: obtaining resource limit information of the target execution environment, the resource limit information comprising at least one of a memory size, a computation timeout threshold, or a third-party library dependency; determining whether the target processing strategy has an execution risk under the resource limit information, and if so, adaptively adjusting an execution step in the target processing strategy that causes the risk to obtain an adjusted target processing strategy for generating the target code script.

[0073] In the embodiments of the present application, the target execution environment can refer to a specific computing platform that will run the target code script.

[0074] The resource limit information can refer to a series of structured data that describes the running boundary of the target execution environment. These information can be obtained in various ways, for example: statically read from a configuration file, or communicate with the interface of the target execution environment to obtain real-time resource state information. The resource limit information can include but is not limited to: Memory size: can refer to the maximum amount of physical memory or virtual memory available to a single task in the target execution environment. This is the first risk factor to consider for logical operators that need to load a large amount of data into memory for processing.

[0075] Computation timeout threshold: can refer to the maximum time allowed for a single task to run continuously in the target execution environment. If a task executes beyond this computation timeout threshold, it may be forcibly terminated. This constitutes an execution risk for logical operators that may involve complex algorithms, iterative calculations, or processing of massive data.

[0076] Third-party library dependency: can refer to a list of installed external code libraries and their version numbers in the target execution environment. If the code snippet generated by the target processing strategy depends on a specific version of the library function, and the target environment does not install the library or the version is incompatible, it will cause the execution to fail, so this information also needs to be considered.

[0077] After obtaining the resource limit information of the target execution environment, it can be determined whether the target processing strategy has an execution risk based on this. The embodiments of the present application do not limit the determination method. In an optional embodiment, the potential resource consumption of each execution step in the target processing strategy can be quantitatively evaluated and compared with the obtained resource limit information to determine whether there is an execution risk. Specifically, the evaluation methods include but are not limited to the following: Rule-based evaluation: use a cost estimation model to establish resource consumption rules for logical operators.

[0078] Evaluation based on historical data: The actual resource consumption of similar operations in history can be recorded, and a machine learning model can be used to predict the consumption of the current operation.

[0079] Dependency check: The code templates that the logical operators in the target processing strategy will be mapped to can be traversed to parse the third-party libraries and functions required by them, and then matched with the third-party library dependency list of the target execution environment.

[0080] When the estimated resource consumption of any one execution step exceeds the corresponding limit threshold, or the library it depends on does not exist or the version does not match, it can be determined that the target processing strategy has an execution risk, and the target processing strategy needs to be adjusted adaptively according to the specific situation. Exemplarily, For the risk of exceeding the memory limit, the execution step that exceeds the content size can be decomposed into a series of sub-steps of "read part of the data-process-write temporary results-merge results". Based on this method, the instantaneous memory occupation of each step can be reduced.

[0081] For the risk of calculation timeout, an approximate calculation logical operator can be introduced to replace a complex logical operator. For example, HyperLogLog can be used for deduplication counting instead of the accurate but slow COUNT(DISTINCT); or the serial calculation can be adjusted to multiple independent calculation steps that can be executed in parallel.

[0082] For the risk of library dependency: Equivalent implementation methods of functions with execution risks can be found. For example, the target processing strategy uses a new feature function of pandas version 2.0, but the environment only has pandas 1.5. A set of execution steps composed of pandas 1.5 version function combinations can be used to replace the original single step.

[0083] After the above adjustment, the adjusted target processing strategy is obtained, which is logically equivalent to the original strategy, but its execution path has been optimized to adapt to the resource limit of the target execution environment. Then, the adjusted target processing strategy is sent to the code generation model, so as to ensure that the finally generated target code script has high environmental adaptability and execution success rate.

[0084] In an optional embodiment, in order to ensure sufficient awareness and analysis capability for exceptions in the code execution process, the code generation model can internally build a standard exception handling and logging mechanism when generating the target code script. When the code interpreter encounters any runtime error during the execution of the target code script, the built-in exception handling mechanism can capture the error to prevent the program from being unexpectedly interrupted. The logging mechanism can be immediately started to process the captured error details (such as error type, error information, stack trace, error code line number, etc.) to generate execution error information. The execution error information is returned by the code interpreter to the information processing system as the core basis for subsequent root cause analysis.

[0085] In an optional embodiment, the target processing strategy includes at least one parameterized execution step, and if the data analysis result includes execution error information, the target processing strategy is modified according to the execution error information to obtain a new target processing strategy, including: obtaining and analyzing the execution error information to obtain analyzed execution error information, the execution error information including error type and error location in the target code script; inputting the target processing strategy, the target code script, and the analyzed execution error information into the strategy modification model, and performing the following operations in the strategy modification model: associating the error location back to the parameterized execution step in the target processing strategy, and combining the error type and context information to infer the root cause of the error to generate a target modification strategy; and modifying the target processing strategy according to the target modification strategy to obtain a new target processing strategy.

[0086] In the embodiments of the present application, the execution error information can be analyzed to obtain error type and error location. The error type can refer to the classification of the error in the target programming language, which provides a classification guide for the nature of the problem, for example, whether it is a syntax error, a type mismatch error, or a reference error, etc. Each type points to a specific problem domain. The error location can refer to the specific location where the error occurs in the target code script, which provides the coordinates of the error at the execution level and is the starting point for backtracking analysis.

[0087] After obtaining the error location, the error location can be associated back to the parameterized execution step in the target processing strategy.

[0088] The context information is the basis for the inference of the strategy modification model, and can include information used or generated in the foregoing stages, such as data table metadata, target processing strategy, execution environment, etc.

[0089] The target modification strategy can refer to operation information for adjusting the target processing strategy, and is used to repair an execution step with a defect in the target processing strategy. The target modification strategy includes at least one modification instruction, and each modification instruction can include an operation type, an operation target, and operation content. The operation target can be used to uniquely identify and locate an identifier of a specific element or position of an execution step that needs to be modified and parameterized in the target processing strategy, to ensure that the modification action can accurately act on the expected execution step or internal parameters thereof. The operation type can refer to a predefined enumeration value, which defines what kind of modification action is to be performed on the operation target, such as updating, adding, or deleting a basic operation. The operation content can provide detailed data necessary for performing a specific modification action, and the content and structure thereof depend on the specific definition of the operation type.

[0090] In an optional embodiment, the error location backtracking is associated with the parameterized execution step in the target processing strategy, and the root cause of the error is inferred in combination with the error type and context information, which can include: locating the error in the target processing strategy according to the error location; analyzing the error type to generate a preliminary hypothesis; verifying the preliminary hypothesis by using context information, which includes upstream execution steps of the error execution step in the target processing strategy, and metadata about the input data table; if the verification result shows that the preliminary hypothesis is inconsistent with the context information, locating the root cause of the error according to the inconsistency; if the verification result shows that the preliminary hypothesis is not inconsistent with the context information, expanding the analysis range, introducing additional context information, and generating a new hypothesis until the root cause is located.

[0091] In the embodiments of the present application, the additional context information can include: ① data sample context: instead of relying only on metadata, several rows of sample data in the input data table are extracted, the logic before the error step is simulated, the form of intermediate data is observed, and whether there are abnormal values, null values, or data that do not conform to the expected format is checked. ② Execution environment context: check more detailed environment information, such as the version of the third-party library.

[0092] The deeper new hypothesis can include: ① algorithm logic hypothesis: a hypothesis about the inapplicability of the algorithm itself can be generated. ② Cross-step logic association hypothesis: the error can not be caused by a single step, but a side effect generated by the interaction between multiple steps. The logic chain of the entire target processing strategy can be analyzed.

[0093] According to the new hypothesis, verification is performed again until the root cause is found.

[0094] After the root cause is obtained, the target modification strategy can be generated based on the target modification strategy.

[0095] In an optional embodiment, the adjustment of the target processing strategy according to the target modification strategy can include: parsing the target modification strategy to obtain a modification instruction, the modification instruction including an operation type, an operation target, and an operation content; locating a target execution step to be modified among a plurality of execution steps of the target processing strategy according to the operation target in the modification instruction; and performing a corresponding modification operation on the target execution step according to the operation type and the operation content in the modification instruction, thereby generating a new target processing strategy.

[0096] According to the new target processing strategy, the code generation model and the code interpreter are re-invoked to generate a response result corresponding to the data analysis requirement.

[0097] In an optional embodiment, the invocation of the code generation model for translation processing to generate a target code script according to the target processing strategy can include: based on a current data analysis requirement, filtering a target historical data analysis requirement with a similarity to the current data analysis requirement exceeding a first threshold from an information knowledge base, and obtaining a corresponding target historical code script, a target generation timestamp, and a metadata change event occurring after the target generation timestamp; based on an analysis strategy, judging whether the metadata change event has a substantial impact on an execution logic of the target historical code script; if not, directly reusing or adaptively adjusting the target historical code script as the target code script; and if so, invoking the code generation model for translation processing to generate a target code script according to the target processing strategy.

[0098] In the embodiments of the present application, the current data analysis requirement can refer to a data analysis requirement currently input by a user. The information knowledge base can refer to a persistent storage database storing historical data analysis requirements, historical code scripts, generation timestamps, and metadata change events. The target historical data analysis requirement can refer to a historical data analysis requirement with a similarity to the current data analysis requirement exceeding a first threshold. The target historical code script can refer to a historical code script corresponding to the target historical data analysis requirement. The target generation timestamp can refer to a generation time of the target historical code script. The metadata change event can refer to a structural log record for operating on the structure of multi-table data. The metadata change event can include an event type, a change entity, a change content, and a timestamp of event occurrence.

[0099] The analysis strategy can refer to a preset algorithm or rule set for judging whether the metadata change event has a substantial impact on the execution logic of the target historical code script.

[0100] The judgment of whether the set of change events has a substantial impact can include: performing static analysis on the target historical code script, extracting all data entities directly referenced by the target historical code script to form a code dependency set; traversing the metadata change event, extracting change entities directly operated by the metadata change event, and collecting all the change entities to form a change entity set; judging whether the intersection of the code dependency set and the change entity set is non-empty; and if the intersection is non-empty, determining that a substantial impact is generated.

[0101] In the embodiments of the present application, when the entity affected by the occurred metadata change event appears in the code dependency set, the entity may have an impact on the execution of the code. This greatly filters out a large number of irrelevant changes and avoids unnecessary code regeneration. The complex logical judgment problem is converted into a simple set operation problem, so that the entire decision-making process can be efficiently and accurately automated, significantly improving the decision-making efficiency and accuracy.

[0102] In an optional embodiment, before or after receiving the current data analysis requirement, the method further includes: inputting the current multi-table data into a question recommendation model, and performing the following operations inside the question recommendation model: performing data feature mining based on the current multi-table data to extract key data features; constructing a user behavior model based on historical user interaction data recorded in a behavior knowledge base; fusing the key data features and the user behavior model to generate a set of candidate recommended questions; and presenting the candidate recommended questions to the user.

[0103] In the embodiments of the present application, data feature mining can refer to multi-dimensional mining of features of the current multi-table data, such as statistical distribution dimension, association rule dimension, and anomaly point detection dimension.

[0104] The user behavior model can be used to express the preferences and habits of users or user groups, so that the question recommendation is more personalized and predictable. The user behavior model is dynamically updateable, and can continuously learn the interactive behaviors of users to adaptively adjust the model parameters to reflect the evolution of user preferences. The user behavior modeling can be obtained by constructing a user portrait model and mining user query sequence patterns.

[0105] In an implementation manner, constructing the user portrait model can include: generating an analysis role label for a user or a user group based on data fields and key indicators involved in historical user interaction data recorded in the behavior knowledge base; and quantifying attention scores of the user or the user group to different data indicators, analysis dimensions, or data sources based on frequencies of query, click, or download and other interactive behaviors in the historical user interaction data.

[0106] In an implementation, mining the user query sequence pattern can include learning transition probabilities between different analysis steps in the user's data exploration using a Markov chain model or a recurrent neural network model to capture the user's coherent analysis thought chain.

[0107] After obtaining the key data features and the user behavior model, the key data features and the user behavior model can be fused to generate a set of candidate recommended questions. In an implementation, generating the set of candidate recommended questions can include screening high-value features from the key data features, the high-value features can include statistical distribution outliers or strong association rules; generating a first type of candidate recommended questions that match the high-value features and the analysis role labels and the attention scores in the user portrait model; predicting the analysis step that the user is likely to perform next based on the transition probabilities learned from the user query sequence pattern; generating a second type of candidate recommended questions for guiding the user to perform in-depth exploration in combination with the predicted analysis step and the relevant key data features; and merging and sorting the first type of candidate recommended questions and the second type of candidate recommended questions to generate a final set of candidate recommended questions.

[0108] When the user sees the candidate recommended questions, the user can directly click to input the question as a new current data analysis requirement and start the subsequent processing flow. This greatly reduces the threshold for the user to perform in-depth data exploration and achieves the ultimate goal of effectively guiding the user to discover potential business risks and opportunity points.

[0109] In the embodiments of the present application, in the initial stage of interaction, instead of passive waiting, a set of high-quality default recommended questions are actively presented. These questions are generated based on preliminary mining of the current multi-table data or general analysis paradigms, and are intended to serve as a starting point for user analysis, effectively stimulate the user's exploration ideas, and help the user quickly familiarize themselves with the available data. When the user inputs an explicit data analysis requirement through natural language (or clicks on a recommended question), the data analysis requirement and the associated current multi-table data can be input to generate a corresponding data analysis result.

[0110] In an optional embodiment, after obtaining the data analysis result, the method further comprises: parsing and extracting a user-specified output format identifier from the data analysis requirement; the output format identifier comprises at least one of a table, a chart or text; performing format processing on the data analysis result according to the output format identifier to obtain to-be-presented data, wherein, in the case that the output format identifier comprises a table, the to-be-presented data comprises the data analysis result; in the case that the output format identifier comprises a chart, mapping the data analysis result into a set of chart rendering parameters required for generating a chart to obtain the to-be-presented data, the to-be-presented data comprising the chart rendering parameters; in the case that the output format identifier comprises text, converting the data analysis result into a piece of text summary to obtain the to-be-presented data, the to-be-presented data comprising the text summary; and invoking a rendering engine to present the to-be-presented data on a user interface.

[0111] In the embodiments of the present application, after obtaining the data analysis result, an appropriate visualization chart type (such as a table, a line chart, a column chart, etc.) can be intelligently selected for dynamic rendering, and complex data can be converted into intuitive visual insights. Meanwhile, the front-end interface provides convenient one-key copy, export (such as Excel, CSV format) and print functions. It is ensured that the analysis result can be easily and seamlessly integrated into the user's report, presentation or other business documents, greatly improving the flow efficiency and reuse value of the data analysis result.

[0112] In an optional embodiment, an agent can be integrated in the user interface, which can perceive the environment to realize an end-to-end automated process from context data collection, natural language query, intelligent analysis to multi-modal result presentation. The agent can exist in various software forms, and the following will be described in detail taking the agent as an intelligent assistant form as an example: As shown in Figure 3 The intelligent assistant is presented as an instant available interactive entrance in the client interface, such as a floating button, a side bar or an icon integrated in a specific toolbar, and integrates the following core functional modules: Intelligent extraction and analysis module: the intelligent assistant can access the structured multi-table data of the client in real time and dynamically, has context perception capability, can automatically identify and capture the core table data in the current business scenario of the user without manual export or upload, and calls the server to perform deep analysis to generate insight results.

[0113] Natural language interaction module: provides a conversational interface allowing the user to submit data analysis requirements by inputting daily language text or voice.

[0114] Result optimization and presentation module: responsible for presenting the analysis results returned by the server to the user in various forms (text, chart, voice), and providing convenient functions such as result optimization, copying and exporting.

[0115] The user activates the assistant by clicking or equivalent interaction events, which starts the information processing process of the AI processing engine.

[0116] When the user activates the intelligent assistant and submits an analysis request, the client and server will jointly perform the following automated processes: Stage one: collection of current multi-table data and data analysis request of the client The client script first scans the document object model of the current page, automatically identifies and locates one or more target data tables within the user's current view through preset rules.

[0117] After locating the target data table, the client script can extract its structured information. Specifically, the table header information of all columns can be obtained by parsing the table header part of the target data table, and the data in each cell of each row can be extracted by traversing the table body part, forming the current multi-table data.

[0118] Monitor the user's input in the intelligent assistant to obtain the data analysis requirements.

[0119] Stage two: data transmission and intelligent analysis on the server side The client sends the current multi-table data and data analysis request to the server by calling the interface.

[0120] After receiving the current multi-table data and data analysis request, the server calls the strategy generation model, code generation model, strategy correction model and other models for processing, generating structured data analysis results. In this process, the multi-table data obtained can be automatically transmitted to the AI processing engine. The AI processing engine can standardize the heterogeneous multi-table data, identify its data types, field meanings and internal logical relationships. Then, an efficient multi-dimensional indexing model can be built for the parsed data. This process not only stores the data simply, but also deeply understands and structures it, laying a foundation for fast and accurate query, ensuring efficient and stable query process.

[0121] Stage three: result presentation and multi-modal interaction of the client The client receives the structured data analysis results returned by the server, which can be visualized and output according to user preferences or system presets: Text and chart presentation: display the core conclusions in the form of a text summary, and render the structured data into appropriate charts (such as pie charts, bar charts) to achieve intuitive visual insight.

[0122] Voice broadcast: Provide voice broadcast options, call the text-to-speech engine of the client, convert the text result into voice output, meet the information acquisition needs of users in inconvenient screen scenarios.

[0123] Re-answer optimization: If the user is not satisfied with the data analysis result, you can click the regenerate button. The client will send a new request containing the original request and optimization instructions to the server to get more accurate answers.

[0124] One-click copy: Provide a convenient one-click copy function, allowing users to quickly copy text summaries, table data or chart images to the clipboard, seamlessly connecting to report writing or other applications.

[0125] Further, as shown in Figure 4 , the intelligent assistant can provide intelligent question recommendations for the current multi-table data before or after receiving the current data analysis requirement.

[0126] Further, as shown in Figure 5 , the intelligent assistant can provide dynamic queries for the current multi-table data. Users do not need to master complex query syntax, but can input query requests in natural language form. Accurately locate the target data, and feed back the data analysis result to the user in a natural language manner, significantly reducing the operation threshold and improving data acquisition efficiency.

[0127] Further, as shown in Figure 6 , the intelligent assistant can provide integrated analysis capabilities for the current multi-table data, which can automatically perform a series of advanced analysis operations including but not limited to data aggregation calculation, classification statistics, time series analysis, and outlier detection. And present the change rule of key indicators in an intuitive way. Through this mechanism, users do not need to have a professional data analysis background, but can easily complete the whole process from raw data to business insight, greatly improving data utilization efficiency and decision response speed.

[0128] In the embodiments of the present application, by introducing natural language interaction, AI-driven dynamic analysis and intelligent question recommendation mechanism, the efficiency and intelligence level of table data processing in business scenarios are significantly improved, and the pain points of low efficiency, dependence on manual operation and single analysis capability are solved.

[0129] The specific technical effects are as follows: Natural language-driven efficient data query: Users can directly submit query requirements in everyday language (such as "Count the top 5 devices with the most equipment failure times in 2025") without needing to master complex instructions or operation procedures. Combined with context understanding, intent recognition, and entity extraction technology, the target data is accurately located and structured results are returned, with response time shortened to seconds and operation threshold reduced by more than 90%. Supports multi-dimensional free combination of query conditions (such as fields, time range, keywords, etc.), meets individual needs in different business scenarios, and improves data retrieval efficiency by 3-5 times.

[0130] Multi-dimensional intelligent data analysis: Based on a deep learning architecture, the data analysis engine can automatically perform complex tasks such as aggregate calculations (such as sum, average), classification statistics, time series analysis (such as trend prediction), and outlier detection. The analysis results are presented in natural language to visually show the changes in key indicators (such as "Device A's failure rate has increased by 20% compared to last month"), helping users quickly identify business risks and opportunity points, and improving decision-making efficiency by more than 50%.

[0131] Intelligent problem recommendation and predictive interaction: AI models actively analyze data features (such as key indicator fluctuations, abnormal patterns) to generate multiple candidate recommended questions (such as "Which devices have the highest maintenance costs?" "Is the failure cycle of device B stable?"), guiding users to explore data value in depth. Through user behavior modeling, the problem recommendation logic is optimized, with a problem identification accuracy of over 85%, reducing user trial and error costs and improving human-computer interaction experience.

[0132] Real-time dynamic updating and high scalability: Supports real-time data synchronization and dynamic access, ensuring that query results are always based on the latest business status, adapting to high-frequency, real-time analysis needs. Modular design adapts to various industry scenarios (such as equipment management, production monitoring, supply chain analysis), allowing for quick deployment to different enterprise information systems, reducing custom development costs.

[0133] The following provides an application scenario case: Scenario: Fault analysis of the equipment management system Background: A manufacturing enterprise has over 1000 devices and needs to regularly analyze fault data to optimize maintenance strategies.

[0134] Current pain points: Manual statistical failure records take 3 hours per time and are prone to missing key information. Traditional reports cannot dynamically adjust output structures, making it difficult to meet management's analysis needs in different dimensions.

[0135] Solution: Natural language query: User input: "What are the top 10 devices with the most equipment failure times in Q2 2025?" Automatically extract fault records from the database, generate a dynamic ranking table and mark abnormal values (such as the failure rate of a certain device exceeding the threshold).

[0136] Intelligent problem recommendation: "What is the average maintenance time of these devices?", "How to reduce the maintenance cost of high-failure devices?".

[0137] Dynamic result output: User selects "generate table + text report": output top 10 device list and maintenance cost analysis, and provide voice broadcast function. Users can directly copy the results to the meeting report, or directly compare the fault trends of different devices through charts.

[0138] Effect: Data analysis time is reduced from 3 hours to 30 seconds, and labor cost is reduced by 90%.

[0139] Through active problem recommendation, it is found that 3 devices have potential failure risks, maintenance is arranged in advance to avoid economic loss of more than 500,000 yuan.

[0140] Figure 7 The internal structure of the processing device provided for another exemplary embodiment of the present application is shown. As shown in Figure 7 The processing device includes a memory 54, a processor 55, and a communication component 56.

[0141] The memory 54 is used to store computer programs, and can be configured to store various data to support operations on the processing device. Examples of these data include instructions for any application or method operating on the processing device, contact data, phonebook data, messages, pictures, videos, etc.

[0142] The memory 54 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0143] Processor 55, coupled to memory 54, is used to execute computer programs in memory 54 for: acquiring current multi-table data and user data analysis requirements for the current multi-table data; based on the current multi-table data and the data analysis requirements, invoking a strategy generation model to perform intent recognition and mapping reasoning processing to obtain a target processing strategy, the target processing strategy including execution steps required to complete the data analysis requirements; based on the target processing strategy, invoking a code generation model to perform translation processing to generate a target code script; invoking a code interpreter to execute the target code script to perform data analysis on the current multi-table data according to the execution steps to obtain data analysis results; if the data analysis results include execution error information, invoking a strategy correction model to correct the target processing strategy based on the execution error information to obtain a new target processing strategy; based on the new target processing strategy, re-executing the steps of generating the target code script, obtaining the data analysis results, and determining whether the data analysis results include execution error information, until a data analysis result that does not include execution error information is obtained.

[0144] The detailed implementation methods and beneficial effects of each step in this embodiment have been described in detail in the foregoing embodiments, and will not be elaborated here.

[0145] Furthermore, such as Figure 7 As shown, the processing device also includes other components such as a communication component 56, a display 57, a power supply component 58, and an audio component 59. Figure 7 The diagram only shows some components and does not imply that the processing device includes only these components. Figure 7 The components shown.

[0146] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed, can perform the steps that can be executed by a processing device in the above method embodiments.

[0147] The above Figure 7 The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA) technology, Ultra-Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0148] The above Figure 7 The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of the touch or swipe action, but also the duration and pressure associated with the touch or swipe operation.

[0149] The above Figure 7 The power supply component provides power to various components within the device in which it resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which it resides.

[0150] The above Figure 7 The audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0151] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product 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.

[0152] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0153] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0154] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 Figure 1 The steps of the function specified in one or more boxes.

[0155] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0156] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0157] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0158] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, 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 a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0159] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method of information processing based on tabular data, characterized by, The method comprises: S1, acquiring current multi-table data and a data analysis requirement of a user on the current multi-table data; S2, according to the current multi-table data and the data analysis requirement, calling a strategy generation model to perform intention recognition and mapping inference processing, to obtain a target processing strategy, the target processing strategy comprising execution steps required to complete the data analysis requirement; S3, according to the target processing strategy, calling a code generation model to perform translation processing to generate a target code script; S4, calling a code interpreter to execute the target code script, to perform data analysis on the current multi-table data according to the execution steps, to obtain a data analysis result; S5, if the data analysis result comprises execution error information, according to the execution error information, calling a strategy correction model to correct the target processing strategy, to obtain a new target processing strategy; S6, according to the new target processing strategy, re-executing steps S3-S6 until a data analysis result not comprising execution error information is obtained.

2. The method of claim 1, wherein, According to the current multi-table data and the data analysis requirement, calling a strategy generation model to perform intention recognition and mapping inference processing, to obtain a target processing strategy, comprises: performing intra-table mode extraction and inter-table relationship recognition on the current multi-table data, to obtain data structure information; inputting the data structure information and the data analysis requirement into the strategy generation model, and performing the following operations in the strategy generation model: performing intention recognition on the data analysis requirement, to obtain intention information, the intention information comprising operation intention, operation object and operation condition; based on the data structure information, performing mapping inference processing on the intention information, to generate a logical structure expressing the intention information, the logical structure comprising operation logic, data entity and constraint parameter; performing deconstruction and arrangement processing on the logical structure, to obtain an ordered execution sequence, the ordered execution sequence comprising a plurality of execution steps; parameterizing each execution step in the ordered execution sequence, to generate the target processing strategy.

3. The method of claim 2, wherein, Performing deconstruction and arrangement processing on the logical structure, to obtain an ordered execution sequence, comprises: deconstructing the operation logic of the logical structure into a plurality of execution steps, wherein each execution step corresponds to a logical operator; analyzing data dependency relationships between the plurality of execution steps, to obtain an operator dependency graph; performing topological sorting according to the operator dependency graph, to obtain an ordered execution sequence.

4. The method according to any one of claims 1-3, characterized in that, Before according to the target processing strategy, calling a code generation model to perform translation processing to generate a target code script, further comprising: acquiring resource limitation information of a target execution environment, the resource limitation information comprising at least one of memory size, calculation timeout threshold or third-party library dependency; judging whether the target processing strategy has execution risk under the resource limitation information, if so, performing adaptive adjustment on an execution step causing risk in the target processing strategy, to obtain an adjusted target processing strategy for generating a target code script.

5. The method of claim 1, wherein, The target processing strategy includes at least one parameterized execution step, if the data analysis result includes execution error information, according to the execution error information, a strategy correction model is called to correct the target processing strategy to obtain a new target processing strategy, including: Obtain and parse the execution error information to obtain parsed execution error information, the execution error information includes error type and error location in the target code script; The target processing strategy, the target code script and the parsed execution error information are input into the strategy correction model, and the following operations are performed inside the strategy correction model: The error location is associated with the parameterized execution step in the target processing strategy, and the root cause of the error is inferred by combining the error type and context information to generate a target correction strategy; According to the target correction strategy, the target processing strategy is corrected to obtain a new target processing strategy.

6. The method of claim 1, wherein, According to the target processing strategy, a code generation model is called to perform translation processing to generate a target code script, including: Based on the current data analysis requirement, the target historical data analysis requirement with similarity exceeding a first threshold value is screened out from the information knowledge base, and the corresponding target historical code script, target generation timestamp, and metadata change event occurring after the target generation timestamp are obtained; Based on the analysis strategy, it is judged whether the metadata change event has a substantial influence on the execution logic of the target historical code script; If there is no substantial influence, the target historical code script is directly reused or adaptively adjusted as the target code script; If there is a substantial influence, according to the target processing strategy, the code generation model is called to perform translation processing to generate a target code script.

7. The method of any one of claims 1-3, 5-6, wherein, Before or after receiving the current data analysis requirement, further including: Input the current multi-table data into the question recommendation model, and perform the following operations inside the question recommendation model: Based on the current multi-table data, data feature mining is performed to extract key data features; Based on the user historical interaction data recorded in the behavior knowledge base, a user behavior model is constructed; Fuse the key data features and the user behavior model to generate a set of candidate recommended questions; The candidate recommended questions are presented to the user.

8. The method of any one of claims 1-3, 5-6, wherein, After obtaining the data analysis result, further including: From the data analysis requirement, parse and extract the output format identifier specified by the user; The output format identifier includes at least one of table, chart or text; According to the output format identifier, the data analysis result is formatted to obtain to-be-presented data, wherein, in the case that the output format identifier comprises a table, the to-be-presented data comprises the data analysis result; in the case that the output format identifier comprises a chart, the data analysis result is mapped to a set of chart rendering parameters required for generating a chart to obtain the to-be-presented data, and the to-be-presented data comprises the chart rendering parameters; in the case that the output format identifier comprises a text, the data analysis result is converted into a text summary to obtain the to-be-presented data, and the to-be-presented data comprises the text summary; A rendering engine is invoked to present the to-be-presented data on a user interface.

9. An electronic device, comprising: Comprise: a memory and a processor; The memory is used to store a computer program; the processor is coupled with the memory and is used to execute the computer program to implement the steps in the method of any one of claims 1-8.

10. A computer readable storage medium storing computer programs / instructions, characterized in that, When the computer program / instruction is executed by the processor, the processor can implement the steps in the method of any one of claims 1-8.

11. A computer program product, characterised in that, Comprise: a computer program / instruction, when the computer program / instruction is executed by the processor, the processor can implement the steps in the method of any one of claims 1-8.

Images