Method, device, electronic device and computer program product for obtaining indicator data

Abstract

Embodiments of the present disclosure relate to methods, devices, electronic devices, and computer program products for obtaining indicator data. The method includes generating a second query input using a language model based on a first query input, and obtaining a first indicator set, a first dimension set, and a first knowledge set from a preconfigured knowledge base based on the second query input. The method also includes determining a second indicator set, a second dimension set, and a second knowledge set using a language model, and generating a domain-specific language DSL code for obtaining indicator data using the language model. In addition, the method also includes obtaining target indicator data using a data acquisition service in an indicator platform based on the DSL code. Therefore, through the embodiments of the present disclosure, the efficiency and accuracy of indicator queries are improved, and the user experience is significantly improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more particularly to a method, device, electronic device and computer program product for obtaining indicator data. Background Art

[0002] With the rapid development of artificial intelligence technology, language models are becoming increasingly important in various fields. These models make human-computer interaction more intuitive and efficient through natural language processing. Users can use natural language to query and instruct, and the system can quickly understand and respond, greatly improving work efficiency. The application of language models in data analysis and content generation has gradually become an indispensable technology.

[0003] In addition, as data-driven decision-making becomes more prevalent in all walks of life, the importance of indicator platforms is becoming increasingly prominent. Through indicator platforms, enterprises can efficiently extract, analyze and display key business data. Especially in scenarios involving complex data processing, aggregation and comparative analysis, indicator platforms can provide automated data retrieval services and multi-dimensional analysis capabilities, greatly improving data processing efficiency. Summary of the invention

[0004] Embodiments of the present disclosure provide a method, an apparatus, an electronic device, and a computer program product for acquiring indicator data.

[0005] According to a first aspect of the present disclosure, a method for obtaining indicator data is provided. The method includes generating a second query input using a language model based on a first query input from a user, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form. The method also includes obtaining a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a pre-configured knowledge base related to the indicator platform based on the second query input. The method also includes determining a second indicator set, a second dimension set, and a second knowledge set using a language model based on the second query input, the first indicator set, the first dimension set, and the first knowledge set. The method also includes generating a domain specific language (DSL) code for obtaining indicator data using a language model based on the second query input, the second indicator set, the second dimension set, and the second knowledge set. In addition, the method also includes obtaining target indicator data using a data acquisition service in the indicator platform based on the DSL code.

[0006] According to a second aspect of the present disclosure, a device for obtaining indicator data is provided. The device includes a query input generation module, which is configured to generate a second query input based on a first query input from a user using a language model, and the first query input indicates obtaining indicator data from an indicator platform in a natural language form. The device also includes a first information acquisition module, which is configured to obtain a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a pre-configured knowledge base related to the indicator platform based on the second query input. The device also includes a second information determination module, which is configured to determine a second indicator set, a second dimension set, and a second knowledge set based on the second query input, the first indicator set, the first dimension set, and the first knowledge set using a language model. The device also includes a query code generation module, which is configured to generate a domain-specific language DSL code for obtaining indicator data based on the second query input, the second indicator set, the second dimension set, and the second knowledge set using a language model. In addition, the device also includes a target indicator acquisition module, which is configured to obtain target indicator data based on the DSL code using a data acquisition service in the indicator platform.

[0007] According to a third aspect of the present disclosure, an electronic device is provided, comprising a processor and a memory coupled to the processor, wherein the memory has instructions stored therein, and when the instructions are executed by the processor, the electronic device executes the method according to the first aspect.

[0008] In a fourth aspect of the present disclosure, a computer program product is provided. The computer program product is tangibly stored on a non-transitory computer readable medium and comprises computer executable instructions, which when executed cause a computer to perform the steps of the method of the first aspect of the present disclosure.

[0009] The purpose of this Summary is to introduce a selection of concepts in a simplified form that are further described in the Detailed Description below. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to limit the scope of the claimed subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0011] Figure 1 A schematic diagram showing an example environment in which devices and / or methods according to embodiments of the present disclosure may be implemented;

[0012] Figure 2A flow chart of a method for obtaining indicator data according to an embodiment of the present disclosure is shown;

[0013] Figure 3 A flow chart of an intelligent data acquisition process according to an embodiment of the present disclosure is shown;

[0014] Figure 4 A schematic diagram showing the architecture of a self-service analysis tool for an indicator platform according to an embodiment of the present disclosure;

[0015] Figure 5 A schematic diagram showing a process of high-cardinality supplementary recall according to an embodiment of the present disclosure is shown;

[0016] Figure 6 A block diagram of a device for acquiring indicator data according to an embodiment of the present disclosure is shown;

[0017] Figure 7 A block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0018] Throughout the drawings, the same or similar reference numbers denote the same or similar elements. DETAILED DESCRIPTION

[0019] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the type, scope of use, usage scenarios, etc. of the personal information (such as voice) involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0020] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0021] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects, unless explicitly stated. Other explicit and implicit definitions may also be included below.

[0022] In relevant indicator acquisition and data query systems, users often initiate query requests in natural language. However, traditional systems have difficulty effectively processing complex natural language queries, especially in query scenarios involving multiple rounds of dialogue and contextual associations. Existing systems are usually unable to accurately parse user intentions, resulting in inaccurate recall of indicators and dimensions. In addition, because the knowledge base of existing systems is not perfect and cannot effectively integrate multiple business knowledge, the system is unable to provide comprehensive and accurate query results when faced with complex business needs.

[0023] To this end, this application proposes a solution for obtaining indicator data. First, the user query can be rewritten using a language model, and then a recall is performed based on the user query using a pre-configured vector library, and a second recall is performed using the language model based on the result of the first recall, and a domain-specific language DSL code is generated using the language model based on the result of the second recall, and finally the DSL code is executed using the data acquisition service to obtain the target indicator data.

[0024] Therefore, through the embodiments of the present disclosure, by combining the knowledge base and the language model for multiple rounds of recall, the indicators, dimensions and knowledge related to the user query are accurately obtained, thereby providing reliable and rich context information for the generation of DSL code, ensuring that the generated DSL code is more accurate and can effectively reflect the user's query needs. This solution not only improves the efficiency and accuracy of query and data extraction, but also simplifies the user's operation process and significantly improves the user's overall usage experience.

[0025] Figure 1 Schematic diagram of an example environment in which the device and / or method according to an embodiment of the present disclosure may be implemented. Figure 1 As shown, the example environment 100 may include a computing device 110, which may be a user terminal, a mobile device, a computer, etc., or a computing system, a single server, a distributed server, or a cloud-based server. The computing device 110 may receive a query input 112. The query input 112 may be a request that a user may make in natural language to obtain indicator data from a data platform. For example, a user query may be "weekly average of clicks in the last 30 days", where "clicks" is an indicator, "last 30 days" is a specific value in the time dimension, and "weekly average" is a summary method. The computing device 110 may generate a query input 116 based on the query input 112 from the user using a language model 114. For example, the query input 112 may be rewritten for multiple rounds of conversations using a language model to more accurately understand the user's query intent. During the rewriting process, the language model 114 may optimize the expression in the query input 112 in combination with contextual information and historical query records, thereby generating a query input 116 that is more in line with the user's expectations.

[0026] The computing device 110 may obtain the indicator set 122, dimension set 124, and knowledge set 126 related to the query input 116 from the pre-configured knowledge base 118 related to the indicator platform based on the query input 116. For example, the computing device 110 may retrieve the indicators, dimensions, and knowledge matching the query content in the knowledge base 118 through vector matching. The computing device 110 may further optimize and determine the indicator set 126, dimension set 128, and knowledge set 130 based on the query input 116, the indicator set 120, the dimension set 122, and the knowledge set 124 using the language model 114. For example, the computing device 110 may perform a secondary search through the language model 114 to perform deep semantic understanding and association analysis on the initially recalled data, thereby generating an accurate set that better meets the user's query requirements.

[0027] The computing device 110 can generate a domain specific language DSL code 132 for obtaining indicator data based on the query input 116, the indicator set 126, the dimension set 128, and the knowledge set 130 using the language model 114. For example, the computing device 110 can automatically generate the DSL code 132 by analyzing the semantic information in the query input 116 and combining the recalled indicators, dimensions, and knowledge. The computing device 110 can obtain the target indicator data 134 based on the DSL code 132 using the data retrieval service in the indicator platform. For example, the DSL code 132 can be executed by the data retrieval service to extract the target indicator data 134 related to the user query from a predefined data source. The data retrieval service can accurately identify the required indicators according to the logic in the DSL code and perform related data operations.

[0028] It should be understood that the architecture and functions in the example environment 100 are described for exemplary purposes only and do not imply any limitation on the scope of the present disclosure. Embodiments of the present disclosure may also be applied to other environments with different structures and / or functions.

[0029] The following will combine Figures 2 to 7 The process according to the embodiment of the present disclosure is described in detail. For ease of understanding, the specific data mentioned in the following description are exemplary and are not intended to limit the scope of protection of the present disclosure. It is understood that the embodiments described below may also include additional actions not shown and / or the actions shown may be omitted, and the scope of the present disclosure is not limited in this respect.

[0030] Figure 2A flowchart of a method 200 for obtaining indicator data according to an embodiment of the present disclosure is shown. The method can be executed by a computing device, which can be a user terminal, a mobile device, a computer, etc., or a computing system, a single server, a distributed server, or a cloud-based server. At box 202, a second query input can be generated using a language model based on a first query input from a user, and the first query input indicates obtaining indicator data from an indicator platform in a natural language form. For example, refer to Figure 1 The computing device 110 may generate a query input 116 using a language model 114 based on a query input 112 from a user, where the query input 112 indicates obtaining indicator data from an indicator platform in a natural language form.

[0031] At block 204, based on the second query input, a first set of indicators, a first set of dimensions, and a first set of knowledge related to the second query input may be obtained from a preconfigured knowledge base related to the indicator platform. Figure 1 , the computing device 110 may obtain, based on the query input 116, a set of indicators 120, a set of dimensions 122, and a set of knowledge 124 related to the query input 116 from a preconfigured knowledge base 118 related to the indicator platform. At block 206, a second set of indicators, a second set of dimensions, and a second set of knowledge may be determined using a language model based on the second query input, the first set of indicators, the first set of dimensions, and the first set of knowledge. For example, referring to Figure 1 The computing device 110 may determine the indicator set 126 , the dimension set 128 , and the knowledge set 130 using the language model 114 based on the query input 116 , the indicator set 120 , the dimension set 122 , and the knowledge set 124 .

[0032] At block 208, a DSL code for obtaining indicator data may be generated using a language model based on the second query input, the second indicator set, the second dimension set, and the second knowledge set. Figure 1 The computing device 110 may generate a DSL code 132 for obtaining indicator data based on the query input 116, the indicator set 126, the dimension set 128, and the knowledge set 130 using the language model 114. At block 210, the target indicator data may be obtained based on the DSL code using the data acquisition service in the indicator platform. For example, referring to Figure 1 The computing device 110 may obtain the target indicator data 134 based on the DSL code 132 by utilizing the data acquisition service in the indicator platform.

[0033] Therefore, according to the method 200 of the embodiment of the present disclosure, by combining the knowledge base and the language model for multiple rounds of recall, the indicators, dimensions and knowledge related to the user query are accurately obtained, thereby providing reliable and rich context information for the generation of DSL code, ensuring that the generated DSL code is more accurate and can effectively reflect the user's query needs. This method not only improves the efficiency and accuracy of query and data extraction, but also simplifies the user's operation process and significantly improves the user's overall usage experience.

[0034] Figure 3 FIG. 3 is a flow chart of an intelligent data acquisition process 300 according to an embodiment of the present disclosure. Figure 3 As shown, at box 302, obtaining a user query may be performed. The user query may be an original query directly input by the user (i.e., the first query input), and the user may make a business-related data request in natural language. For example, the user query may be "the weekly average of clicks in the last 30 days." The query may also be rewritten and generated based on the context of multiple rounds of conversations. For example, the user entered "yesterday's clicks on the video app" in a previous query. Through multiple rounds of conversations, the user's subsequent query "the weekly average of clicks in the last 30 days" may be rewritten as "the weekly average of clicks on the video app in the last 30 days," and the rewritten query is the second query input.

[0035] In addition, the query input can also be inferred and rewritten by an agent based on the ReAct (Reasoning and Acting) architecture. For example, through reasoning, action, and feedback mechanisms, the agent can automatically analyze the context of the user's query, automatically supplement specific date and other information, and rewrite the original query into a more specific form. For example, the original query can be rewritten as "the weekly average of clicks from xx / xx / xx to yy / yy / yy" to facilitate subsequent queries. The ReAct architecture is a model framework that combines reasoning and action, which can enhance the reasoning ability and decision-making flexibility of language models in complex tasks.

[0036] At box 304, a recall can be performed. For example, based on the user query, the indicators, dimension metadata, and industry knowledge that have been pre-vectorized into the vector database can be preliminarily screened. The vector database can store indicators and dimension information related to the business, and can be updated regularly. When the user query is triggered, a search is first performed in the vector database to find the indicators, dimensions, and business knowledge that are most relevant to the user query. For example, for the query "weekly average of clicks in the last 30 days" entered by the user, the first indicator set to be recalled may include total clicks, valid clicks, and independent user clicks, etc., the first dimension set to be recalled may include time dimension, platform dimension, and regional dimension, and the first knowledge set to be recalled may include the business definition of clicks, daily average calculation method, statistical period method, etc. It should be understood that the above examples are for illustrative purposes only, and the actual recall results may include less or more content.

[0037] At box 306, a secondary recall can be performed. For example, a prompt word (i.e., a first prompt word) can be determined according to the indicator disassembly specification in the indicator platform, and the indicator disassembly specification can be an example of one or more indicator disassembly to guide the language model to recall indicators and dimension information that conform to the user's query semantics. In addition, the user's query content, the indicators and dimension information of the first recall, and the relevant knowledge can be composed of context information (i.e., the first context information). Then, the context information and prompt words can be input into the language model to instruct the language model to obtain content with higher relevance to the user's query from the result of the first recall as the result of the second recall. The first recall stage is to conduct a wide search from the knowledge base to preliminarily screen out indicators, dimensions and industry knowledge related to the user's query. This stage mainly obtains a wide range of content related to the user's query to ensure that the scope of the recall is wide enough to cover various information that may be related to the user's needs. The second recall is based on the result of the first recall for more accurate screening, and the language model is used to extract more accurate indicators and dimensions that are more in line with the user's query semantics from the results of the preliminary screening.

[0038] At block 308, a DSL code may be generated. For example, based on the indicators and dimension information obtained in the secondary recall, the relevant business knowledge, and the user's query content, these contents may be used as context information (i.e., second context information). At the same time, the DSL rules defined in the indicator platform may be combined and these rules may be used as prompt words (i.e., second prompt words). By inputting the context information and prompt words into the language model, the language model may be instructed to generate a DSL code that matches the user's query.

[0039] At box 310, the required indicator data can be obtained through the data retrieval service. For example, the generated DSL query code will be pushed down to the data retrieval service module in the indicator platform to perform specific data retrieval operations. The data retrieval service module performs data query based on the DSL code to ensure that the indicator data related to the user's query can be extracted. Through the execution of the DSL code, the system can accurately follow the user's query logic and automatically obtain relevant data without direct user intervention. In addition, the data retrieval service not only supports data query, but also can perform a variety of calculation functions, such as common statistical operations such as year-on-year and month-on-month. These functions enable the system to handle more complex query requirements and provide users with richer and more comprehensive analysis results. By automatically performing these operations, the system can reduce the user's manual operation requirements and further improve the accuracy and response speed of the query to ensure that the user's diverse business needs are met.

[0040] The data retrieval service uses the model routing module to shield the selection logic of the underlying data table, which means that the specific selection of the underlying data table is transparent to users and language models. The system automatically matches the corresponding data sources and tables according to the query requirements through the model routing module to ensure the accuracy of data extraction and eliminate the errors that may be caused by manual selection of data tables. For example, in response to the user's query "weekly average of clicks in the last 30 days", the data retrieval service will automatically select the relevant time dimension, platform dimension, and click volume table according to the DSL code, and perform the corresponding query operation. In this way, users do not need to care about the underlying data structure, but only need to focus on business logic and query intent, thereby improving data retrieval efficiency and ensuring data accuracy. Finally, the data retrieval service returns the execution results to the system to provide support for subsequent data display and analysis.

[0041] At box 312, the obtained indicator data can be displayed on the terminal. For example, the results returned by the data retrieval service will be automatically presented through the terminal of the system. According to the data type returned, the most suitable chart form is selected for display to ensure that the user can intuitively understand and analyze the data. Common display forms include bar charts, line charts, and pie charts, which can display data in a clear and concise manner. For example, for the user's query "weekly average of clicks in the last 30 days", the system can display the daily click data and present its weekly average in the form of a chart to help users better understand the trend and distribution of the data. In addition, the analysis ideas and processing steps of the language model in the query processing process can also be displayed synchronously. For example, the logical process of data query can be presented in a visual way (such as capsule display) to show how to convert the user's query into a specific analysis result. Users can not only view the final results, but also track the query path and logical steps used to generate these results. This transparent display method not only enhances the user's understanding and trust in the data analysis process, but also provides a way to verify the data processing logic.

[0042] Figure 4 A schematic diagram of the architecture 400 of a self-service analysis tool for an indicator platform according to an embodiment of the present disclosure is shown. The self-service analysis tool can implement various functions through a hybrid architecture of RAG and ReAct, such as operation support query, intelligent data acquisition, query indicator caliber, and attribution analysis. Figure 4 As shown, user query 402 represents a query request input by a user, and the user can raise a query requirement related to the business through natural language. The dialogue processor 404 is used to process the user's multi-round conversation, and can rewrite the user query for multiple rounds of conversations. For example, the multi-round conversation rewriting module can understand the context based on the user's multi-round query and rewrite the query to ensure the accuracy and consistency of the query. The business information loading module can load business information related to the user query 402 (for example, "GMV" means "payment order amount") to enhance the accuracy of the query.

[0043] At the routing module 406, an appropriate processing method can be selected based on the complexity and context of the user's query. For example, when a user asks a relatively simple question, such as querying a single indicator or executing a basic data query request, the RAG module 408 will be given priority for processing. RAG (Retrieval-Augmented Generation) is a model architecture that combines retrieval and generation. It can quickly generate the answers required by users by retrieving relevant information from a pre-embedded knowledge base. For example, "the number of clicks in the last 7 days" is a single-indicator data request and does not involve complex data calculations or multi-dimensional analysis, so it can be quickly responded to through the RAG module 408. The RAG model can efficiently handle such requests, directly retrieve relevant answers from the pre-embedded knowledge base and return results, thereby improving response speed and performance.

[0044] However, for queries involving complex calculations or requiring further processing, it is necessary to select the ReAct module 418 for processing. For example, the query proposed by the user, "weekly average of clicks in the last 30 days", may require secondary processing based on the original data because the underlying indicator platform cannot directly support the calculation of the weekly average. The ReAct module 418 can integrate complex data acquisition logic, perform data clarification, generate calculation logic and perform attribution analysis through the reasoning and operation capabilities of the intelligent agent. It is suitable for complex multi-step queries and scenarios that require dynamic calculations, ensuring that accurate and effective results can still be provided when faced with complex queries. In addition, through this hybrid architecture, a balance can be achieved between performance and processing power. The efficient RAG model is used for simple queries, while the ReAct module provides stronger analysis and calculation support for complex queries.

[0045] RAG module 408 can perform operations based on the model generated by retrieval enhancement, including intent recognition module 410, which is used to recognize the user's query intent. For example, the user's query can be analyzed by intent recognition module 410, and corresponding operations can be performed according to different intents. If it is recognized that the user's intention is to obtain instant help on the operation method during use, the operation support module 412 can be called to provide help information related to the user's operation. If it is recognized that the user's intention is to obtain indicator data related to the business, the intelligent data acquisition module 414 can be called to extract relevant indicator data from the indicator platform. This process can be Figure 3 In addition, if it is recognized that the user intends to query the calculation caliber of a certain indicator, the query caliber module 416 can be called to obtain and return the calculation method and related definition of the indicator to ensure that the user has an accurate understanding of the indicator.

[0046] The ReAct module 418 includes multiple submodules. For example, the input module 420 is used to receive user queries and metadata information. The agent 422 includes a cyclic process of thinking, observing and acting, and can analyze the user's query and make corresponding inferences and feedback. The memory module 424 stores the user's historical queries and context information to help optimize subsequent queries. The sandbox module 426 provides a safe execution environment for programming languages ​​such as Python, which can perform data operations and code execution. The tool module 428 can provide specific functions, including intelligent data acquisition tools and rendering tools, which are helpful for data acquisition, cleaning and processing. The code generation tool is responsible for generating code related to the user query, and the attribution analysis tool is used to analyze the causal relationship of the query data to ensure the accuracy and completeness of the query results. For example, if the user query is "the weekly average of the transaction amount in the last 30 days", and the indicator platform may not support the weekly average operator, the target indicator data (i.e., the indicator required to calculate the weekly average) can be first obtained using the intelligent data acquisition tool, and then the executable code (e.g., Python code) can be generated using the code generation tool, and the additional indicator data required for the user query can be generated by executing the executable code in the sandbox environment.

[0047] In the basic capability module 430, the user's query can be preliminarily processed by the first recall module 430 to filter out the indicators, dimensions and business knowledge related to the query. Then, more accurate processing can be performed by the second recall module 432, and the indicators, dimensions and business knowledge recalled can be ensured to accurately meet the user's needs by further optimizing the results of the first recall. The second recall module 432 further refines and filters the data of the first recall based on the semantics and business requirements of the user's query to ensure that the indicators and dimensions finally recalled can more accurately meet the user's needs. After obtaining accurate indicators, dimensions and business knowledge, these text information can be converted into corresponding domain-specific languages ​​DSL through the Text2DSL module 434. The other tool module 436 includes multiple auxiliary tools to support other data processing functions of the system. For example, the clarification tool is used to further clarify unclear requests when the system parses user queries to ensure the accuracy of query results. The time extraction tool is used to extract information related to time to ensure that queries on the time dimension are correctly processed and responded to.

[0048] The indicator module 438 may include a metadata module 400, which may include indicator metadata and dimension metadata. For example, the indicator metadata may include detailed information related to each indicator, such as the Chinese name, English name, alias, supported dimension list, and whether the indicator has an accumulable attribute. The dimension metadata may include information related to each dimension, such as the time dimension, geographic dimension, device dimension, etc. The indicator data acquisition service module 442 provides the required indicator data to the user through the model routing function, combined with the indicator data acquisition DSL module and the operator set, and executes the corresponding calculation logic. Model routing enables the system to automatically select appropriate data sources and tables, reducing or even eliminating the need for manual user intervention. The indicator data acquisition DSL module is responsible for executing the generated DSL code to obtain indicator data related to the user query. By executing the DSL code, the module can efficiently convert the query logic into a specific data acquisition operation, thereby obtaining accurate business data. The operator set 446 provides a variety of commonly used calculation functions, such as statistical operations, aggregation calculations, etc., to support more complex query requirements and ensure that users can obtain complete and accurate analysis results from queries.

[0049] Figure 5A schematic diagram of a process 500 for supplementary recall of high-cardinality dimensions according to an embodiment of the present disclosure is shown. The main purpose of supplementary recall of high-cardinality dimensions is to cope with complex query scenarios, especially data requests involving high-cardinality dimensions such as goods and stores. Due to the wide variety of high-cardinality dimension data, such as product ID or store ID, the number of dimension values ​​after deduplication is huge, and it is impossible to completely retrieve all relevant dimensions and their values ​​in one recall phase. In addition, high-cardinality dimension data often belongs to the ID type, and the recall effect through the vectorized model is not good. Therefore, in order to ensure that data related to high-cardinality dimensions can be accurately obtained in queries, this embodiment proposes an auxiliary recall solution for high-cardinality dimensions.

[0050] like Figure 5 As shown, the high cardinality dimension supplement recall module 504 may include multiple functional modules for processing high cardinality dimension information. First, the query entity extraction module 506 automatically identifies specific high cardinality dimension entities, such as product names, store names, etc., from the user query 502. These entities are usually key information that is highly relevant to the user query, but due to their large number and dynamic changes, it is often difficult to process them completely in one recall phase. Then, the high cardinality dimension recall module 510 retrieves the high cardinality dimension information related to these entities by accessing the high cardinality dimension feature library 508. The high cardinality dimension feature library 508 stores a large number of pre-processed dimensional features, which can support the system to quickly and accurately obtain high cardinality dimension information that matches the user query requirements. In addition, in order to ensure that the system can provide more accurate dimension values, the dimension value extraction module 514 can extract specific dimension values ​​from the high-frequency dimension value library 512. These high-frequency dimension values ​​are extracted based on historical queries and common business scenarios, which can accelerate processing and improve the accuracy of query results.

[0051] Based on the user query 502, a recall 516 can be performed first. During this process, the system will preliminarily screen out the basic indicators and dimensional information related to the query. Subsequently, based on the results of the first recall, a secondary recall 518 can be further performed to improve the accuracy of the recall. In addition, the high-cardinality dimensional information from the high-cardinality dimensional supplementary recall module 504 will be integrated with the recall results in the secondary recall stage to ensure that the final indicators and dimensions are more comprehensive and more in line with the needs of the user's query. Subsequently, the system will generate corresponding DSL codes for the above query results and dimensional information through the Text2DSL module 520. The DSL code can abstract the user's query logic and facilitate the subsequent execution of complex data extraction operations. Finally, at box 522, the specific indicator data corresponding to the user's query requirements can be obtained through the data retrieval service according to the generated DSL code.

[0052] Figure 6 FIG. 6 is a block diagram of an apparatus 600 for acquiring indicator data according to an embodiment of the present disclosure. Figure 6As shown, the apparatus 600 includes a query input generating module 602, which is configured to generate a second query input using a language model based on a first query input from a user, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form. The apparatus 600 also includes a first information obtaining module 604, which is configured to obtain a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a pre-configured knowledge base related to the indicator platform based on the second query input.

[0053] The device 600 also includes a second information determination module 606, which is configured to determine the second indicator set, the second dimension set and the second knowledge set based on the second query input, the first indicator set, the first dimension set and the first knowledge set using the language model. The device 600 also includes a query code generation module 608, which is configured to generate a domain specific language DSL code for obtaining indicator data based on the second query input, the second indicator set, the second dimension set and the second knowledge set using the language model. In addition, the device 600 also includes a target indicator acquisition module 610, which is configured to obtain target indicator data based on the DSL code using the data acquisition service in the indicator platform.

[0054] Figure 7 A block diagram of an electronic device 700 is shown in accordance with certain embodiments of the present disclosure. Figure 7 FIG. 1 is a block diagram of an electronic device 700 according to some embodiments of the present disclosure. The device 700 may be a device or apparatus described in an embodiment of the present disclosure. Figure 7 As shown, the device 700 includes a central processing unit (CPU) and / or a graphics processing unit (GPU) 701, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 702 or loaded from a storage unit 708 to a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The CPU / GPU 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704. Although not shown in FIG. Figure 7 As shown in FIG. 7 , device 700 may further include a coprocessor.

[0055] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0056] The various methods or processes described above may be performed by the CPU / GPU 701. For example, in some embodiments, the methods may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the CPU / GPU 701, one or more steps or actions in the methods or processes described above may be performed.

[0057] In some embodiments, the methods and processes described above may be implemented as a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present disclosure.

[0058] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0059] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0060] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, and conventional procedural programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as a separate software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0061] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0062] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0063] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the equipment, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each frame in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the frame can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous frames can actually be executed substantially in parallel, and they can also be executed in the opposite order sometimes, depending on the functions involved. It should also be noted that each frame in the block diagram and / or flow chart, and the combination of frames in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0064] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

[0065] Some example implementations of the present disclosure are listed below.

[0066] Example 1. A method for obtaining indicator data, comprising:

[0067] Based on a first query input from a user, generating a second query input using a language model, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form;

[0068] Based on the second query input, obtaining a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a preconfigured knowledge base related to the indicator platform;

[0069] Determine a second indicator set, a second dimension set, and a second knowledge set using the language model based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0070] Based on the second query input, the second indicator set, the second dimension set, and the second knowledge set, using the language model to generate a domain specific language DSL code for obtaining indicator data; and

[0071] Based on the DSL code, the target indicator data is obtained by utilizing the data acquisition service in the indicator platform.

[0072] Example 2. The method according to Example 1, wherein generating a second query input using the language model comprises:

[0073] Obtaining previous query input related to the first query input;

[0074] Based on the previous query input, the second query input is generated by rewriting the first query input using the language model.

[0075] Example 3. The method according to any one of Examples 1-2, further comprising:

[0076] Based on the second query input, generating executable code using the language model; and

[0077] Additional indicator data related to the second query input is generated by processing the target indicator data using the code.

[0078] Example 4. The method according to any one of Examples 1-3, wherein obtaining the first indicator set, the first dimension set, and the first knowledge set related to the second query input comprises:

[0079] generating a query vector corresponding to the second query input using a vectorization model; and

[0080] Based on the query vector, the first indicator set, the first dimension set, and the first knowledge set are determined from the knowledge base through vector similarity matching.

[0081] Example 5. The method according to any one of Examples 1-4, wherein determining the second indicator set, the second dimension set, and the second knowledge set using the language model comprises:

[0082] Determine first context information based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0083] Determining a first prompt word based on the indicator disassembly specification in the indicator platform; and

[0084] Based on the context information and the first prompt word, the second indicator set, the second dimension set and the second knowledge set are determined using the language model.

[0085] Example 6. According to any one of Examples 1-5, the method of generating the DSL code for obtaining indicator data using the language model comprises:

[0086] Determine second context information based on the second query input, the second indicator set, the second dimension set, and the second knowledge set;

[0087] Based on the DSL specification in the indicator platform, determining a second prompt word for guiding the language model to generate the DSL; and

[0088] The DSL code is generated using the language model based on the second context information and the second prompt word.

[0089] Example 7. The method according to any one of Examples 1-6, further comprising:

[0090] In response to determining that the second query input has a high cardinality, determining a high cardinality value from a high cardinality library; and

[0091] The high cardinality dimension value is added to the second context information.

[0092] Example 8. According to any one of the methods of Examples 1-7, obtaining the target indicator data by using the data acquisition service in the indicator platform includes:

[0093] Based on the DSL code, using the model routing module in the data acquisition service to determine a data source related to the target indicator data; and

[0094] The target indicator data is obtained from the data source by executing the DSL code.

[0095] Example 9. A device for obtaining indicator data, comprising:

[0096] A query input generation module is configured to generate a second query input using a language model based on a first query input from a user, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form;

[0097] A first information acquisition module is configured to acquire, based on the second query input, a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a preconfigured knowledge base related to the indicator platform;

[0098] A second information determination module is configured to determine a second indicator set, a second dimension set, and a second knowledge set using the language model based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0099] a query code generation module, configured to generate a domain specific language (DSL) code for acquiring indicator data using the language model based on the second query input, the second indicator set, the second dimension set, and the second knowledge set; and

[0100] The target indicator acquisition module is configured to acquire target indicator data based on the DSL code and using the data acquisition service in the indicator platform.

[0101] Example 10. The apparatus according to Example 9, wherein the query input generation module comprises:

[0102] a previous query acquisition module, configured to obtain previous query inputs related to the first query input;

[0103] The query input second generation module is configured to generate the second query input by rewriting the first query input based on the previous query input by using the language model.

[0104] Example 11. The apparatus according to any one of Examples 9-10, further comprising:

[0105] an executable code generation module, configured to generate executable code based on the second query input using the language model; and

[0106] The additional indicator generation module is configured to generate additional indicator data related to the second query input by processing the target indicator data using the code.

[0107] Example 12. The apparatus according to any one of Examples 9-11, wherein the first information acquisition module comprises:

[0108] a query vector generating module, configured to generate a query vector corresponding to the second query input by using a vectorization model; and

[0109] The first information second acquisition module is configured to determine the first indicator set, the first dimension set and the first knowledge set from the knowledge base through vector similarity matching based on the query vector.

[0110] Example 13. The apparatus of any one of Examples 9-12, wherein the second information determination module comprises:

[0111] a first context determination module, configured to determine first context information based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0112] A first prompt word determination module is configured to determine a first prompt word based on an indicator disassembly specification in the indicator platform; and

[0113] The second information second determination module is configured to determine the second indicator set, the second dimension set and the second knowledge set based on the context information and the first prompt word using the language model.

[0114] Example 14. The apparatus according to any one of Examples 9-13, wherein the query code generation module comprises:

[0115] a second context determination module, configured to determine second context information based on the second query input, the second indicator set, the second dimension set, and the second knowledge set;

[0116] A second prompt word determination module is configured to determine a second prompt word for guiding the language model to generate a DSL based on the DSL specification in the indicator platform; and

[0117] The query code second generation module is configured to generate the DSL code by using the language model based on the second context information and the second prompt word.

[0118] Example 15. The apparatus of any one of Examples 9-14, further comprising:

[0119] a high cardinality value determination module, configured to determine a high cardinality value from a high cardinality library in response to determining that the second query input has a high cardinality; and

[0120] The high cardinality dimension value adding module is configured to add the high cardinality dimension value to the second context information.

[0121] Example 16. According to the device of any one of Examples 9-15, the target indicator data acquisition module includes:

[0122] A data source determination module is configured to determine a data source related to the target indicator data based on the DSL code and using a model routing module in the data acquisition service; and

[0123] The target indicator data second acquisition module is configured to acquire the target indicator data from the data source by executing the DSL code.

[0124] Example 17. An electronic device comprising:

[0125] Processor; and

[0126] A memory coupled to the processor, the memory having instructions stored therein, wherein when the instructions are executed by the processor, the electronic device performs actions, the actions comprising:

[0127] Based on a first query input from a user, generating a second query input using a language model, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form;

[0128] Based on the second query input, obtaining a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a preconfigured knowledge base related to the indicator platform;

[0129] Determine a second indicator set, a second dimension set, and a second knowledge set using the language model based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0130] Based on the second query input, the second indicator set, the second dimension set, and the second knowledge set, using the language model to generate a domain specific language DSL code for obtaining indicator data; and

[0131] Based on the DSL code, the target indicator data is obtained by utilizing the data acquisition service in the indicator platform.

[0132] Example 18. The electronic device of Example 17, wherein generating a second query input using the language model comprises:

[0133] Obtaining previous query input related to the first query input;

[0134] Based on the previous query input, the second query input is generated by rewriting the first query input using the language model.

[0135] Example 19. The electronic device according to any one of Examples 17-18, wherein the actions further include:

[0136] Based on the second query input, generating executable code using the language model; and

[0137] Additional indicator data related to the second query input is generated by processing the target indicator data using the code.

[0138] Example 20. The electronic device according to any one of Examples 17-19, wherein obtaining the first indicator set, the first dimension set, and the first knowledge set related to the second query input comprises:

[0139] generating a query vector corresponding to the second query input using a vectorization model; and

[0140] Based on the query vector, the first indicator set, the first dimension set, and the first knowledge set are determined from the knowledge base through vector similarity matching.

[0141] Example 21. The electronic device according to any one of Examples 17-20, wherein determining the second indicator set, the second dimension set, and the second knowledge set using the language model comprises:

[0142] Determine first context information based on the second query input, the first indicator set, the first dimension set, and the first knowledge set;

[0143] Determining a first prompt word based on the indicator disassembly specification in the indicator platform; and

[0144] Based on the context information and the first prompt word, the second indicator set, the second dimension set and the second knowledge set are determined using the language model.

[0145] Example 22. The electronic device according to any one of Examples 17-21, wherein generating the DSL code for obtaining indicator data using the language model comprises:

[0146] Determine second context information based on the second query input, the second indicator set, the second dimension set, and the second knowledge set;

[0147] Based on the DSL specification in the indicator platform, determining a second prompt word for guiding the language model to generate the DSL; and

[0148] The DSL code is generated using the language model based on the second context information and the second prompt word.

[0149] Example 23. The electronic device according to any one of Examples 17-22, wherein the actions further include:

[0150] In response to determining that the second query input has a high cardinality, determining a high cardinality value from a high cardinality library; and

[0151] The high cardinality dimension value is added to the second context information.

[0152] Example 24. The electronic device according to any one of Examples 17-23, using the data acquisition service in the indicator platform to obtain the target indicator data includes:

[0153] Based on the DSL code, using the model routing module in the data acquisition service to determine a data source related to the target indicator data; and

[0154] The target indicator data is obtained from the data source by executing the DSL code.

[0155] Example 25. A computer-readable storage medium having one or more computer instructions stored thereon, wherein the one or more computer instructions are executed by a processor to implement the method according to any one of Examples 1 to 8.

[0156] Example 26. A computer program product tangibly stored on a computer-readable medium and comprising computer-executable instructions that, when executed by a device, cause the device to perform the method according to any one of Examples 1 to 8.

[0157] Although the disclosure has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Instead, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A method for obtaining indicator data, comprising: Based on a first query input from a user, generating a second query input using a language model, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form; Based on the second query input, obtaining a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a preconfigured knowledge base related to the indicator platform; Based on the second query input, the first indicator set, the first dimension set, and the first knowledge set, using the language model to determine a second indicator set, a second dimension set, and a second knowledge set, the determining comprising: Determine first context information based on the second query input, the first indicator set, the first dimension set, and the first knowledge set; Determining a first prompt word based on the indicator disassembly specification in the indicator platform; and Based on the context information and the first prompt word, using the language model to determine the second indicator set, the second dimension set, and the second knowledge set; Based on the second query input, the second indicator set, the second dimension set, and the second knowledge set, using the language model to generate a domain specific language DSL code for obtaining indicator data; and Based on the DSL code, the target indicator data is obtained by utilizing the data acquisition service in the indicator platform.

2. The method according to claim 1, wherein generating the second query input using the language model comprises: Obtaining previous query input related to the first query input; Based on the previous query input, the second query input is generated by rewriting the first query input using the language model.

3. The method according to claim 2, further comprising: Based on the second query input, generating executable code using the language model; as well as Additional indicator data related to the second query input is generated by processing the target indicator data using the code.

4. The method according to claim 1, wherein obtaining the first indicator set, the first dimension set, and the first knowledge set related to the second query input comprises: generating a query vector corresponding to the second query input using a vectorized model; as well as Based on the query vector, the first indicator set, the first dimension set, and the first knowledge set are determined from the knowledge base through vector similarity matching.

5. The method according to claim 1, wherein generating the DSL code for obtaining indicator data using the language model comprises: Determine second context information based on the second query input, the second indicator set, the second dimension set, and the second knowledge set; Based on the DSL specification in the indicator platform, determining a second prompt word for guiding the language model to generate the DSL; as well as The DSL code is generated using the language model based on the second context information and the second prompt word.

6. The method according to claim 5, further comprising: In response to determining that the second query input has a high cardinality, determining a high cardinality value from a high cardinality library; as well as The high cardinality dimension value is added to the second context information.

7. According to the method of claim 1, obtaining the target indicator data by using the data acquisition service in the indicator platform comprises: Based on the DSL code, determine the data source related to the target indicator data using the model routing module in the data acquisition service; as well as The target indicator data is obtained from the data source by executing the DSL code.

8. A device for obtaining indicator data, comprising: A query input generation module is configured to generate a second query input using a language model based on a first query input from a user, wherein the first query input indicates obtaining indicator data from an indicator platform in a natural language form; A first information acquisition module is configured to acquire, based on the second query input, a first indicator set, a first dimension set, and a first knowledge set related to the second query input from a preconfigured knowledge base related to the indicator platform; The second information determination module is configured to determine a second indicator set, a second dimension set, and a second knowledge set using the language model based on the second query input, the first indicator set, the first dimension set, and the first knowledge set, the second information determination module comprising: a first context determination module, configured to determine first context information based on the second query input, the first indicator set, the first dimension set, and the first knowledge set; A first prompt word determination module is configured to determine a first prompt word based on an indicator disassembly specification in the indicator platform; and A second information set determination module, configured to determine the second indicator set, the second dimension set and the second knowledge set by using the language model based on the context information and the first prompt word; a query code generation module, configured to generate a domain specific language (DSL) code for acquiring indicator data using the language model based on the second query input, the second indicator set, the second dimension set, and the second knowledge set; and The target indicator acquisition module is configured to acquire target indicator data based on the DSL code and using the data acquisition service in the indicator platform.

9. An electronic device, comprising: processor; as well as A memory coupled to the processor, the memory having instructions stored therein, wherein when the instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 7.

10. A computer program product tangibly stored on a non-transitory computer readable medium and comprising computer executable instructions for performing the method according to any one of claims 1 to 7.

Images