A data processing method, device, computer equipment, medium and program product

Abstract

The embodiment of the application discloses a data processing method and device, computer equipment, medium and program product, after setting a timing task for a target application, object data can be obtained from the target application based on the timing task, and automatic data acquisition is realized. By setting multiple prompt information, different prompt information is used to instruct the large language model to execute different data processing procedures, that is, based on the guidance of multiple prompt information, the large language model can execute different data processing procedures on the object data, thereby obtaining multiple data processing results, not only realizing automatic analysis of data, but also being able to execute multiple data processing procedures through multiple prompt information, which is convenient and fast, and has high recognition accuracy. Store the data processing results corresponding to the multiple prompt information respectively, and when a query request for the target application is obtained, find the data processing result corresponding to the query request from the multiple data processing results, and realize data storage and query.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a data processing method, apparatus, computer equipment, medium, and program product. Background Technology

[0002] With the development of computer technology, users can publish articles, comment on articles, like, and forward them on platforms such as communities, thereby generating various types of object data.

[0003] In related technologies, it is generally necessary to manually acquire object data and perform various data processing procedures on the object data to obtain data processing results such as user comments and ratings on the target application.

[0004] However, this method, which relies on manual analysis of user feedback on the target application, has relatively low data processing efficiency. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a data processing method, apparatus, computer equipment, medium, and program product for improving data processing efficiency.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide a data processing method, the method comprising:

[0008] Obtain multiple prompt messages and scheduled tasks for the target application, where different prompt messages are used to instruct the execution of different data processing flows;

[0009] According to the scheduled task, object data is obtained from the target application, and the object data is used to characterize object-related behavioral data generated based on the target application.

[0010] Based on the multiple prompts, the large language model is instructed to perform different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts.

[0011] Store multiple data processing results;

[0012] If a query request for the target application is obtained, the data processing result corresponding to the query request is searched from the multiple stored data processing results.

[0013] On the other hand, embodiments of this application provide a data processing apparatus, the apparatus comprising: an acquisition unit, an analysis unit, a storage unit, and a query unit;

[0014] The acquisition unit is used to acquire multiple prompt messages and timed tasks for the target application, and different prompt messages are used to instruct the execution of different data processing flows;

[0015] The acquisition unit is used to acquire object data from the target application according to the scheduled task, and the object data is used to characterize object-related behavioral data generated based on the target application.

[0016] The analysis unit is used to instruct the large language model to perform different data processing procedures on the object data based on the multiple prompts, so as to obtain the data processing results corresponding to the multiple prompts respectively.

[0017] The storage unit is used to store multiple data processing results;

[0018] The query unit is configured to, if a query request for the target application is obtained, search for the data processing result corresponding to the query request from a plurality of stored data processing results.

[0019] On the other hand, embodiments of this application provide a computer device, the computer device including a processor and a memory:

[0020] The memory is used to store computer programs and to transfer the computer programs to the processor;

[0021] The processor is configured to execute the methods described above according to instructions in the computer program.

[0022] On the other hand, embodiments of this application provide a computer-readable storage medium for storing a computer program for performing the methods described above.

[0023] On the other hand, embodiments of this application provide a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform the methods described above.

[0024] As can be seen from the above technical solution, by setting up scheduled tasks for the target application, object data can be retrieved from the target application at regular intervals based on the scheduled tasks, thus achieving automatic data acquisition. By setting multiple prompt messages, different prompt messages are used to instruct the large language model to execute different data processing flows. That is, based on the guidance of multiple prompt messages, the large language model can execute different data processing flows on the object data respectively, thereby obtaining multiple data processing results. This not only achieves automatic data analysis, but also enables the execution of multiple data processing flows through various prompt messages, which is convenient, fast, and has high recognition accuracy. The data processing results corresponding to the multiple prompt messages are stored. When a query request for the target application is obtained, the data processing result corresponding to the query request is found from the multiple data processing results, realizing data storage and retrieval. Therefore, through scheduled tasks, multiple prompt messages, a large language model, and storage, automatic data acquisition, analysis, storage, and retrieval are achieved. The automated process reduces manual intervention, quickly processes and analyzes large amounts of data, and improves data processing efficiency. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic diagram of a computer system for a data processing method provided in an embodiment of this application;

[0027] Figure 2 This is a schematic diagram illustrating an application scenario of a data processing method provided in an embodiment of this application;

[0028] Figure 3 A flowchart illustrating a data processing method provided in an embodiment of this application;

[0029] Figure 4A A flowchart for obtaining object data is provided as an embodiment of this application;

[0030] Figure 4B A flowchart for obtaining object data is provided as an embodiment of this application;

[0031] Figure 5 A schematic diagram of a data processing system provided in an embodiment of this application;

[0032] Figure 6 A schematic diagram illustrating a target application's overall market analysis as provided in an embodiment of this application;

[0033] Figure 7 A schematic diagram illustrating the details of an indicator provided in an embodiment of this application;

[0034] Figure 8 A schematic diagram illustrating one advantage and disadvantage of an embodiment of this application;

[0035] Figure 9 A schematic diagram of a subject matter provided for an embodiment of this application;

[0036] Figure 10 A schematic diagram of heat provided for an embodiment of this application;

[0037] Figure 11 A schematic diagram illustrating a thematic activity analysis provided in an embodiment of this application;

[0038] Figure 12 A schematic diagram illustrating the details of a themed activity as provided in an embodiment of this application;

[0039] Figure 13 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application;

[0040] Figure 14 This application provides a schematic diagram of the structure of a server according to an embodiment of the present application.

[0041] Figure 15 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0043] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0044] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0045] It should be understood that although the terms first, second, etc., may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, a first parameter may also be referred to as a second parameter, and similarly, a second parameter may also be referred to as a first parameter. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0046] It should be noted that this application may display prompt interfaces, pop-ups, or output voice prompts before and during the collection of user-related data (such as object data). These prompt interfaces, pop-ups, or voice prompts are used to inform the user that their relevant data is being collected. This ensures that the application only begins the steps for collecting user-related data after receiving confirmation from the user regarding the prompt interface or pop-up; otherwise (i.e., without receiving confirmation from the user), the steps for collecting user-related data end, meaning no user-related data is collected. In other words, all user data obtained by this application is acquired with the user's consent and authorization, and the collection, use, and processing of relevant user data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0047] To improve data processing efficiency, this application provides a data processing method that achieves automatic data acquisition, analysis, storage, and querying through scheduled tasks, multiple prompts, a large language model, and storage. The automated process reduces manual intervention, quickly processes and analyzes large amounts of data, and improves data processing efficiency.

[0048] To facilitate understanding of the data processing method provided in the embodiments of this application, the computer system of the data processing method will be described first below.

[0049] See Figure 1 This figure is a schematic diagram of a computer system for a data processing method provided in an embodiment of this application. The computer system 100 includes one or more terminal devices 110 and one or more servers 120, and the terminal devices 110 and servers 120 can communicate through a communication network, etc.

[0050] The communication network uses standard communication technologies and / or protocols, typically the Internet, but can also be any network, including but not limited to Bluetooth, local area network (LAN), metropolitan area network (MAN), wide area network (WAN), mobile, private network, or any combination of virtual private network. In some embodiments, custom or dedicated data communication technologies may be used to replace or supplement the aforementioned data communication technologies.

[0051] The terminal device can be a device such as a smartphone, in-vehicle terminal, wearable device, personal computer (PC), intelligent voice interaction device, smart home appliance, aircraft, or unmanned vending terminal. A client application for the first application can be installed and run on the terminal device. This first application can be an application that supports data processing, or other applications that support displaying data processing results, data analysis, and data storage; this application does not limit its scope. Furthermore, this application does not limit the form of the first application, including but not limited to applications (Apps), mini-programs, etc., installed on the terminal device, and can also be in the form of a webpage.

[0052] A server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services such as cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and basic cloud computing services such as big data. The server can be a backend server for the aforementioned target application, providing backend services to the client of the target application.

[0053] To facilitate understanding of the data processing method provided in the embodiments of this application, the following example uses a server as the execution subject of the data processing method to illustrate the application scenarios of the data processing method.

[0054] See Figure 2 This figure is a schematic diagram illustrating an application scenario of a data processing method provided in an embodiment of this application. Figure 2 In this application scenario, we take two terminal devices and one server as an example, where terminal device 111 and terminal device 112 are... Figure 1 The terminal device 110.

[0055] The terminal device 111 has a target application related to the game installed. Object A can use the target application through the terminal device 111 to generate object data such as publishing game-related articles, liking articles, and commenting on articles.

[0056] Terminal device 112 is equipped with a first application for performing data processing. Object B can use the first application through terminal device 112 to automatically process object data. For example, object B can set up scheduled tasks and multiple notification messages for the target application through the first application. Terminal device 112 sends the multiple notification messages and scheduled tasks to server 120.

[0057] Server 120 receives multiple prompts and scheduled tasks from terminal device 112. When a scheduled task is triggered, it retrieves object data from the target application, such as retrieving object data daily at 5:00 AM, thus achieving automatic data acquisition based on the scheduled task. Different prompts instruct the large language model to execute different data processing flows, such as extracting themes and sentiment types from the object data. Specifically, server 120 can instruct the large language model to execute corresponding data processing flows based on the prompts, thereby obtaining and storing multiple data processing results corresponding to each prompt. For example, the data processing results could include the number of articles published, the total number of likes, the total number of comments, and the distribution of themes for each topic. Through multiple prompts and the large language model, not only is automatic data analysis achieved, but multiple data processing flows can also be executed through various prompts, which is convenient, fast, and has high recognition accuracy.

[0058] Object B can be queried through the first application in terminal device 112. For example, if the content is related to a certain topic, object B can enter that topic in the first application to generate a query request. Terminal device 112 sends the query request to server 120, and server 120 searches for the data processing result corresponding to the query request from multiple stored data processing results, thus realizing data storage and retrieval.

[0059] Therefore, by using scheduled tasks, multiple prompts, a large language model, and storage methods, the system achieves automatic data acquisition, analysis, storage, and retrieval. The automated process reduces manual intervention, quickly processes and analyzes large amounts of data, and improves data processing efficiency.

[0060] The data processing method provided in this application embodiment can be executed by a server. However, in other embodiments of this application, the terminal device may also have similar functions to the server to execute the data processing method provided in this application embodiment, or the terminal device and the server may jointly execute the data processing method provided in this application embodiment. This embodiment does not limit this.

[0061] The following describes in detail a data processing method provided in this application through method embodiments.

[0062] See Figure 3 This figure is a schematic flowchart of a data processing method provided in an embodiment of this application. For ease of description, the following embodiment uses a server as the execution subject of the data processing method as an example. Figure 3 As shown, the data processing method includes S301-S305.

[0063] S301: Retrieves multiple prompt messages and scheduled tasks for the target application.

[0064] A prompt is a piece of information that uses natural language to guide or motivate a large language model to complete a specific task, such as the data processing task described in this application. Further, the prompt can be a text input to the large language model, used to guide it in generating a specific type of response or output. The purpose of the prompt is to provide contextual information to the large language model so that the model understands the user's needs and provides appropriate answers or suggestions. The prompt can be a question, a topic, a task description, etc., and the large language model will generate relevant outputs based on this input.

[0065] Different prompts are used to instruct the large language model to perform different data processing flows. A data processing flow refers to the process of processing data. The basic purpose of data processing is to extract and deduce specific, valuable, and meaningful data from a large amount of data that may be messy and difficult to understand. For example, data processing can be used to derive the central idea of ​​an article.

[0066] A scheduled task for a target application refers to a task that retrieves corresponding data from the target application when preset time conditions are met; that is, a task that retrieves data at regular intervals. This application does not specifically limit the trigger time of the scheduled task; those skilled in the art can set it according to actual needs. For example, retrieving object data from the target application at 5:00 AM every day.

[0067] S302: Obtain object data from the target application according to the scheduled task.

[0068] A scheduled task is a task that retrieves object data based on preset conditions. This application embodiment does not specifically limit the preset conditions, such as meeting a preset time (e.g., 5:00 AM every day) or a preset change (e.g., detecting data updates based on tracking points). When the scheduled task is triggered, object data is retrieved from the target application.

[0069] Object data is used to characterize object-related behavioral data generated based on a target application. Taking a game-related application as an example, this could include articles written by objects for the target application, object behaviors performed on the target application (such as clicks, browsing time for a particular article, type of terminal device used, etc.), comments, and themed events published for the target application.

[0070] Therefore, scheduled tasks can be used to automatically acquire object data from target applications, i.e., automate data acquisition. Data acquisition refers to the process of obtaining raw data from different sources using various methods and tools. This data may come from multiple channels, such as sensors, databases, the internet, experiments, questionnaires, social media, applications, etc. The main purpose of data acquisition is to provide foundational data for subsequent data analysis, research, decision-making, or machine learning models.

[0071] S303: Based on multiple prompts, the large language model is instructed to perform different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts.

[0072] For a detailed explanation of the prompts, please refer to the above text; they will not be repeated here. Different prompts are used to instruct the large language model to perform different data processing flows, thereby enabling various data analyses of the target data. Data analysis is an important process that involves collecting, cleaning, processing, analyzing, and interpreting data to obtain valuable information that supports the decision-making process. Data analysis involves using statistics, mathematical models, algorithms, and data visualization tools to transform raw data into meaningful insights to help solve specific problems, predict future trends, or optimize operations.

[0073] Based on this, a single prompt can be used to instruct the large language model to perform a data processing procedure on the object data. In turn, multiple prompts can be used to instruct the large language model to perform various data processing procedures on the object data, thereby obtaining the data processing results corresponding to each of the multiple prompts.

[0074] As one possible implementation, the multiple prompts can be preset or selected by the user based on the target application; this application does not impose any specific limitations on this.

[0075] This application does not specifically limit the content of the multiple prompt messages, nor how the prompt messages guide the large language model to implement the corresponding data processing flow. The following will explain based on multiple prompt messages, and will not be repeated here.

[0076] Large Language Models (LLMs) are artificial intelligence models that learn to understand and generate human language by learning from large amounts of text data. These models are typically based on a Transformer architecture, learning basic patterns and structures of language through pre-training on large amounts of text data, thereby understanding the syntax, semantics, and contextual relationships of language. After pre-training, large language models can be fine-tuned to adapt to specific tasks or applications, such as text generation, text classification, and sentiment analysis. In a specific embodiment, a large language model can be a language model based on VisualGLM. VisualGLM-6B is an open-source, multimodal dialogue language model supporting images, Chinese, and English. The image part bridges the visual and language models by training BLIP2-Qformer. BLIP-2 (Bootstrapping Language-Image Pre-training) introduces a new visual language pre-training paradigm to arbitrarily combine and fully utilize two pre-trained visual encoders and an LLM, without requiring end-to-end pre-training of the entire architecture. This allows us to achieve state-of-the-art results on multiple visual language tasks while significantly reducing the number of training parameters and pre-training costs.

[0077] Therefore, by combining multiple prompts and a large language model, not only is automatic data analysis achieved, but also multiple data processing flows can be executed through various prompts, which is convenient, fast, and has high recognition accuracy.

[0078] As one possible approach, multiple data processing results can be obtained from frequency statistics, trend analysis, and correlation analysis. Frequency statistics can calculate the frequency of each topic's occurrence, identifying the most common topics. Trend analysis can analyze the changing trends of topics over different time periods, identifying emerging topics or areas of interest. Correlation analysis can assess the correlation between different topics, discovering potential connections or patterns.

[0079] As one possible implementation, since the data processing flow for each prompt message is different, a large language model corresponding to each prompt message can be provided, that is, multiple large language models can be provided. This allows data processing of object data based on a prompt message and a large language model, which not only makes the data processing flow more targeted and improves the accuracy of the data processing results, but also enables the parallel execution of multiple data processing flows, thereby improving data processing efficiency.

[0080] One possible implementation is to determine the large language model based on the type of object data. For example, if the object data is of the stock type, the first large language model is instructed to perform different data processing procedures on the object data based on multiple prompts, resulting in multiple data processing results; if the object data is of the incremental type, the first large language model is instructed to perform different data processing procedures on the object data based on multiple prompts, resulting in multiple data processing results.

[0081] Among these, object data belonging to the storage type can be called existing data, that is, data that already exists or historical data. Object data belonging to the incremental type can be called incremental data, that is, newly added data. The processing speed of the second largest language model is greater than that of the first largest language model. For example, the first largest language model could be the Vicuna model (an open-source chatbot), and the second largest language model could be the Sagemaker model (a model that includes various machine learning algorithms).

[0082] Therefore, incremental data generally requires real-time processing, meaning it has high timeliness requirements, while existing data can be processed offline, meaning it has lower timeliness requirements. This allows for processing of incremental data using a faster second-largest language model, and existing data using a slower first-largest language model. The cost of using the second-largest language model is generally higher than that of the first. This approach reduces costs while ensuring timely processing.

[0083] S304: Stores multiple data processing results.

[0084] After obtaining multiple data processing results, the data processing results are stored so that subsequent query functions can be implemented based on the stored data processing results.

[0085] This application does not specifically limit the specific storage method, such as storing the data processing results to a database file on a disk, magnetic tape, disk array, or cloud storage. The following description uses one such method as an example.

[0086] Specifically, the results of multiple data processing are divided into a first result set and a second result set. The first result set is stored in a first storage space, and the second result set is stored in a second storage space.

[0087] In this embodiment, the query frequency of the data processing results included in the first result set is less than or equal to a frequency threshold, while the query frequency of the data processing results included in the second result set is greater than the frequency threshold. This application does not specifically limit the frequency threshold; those skilled in the art can set it according to actual needs, for example, determining the frequency threshold based on the type of data processing result.

[0088] The cost of using the first storage space is lower than the cost of using the second storage space. Usage cost refers to the expense required to store and process data results in the corresponding storage space. Generally, storage spaces with higher usage costs offer faster data interaction speeds, such as shorter query wait times. For example, the second storage space could be Redis (RemoteDictionary Server, an open-source in-memory database) or MySQL (a relational database management system), while the first storage space could be a public cloud storage service, such as Simple Storage Service (S3).

[0089] Therefore, by classifying and storing data processing results in different storage spaces—that is, storing frequently queried data processing results in higher-cost storage spaces and less frequently queried data processing results in lower-cost storage spaces—it is possible to reduce storage costs while improving subsequent data query speed, thereby enhancing the user experience.

[0090] S305: If a query request for a target application is obtained, the corresponding data processing result for the query request is searched from the multiple stored data processing results.

[0091] A query request is generally intended to obtain a certain data processing result. Therefore, after obtaining the query request, it can be searched from multiple stored data processing results to obtain the data processing result corresponding to the query request.

[0092] As one possible implementation, the data processing results corresponding to the query request can be displayed through charts. This will be discussed later. Figures 6-12 The details will not be elaborated here. Users can search for reviews and ratings of the target application by their peers using keywords, themed activities, etc., to provide feedback on key metrics.

[0093] As one possible implementation, query requests can include topics, keywords, or themed events, and can even be combined with dimensions such as dates. A topic refers to content extracted from relevant information (such as articles) that expresses the meaning of the target application. Keywords can be words from the article, such as high-frequency words. Themed events are activities published by the target application, and queries can be based on links corresponding to these events.

[0094] Based on this, if the query request for the target application includes target keywords, then the data processing result corresponding to the target keywords is searched from multiple data processing results; if the query request for the target application includes target thematic activities, then the data processing result corresponding to the target thematic activities is searched from multiple data processing results. Here, the target keyword is one of multiple keywords related to the target application, and the target thematic activity is one of multiple thematic activities related to the target application.

[0095] Therefore, query requests can be generated based on keywords or thematic activities (such as links), enabling more targeted queries for specific content. Furthermore, the data processing flow generates corresponding data processing results for keywords or thematic activities, facilitating convenient and fast queries based on keywords or thematic activities, and making the queries more targeted.

[0096] As can be seen from the above technical solution, by setting up scheduled tasks for the target application, object data can be retrieved from the target application at regular intervals based on the scheduled tasks, thus achieving automatic data acquisition. By setting multiple prompt messages, different prompt messages are used to instruct the large language model to execute different data processing flows. That is, based on the guidance of multiple prompt messages, the large language model can execute different data processing flows on the object data respectively, thereby obtaining multiple data processing results. This not only achieves automatic data analysis, but also enables the execution of multiple data processing flows through various prompt messages, which is convenient, fast, and has high recognition accuracy. The data processing results corresponding to the multiple prompt messages are stored. When a query request for the target application is obtained, the data processing result corresponding to the query request is found from the multiple data processing results, realizing data storage and retrieval. Therefore, through scheduled tasks, multiple prompt messages, a large language model, and storage, automatic data acquisition, analysis, storage, and retrieval are achieved. The automated process reduces manual intervention, quickly processes and analyzes large amounts of data, and improves data processing efficiency.

[0097] As one possible implementation, this application provides a specific implementation of S302, namely, a specific implementation of obtaining object data from the target application according to a scheduled task, as detailed in A1-A4.

[0098] A1: Obtain source object data from the target application.

[0099] Source object data refers to unprocessed data obtained from the target application. After simple data processing, source object data can be transformed into object data.

[0100] This application does not specifically limit the acquisition method. The following description uses two methods as examples.

[0101] Method (1), offline data. Offline data is usually extracted from the business system or database corresponding to the target application through scheduled tasks (such as daily batch tasks), for example, once a day.

[0102] Method (2), Real-time Data. Real-time data sources mainly come from log data or business databases. For example, the target application will generate various logs, which are generally saved in the form of files. You can choose to extract them periodically or access them in real time.

[0103] A2: Store the source object data in different coarse data storage tables.

[0104] Coarse data storage tables are used to store source object data, and different coarse data storage tables are used to store different categories of source object data. In other words, by storing source object data in different coarse data storage tables, it is possible to classify the source object data.

[0105] One possible approach is to perform preliminary processing on the source object data before storing it in a corresponding coarse data storage table. Preliminary processing may include handling abnormal fields, standardizing field names, unifying time formats, and ensuring data consistency—all data cleaning steps—to facilitate subsequent data processing.

[0106] Data cleaning is the process of processing and optimizing raw data to improve its quality, accuracy, and consistency for better analysis, modeling, or decision-making. Since noise, incompleteness, inconsistencies, or duplication may be introduced during data acquisition, the main task of data cleaning is to identify and correct these problems.

[0107] A3: Create multiple fine data storage tables based on various sub-timed tasks and multiple coarse data storage tables.

[0108] The scheduled task comprises multiple sub-scheduled tasks. Different sub-scheduled tasks create different fine-grained data storage tables, which in turn store different object data. Each fine-grained data storage table is derived from one sub-scheduled task and at least one coarse-grained data storage table. Compared to the coarse-grained data storage tables, the fine-grained data storage tables undergo a slight aggregation, establishing naming conventions and consistent statistical indicators, thus facilitating subsequent data processing.

[0109] For example, once a sub-timed task is triggered, source object data can be retrieved from one or more coarse data storage tables associated with it and processed to obtain the corresponding fine data storage table.

[0110] One possible implementation is to create multiple dimension tables based on various sub-timed tasks and multiple coarse data storage tables, and then create multiple fine data storage tables based on these dimension tables and various crawler tasks. Further details will follow. Figure 4A and Figure 4B This will be illustrated using an example, and will not be elaborated upon further here. It should be noted that web crawling tasks are also a type of sub-timed task.

[0111] As one possible implementation, sub-scheduled tasks can be ETL tasks, such as extracting, transforming, and loading. ETL tasks can perform data preprocessing operations, such as aggregation, filtering, sorting, and data transformation, thereby reducing the workload of subsequent processing and improving data quality, preventing poor results. It should be noted that ETL tasks are also a type of sub-scheduled task.

[0112] A4: Define the data stored in the fine data storage table as object data.

[0113] Therefore, the above process includes data acquisition, storage, and transformation. Data can be obtained from the source (i.e., the target application) through methods such as web crawling. The source object data is categorized and stored based on coarse data storage tables, with different categories of source object data stored in different coarse data storage tables, thus achieving preliminary processing of the source object data. Then, multiple fine data storage tables are created based on various sub-timed tasks and multiple coarse data storage tables, achieving refined processing of the source object data. The entire process not only automates the process from data acquisition to analysis preparation but also reduces the difficulty of subsequent data analysis.

[0114] For ease of explanation, the following describes a specific implementation of A3, taking the source object data as including four categories as an example. Specifically, it describes a specific implementation of creating multiple fine data storage tables based on multiple sub-timed tasks and multiple coarse data storage tables. The details are as follows.

[0115] The source object data is categorized into four types: Category 1 (articles published for the target application), Category 2 (object actions performed for the target application), Category 3 (comments published for the target application), and Category 4 (themed events published through the target application, such as game skin promotions for game applications). The following four examples illustrate these categories.

[0116] (1) If the sub-timed task is to periodically retrieve article detail data, then a first fine data storage table is created based on the first coarse data storage table and the second coarse data storage table. The first coarse data storage table is used to store source object data belonging to the first category, the second coarse data storage table is used to store source object data belonging to the second category, and the first fine data storage table is used to store article detail data related to each article, such as the author of the article, the number of likes, the number of comments, etc.

[0117] (2) If the sub-timed task is to periodically obtain evaluations based on articles, then a second fine data storage table is created based on the first coarse data storage table and the third coarse data storage table. The second coarse data storage table stores source object data belonging to the third category, and the second fine data storage table stores comments published for each article.

[0118] (3) If the sub-timed task is to periodically retrieve the details of thematic activities, then a third fine data storage table is created based on the second coarse data storage table and the fourth coarse data storage table. The fourth coarse data storage table stores source object data belonging to the fourth category, and the third fine data storage table stores activity details data related to each thematic activity.

[0119] (4) If the sub-timed task is to periodically obtain evaluations based on the theme activities, then a fourth fine data storage table is created based on the third coarse data storage table and the fourth coarse data storage table. The fourth fine data storage table is used to store comments related to each theme activity.

[0120] Therefore, a fine data storage table can be constructed based on a sub-timed task and multiple coarse data storage tables, thereby automatically classifying and storing source object data through multiple coarse data storage tables. Based on the sub-timed task and multiple coarse data storage tables, the data can be further processed, such as light summarization, which is convenient and fast.

[0121] As mentioned above, multiple dimension tables can be created based on various sub-timed tasks and multiple coarse data storage tables, and multiple fine data storage tables can be created based on multiple dimension tables and multiple crawler tasks.

[0122] See Figure 4A This figure is a flowchart illustrating an embodiment of this application for obtaining object data. Figure 4A In this context, source object data includes articles published by objects for the target application, object behaviors generated by objects for the target application, comments published for the target application or articles, and topic activities related to the target application. Different types of source object data can be obtained through various crawling tasks, and after preliminary processing, they can be stored in different coarse data storage tables in the Operational Data Store (ODS).

[0123] Specifically, relevant data can be obtained through web scraping tasks. This involves simulating browser behavior to send requests to websites and other locations to obtain webpage content, and then using parsing tools to locate and extract the required data. For example, articles can be retrieved from specified websites or other locations using a web scraping task, such as setting the `get_listing_msalesathr_job` statement to implement a web scraping task for retrieving articles. This data is stored in the coarse data storage table `ods_article_msalesathr` (the name of the data storage table). Object behaviors are retrieved from specified websites or other locations via crawler tasks. For example, the `get_listing_reviewsathr_job` statement is used to implement a crawler task to retrieve object behaviors, and these behaviors are stored in the coarse data storage table `ods_behavior` (the name of the data storage table). Object comments are retrieved from specified websites or other locations via crawler tasks. For example, the `get_listing_reviewsathr_job` statement is used to implement a crawler task to retrieve object comments, and these comments are stored in the coarse data storage table `ods_reviewsathr` (the name of the data storage table). Topical activities are retrieved from websites or other locations via crawler tasks. For example, the `get_activity_top_job` statement is used to implement a crawler task to retrieve topic activities, and these activities are stored in the coarse data storage table `ods_activity` (the name of the data storage table).

[0124] See Figure 4B This figure is a flowchart of an embodiment of the present application for obtaining object data. Figure 4B and Figure 4A Together, they constitute a process for acquiring object data, connected through node A. In Figure 4B In this process, ETL tasks are used to further process the data in the coarse data storage table, updating the dimension table and creating the fine data storage table. It should be noted that... Figure 4B The data obtained through node A in the simplified method is the data stored in the coarse data storage table.

[0125] As mentioned above, ETL tasks involve extracting, transforming, and loading data. ETL tasks enable data preprocessing, reducing the workload of subsequent processing and improving data quality. Specifically, a daily task (hereinafter referred to as the daily task) can be implemented based on the coarse data storage table `ods_article_msalesathr` (the name of the first coarse data storage table) and a daily ETL task. To distinguish it from other ETL tasks, the `add_new_article` statement can be used to implement the ETL task for retrieving new articles, creating a dimension table `dim_articles` (the name of the data storage table). Then, based on the dimension table `dim_articles`, the coarse data storage table `ods_behavior` (the name of the second coarse data storage table), and a crawler task, platform page interfaces are called from websites and other locations daily (i.e., a sub-timed task that periodically retrieves article detail data). For example, the `get_articledetail_job` statement can be used to implement the crawler task for retrieving article-related detail data, creating a fine data storage table `ods_articledetail` (the name of the first fine data storage table). Based on the dimension table dim_articles (the name of the dimension table), the coarse data storage table ods_reviewsathr (the name of the third coarse data storage table), and the crawler task calling the platform page interface to obtain incrementally updated articles every day (i.e., a sub-timed task that periodically obtains reviews based on articles), such as setting the get_review_byarticle statement to implement the crawler task that periodically obtains reviews based on articles, a fine data storage table ods_review_byarticle (the name of the second fine data storage table) is created.

[0126] In addition, based on the dimension table dim_articles (the name of the dimension table), the coarse data storage table ods_activity (the name of the fourth coarse data storage table), and the daily incrementally updated list of topic activities obtained by the crawler task, such as setting the get_activitys_byarticle statement to implement the crawler task to obtain the list of topic activities, a fine data storage table ods_activitys_article (the name of the data storage table) is created.

[0127] Similarly, new themed activities can be retrieved daily based on the coarse data storage table `ods_activity` (i.e., the name of the fourth coarse data storage table), the fine data storage table `ods_activitys_article` (the name of the data storage table), and a sub-scheduled task. This can be represented by `add_new_activity` to indicate the ETL task for retrieving new themed activities. A dimension table `dim_activity` (the name of the dimension table) can be created. Then, a sub-scheduled task can be created based on the dimension table `dim_activity` (the name of the dimension table), the coarse data storage table `ods_behavior` (i.e., the name of the second data storage table), and a crawler task to periodically retrieve detailed data of the themed activities. For example, a crawler task can be set up with the statement `get_activity detail_job` to periodically retrieve detailed data of the themed activities. Based on the themed activities, detailed data related to the themed activities can be retrieved (which can be a daily task that calls the platform page interface). A fine data storage table `ods_activity detail` (i.e., the name of the third fine data storage table) can be created.

[0128] Based on the dimension table dim_activity (the name of the dimension table), the coarse data storage table ods_reviewsathr (the name of the third coarse data storage table), and the crawler task calling the platform page interface to obtain incrementally updated data every day (i.e., a sub-scheduled task that periodically obtains evaluations based on the theme activity), the crawler task obtains evaluations based on the theme activity, such as setting the get_review_byarticle_job statement to implement the periodic retrieval of theme activity details data, and creates the fine data storage table ods_review_byactivity (the name of the fourth fine data storage table).

[0129] In addition, the dimension table dim_articles (the name of the dimension table) can be updated based on the subtask of the dimension table corresponding to the updated article and the fine data storage table ods_articledetail (i.e., the name of the first fine data storage table). The dimension table dim_activity (the name of the dimension table) can be updated based on the subtask of the dimension table corresponding to the updated topic activity and the fine data storage table ods_activity_detail (i.e., the name of the third fine data storage table).

[0130] As one possible implementation, this application provides a specific implementation of A2, which is a specific implementation of storing source object data in different coarse data storage tables, see B1-B2 for details.

[0131] B1: Store source object data belonging to the fifth category in the fifth coarse data storage table.

[0132] The source object data also includes a fifth category, which consists of keywords generated based on the target application. These keywords are stored in the fifth coarse data storage table.

[0133] B2: Search for keywords stored in the fifth coarse data storage table in the target application, and create the sixth coarse data storage table.

[0134] The sixth coarse data storage table is used to store associated data related to keywords.

[0135] See also Figure 4A Keywords can be obtained based on crawling tasks. For example, the `get_keyword_top_job` statement can be used to implement a crawling task that retrieves keywords, and the keywords are stored in the fifth coarse data storage table `ods_keyword` (the name of the data storage table). This can be used to crawl new links (such as keyword-linked links) on pages related to themed activities. Then, based on the crawling task, a search can be performed in the target application according to the keywords. For example, the `get_lision_searchkeyword` statement can be used to implement a crawling task that searches in the target application according to the keywords, and the search results (i.e., related data associated with the keywords) are stored in the sixth coarse data storage table `ods_activity_searchkeyword` (the name of the data storage table).

[0136] B3: The data stored in the fine data storage table and the data stored in the sixth coarse data storage table are identified as object data.

[0137] Not only can the data stored in fine data storage tables such as the first fine data storage table to the fourth fine data storage table be identified as object data, but the data stored in the sixth coarse data storage table can also be identified as object data, so that subsequent processing can continue based on the processed object behavior, thereby obtaining multiple data processing results.

[0138] Therefore, by acquiring data related to keywords and processing it further to obtain the fifth and sixth coarse data storage tables, keyword-based queries can be implemented based on these tables, which is convenient, fast, reduces the learning cost for users, and provides a better user experience.

[0139] As described above, in embodiment S303, the large language model is instructed to perform different data processing procedures on the object data based on multiple prompt messages, thereby obtaining data processing results corresponding to each prompt message. Embodiments of this application do not specifically limit the content of the multiple prompt messages; the following uses various prompt messages as examples to describe different data processing procedures.

[0140] (1) First prompt message.

[0141] The first prompt is used to extract the topic from the object data. The specific data processing flow for the large language model to perform on the object data based on the first prompt is detailed in C1-C7.

[0142] C1: Extract object data to obtain multiple undetermined topics.

[0143] Multiple undetermined topics are extracted from object data. These undetermined topics may be topics, but due to the diversity of language expression, the same semantic expression may be expressed in different forms, resulting in multiple homogeneous and heterogeneous undetermined topics. Based on this, embodiments of this application provide an effective method for merging homogeneous and heterogeneous undetermined topics, as detailed in C2-C7.

[0144] The embodiments of this application do not specifically limit the method of extracting the subject to be determined. For example, the subject to be determined can be obtained by identifying the object data through semantic recognition or other methods, and the subject that can represent the idea of ​​each object data center can be used as the subject to be determined.

[0145] C2: Extract features from multiple undetermined topics to obtain topic vectors corresponding to each undetermined topic.

[0146] This application does not specifically limit the feature extraction method, such as using the Bag of Words (BoW) model, Transformer, Long Short-Term Memory Network (Bidirectional Encoder Representation from Transformers, BERT) model, etc. This application does not specifically limit the method.

[0147] A topic vector is a vector obtained by extracting features from a given topic.

[0148] C3: Cluster multiple topic vectors to obtain multiple vector clusters.

[0149] Clustering is the process of dividing a dataset into different vector clusters based on a specific criterion (such as distance), maximizing the similarity of data objects within the same vector cluster while maximizing the dissimilarity of data objects in different vector clusters. In other words, after clustering, topic vectors with the same semantic meaning will be grouped together, while topic vectors with different semantic meanings will be assigned to different vector clusters.

[0150] C4: The topic vectors whose distance from the center of the target vector cluster is less than the distance threshold are identified as the target topic vectors corresponding to the target vector cluster.

[0151] The following explanation uses one vector cluster from multiple vector clusters, namely the target vector cluster, as an example. The topic vectors whose distance to the center of the target vector cluster is less than a distance threshold, i.e., the topic vectors closest to the center, are determined as the target topic vectors corresponding to the target vector cluster. This application does not specifically limit the distance threshold; for example, it may use the topic vector closest to the center.

[0152] In other words, the target topic vector corresponding to the target vector cluster is close to the center of the target vector cluster, so the target topic vector can represent the category corresponding to the target vector cluster.

[0153] C5: Determine the undetermined topic corresponding to the target topic vector as the topic corresponding to the target vector cluster.

[0154] The target topic vector can represent the category corresponding to the target vector cluster, thereby determining the topic corresponding to the target vector cluster from the undetermined topic corresponding to the target topic vector.

[0155] C6: Take multiple vector clusters as target vector clusters respectively, and obtain the topics corresponding to the multiple vector clusters respectively.

[0156] C7: Define multiple topics as the subject of object data.

[0157] Therefore, by pre-extracting multiple undetermined topics from the object data, and using clustering to identify a representative undetermined topic from each vector cluster as the topic of that category, it is possible to merge words with the same meaning but different forms, or to distinguish the same word expressing different meanings in different contexts, thereby improving the accuracy of the topics.

[0158] (2) Second prompt message.

[0159] D1: Get the second prompt template.

[0160] The second prompt template is a type of prompt template. Prompt templates can define the semantic content, text format, and word count of the prompt information. Filling in the prompt template generates prompt information used to execute a data processing flow. For example, after filling in the second prompt information template, a second prompt information can be obtained, used to extract the theme, sentiment, and intent from the object data.

[0161] As one possible implementation, taking a game as the target application as an example, the second prompt template can be as follows:

[0162] You are an analytics assistant for a mobile game forum. Please extract the following information from user comments and return it in JSON (a lightweight data interchange format):

[0163] 1. The themes involved (

Theme Set

[0164] 2. The subject's emotions (

emotional set

[0165] 3. User intent (

Intent set

[0166] Example:

[0167] { 【】

[0169] }

[0170] D2: Based on the second prompt template, obtain the theme set, sentiment set, and intention set.

[0171] A topic collection comprises multiple topics related to the target application. A topic refers to content extracted from relevant content (such as articles) that expresses its meaning. Taking a game as an example, topics could include articles, products, and themed events, with various topics forming the topic collection. Posts / articles could include game guides, game tournament discussions, and program bug reports. Products could include mobile game skins and merchandise. Themed events could include limited-time events and live tournament broadcasts.

[0172] An emotion set includes multiple emotion types, which represent the polarity of an emotion and can be positive, negative, neutral, etc. Multiple emotion types constitute an emotion set.

[0173] An intent set comprises multiple intent types, each representing the intent of an object. These various intent types together constitute the intent set. Continuing with the example of a game as the target application, intent types could include reporting bugs, sharing strategies, complaining to customer service, participating in events, and providing product reviews.

[0174] D3: Generate a second prompt message based on the second prompt template, theme set, emotion set, and intention set.

[0175] Fill the corresponding positions in the second prompt template with the theme set, emotion set, and intention set respectively to obtain the second prompt message. For example, the second prompt message can be as follows:

[0176] You are an analytics assistant for a mobile game forum. Please extract the following information from user comments and return it in JSON format:

[0177] 1. The topic type (posts, products, events) must match game terminology;

[0178] 2. Subject's emotions (positive / negative / neutral) and keywords indicating the reasons;

[0179] 3. User intent (reporting bugs, sharing tips, complaining to customer service, participating in activities, and reviewing products).

[0180] Example input:

[0181] "The new skin XXX has amazing special effects, but the anniversary event tasks are too difficult, and I can't get the rewards at all!"

[0182] Example output:

[0183] {

[0184] "theme":[

[0185] {"Name":"XXX","Type":"Product","Emotion":"Positive","Reason":["Explosive Special Effects"]},

[0186] {"Name":"Anniversary Celebration Event","Type":"Event","Emotion":"Negative","Reason":["Tasks are too difficult","Can't get rewards"]}

[0187] ],

[0188] "Intent": ["Product Review", "Event Complaint"]

[0189] }

[0190] D4: Based on the second prompt, instruct the large language model to match the object data to obtain the target topic corresponding to the object data, the target sentiment type corresponding to the target topic, and the target intent type corresponding to the target topic.

[0191] The target topic is one of the multiple topics included in the topic set, the target emotion type is one of the multiple emotion types included in the emotion set, and the target intention type is one of the multiple intention types included in the intention set.

[0192] The second prompt information guides the large language model to learn, thereby matching the object data with the topic set, sentiment set, and intent set in the second prompt information. This allows the model to obtain the target topic for representing the object data from the topic set, the target sentiment type for representing the object data from the sentiment set, and the target intent type for representing the object data from the intent set.

[0193] Therefore, by designing a second prompt template and matching the theme set, sentiment set, and intent set for each application, the theme set, sentiment set, and intent set matching the target application can be automatically populated into the second prompt template to generate the second prompt information. Subsequently, based on the second prompt information, the large language model can be guided to output the target theme, target sentiment type, and target intent type corresponding to the object data. This not only enables automatic identification of the main content in the object data, reducing manual intervention and improving efficiency, but also automatically supplements the corresponding theme set, sentiment set, and intent set for different target applications, making it convenient, fast, and more accurate.

[0194] As one possible implementation, intent recognition can also be performed separately. Continuing with the example of a game as the target application, intent types could include product reviews (e.g., skin feel, merchandise quality), activity participation (e.g., rule inquiries, reward collection), game discussions (e.g., hero strength, team composition), and customer service complaints (e.g., delivery delays, refund issues). The corresponding prompts could then be:

[0195] Please determine the intent behind the user's comment by selecting from the following tags:

[0196] [Strategy Guides][Bug Feedback][Event Inquiries][Skin Recommendations][Customer Complaints]

[0197] Type: "Game character A's E skill often lags when passing through walls, when will it be fixed?"

[0198] Output: {"Intent":["BUG Feedback"]}.

[0199] (3) Third prompt message.

[0200] E1: Get the third prompt template.

[0201] The third prompt template is a type of prompt template. For example, after the third prompt information template is filled, third prompt information can be obtained, which is used to mine the logical chain between entities in the object data. For example, the third prompt information can be used to extract the "entity A-entity relationship-entity B" triple from the object data and mark the entity relationship, emotional influence intensity, etc.

[0202] As one possible implementation, the third-party prompt template could be as follows:

[0203] You are an analyst at a mobile game forum. Please extract the relationships between entities from user comments and return them in JSON format:

[0204] Please analyze user reviews and complete the following tasks:

[0205] **Task 1: Entity Recognition**

[0206] - Type range: [[Entity type set]]

[0207] - Special Note: The skin name must be complete (e.g., "XXX" not "X").

[0208] **Task 2: Relationship Extraction**

[0209] - Relationship type: [

Entity Relationship Collection

[0210] Output format: "Entity A → Relationship → Entity B"

[0211] **Task 3: Sentiment Analysis**

[0212] -Score each entity based on its different parameter dimensions (

Parameter Dimensions

[0213] Example:

[0214] { 【】

[0216] }

[0217] E2: Based on the third prompt template, obtain the entity relationship set, entity type set, and the parameter dimensions corresponding to each entity type.

[0218] Each entity refers to an object related to the target application, such as the name of various entities, and multiple entities constitute an entity set.

[0219] An entity relationship set includes various entity relationships. Each entity relationship describes the relationship between two objects, such as purchase, participation, discussion, comparison, complaint, etc. Multiple entity relationships constitute an entity relationship set.

[0220] An entity type set refers to the type of an entity, such as skin, event, merchandise, bug, player, etc. Multiple entity types constitute an entity type set.

[0221] Parameter dimensions are used to describe the dimensions used to evaluate entities. For example, the evaluation dimensions for entities belonging to the surrounding area are quality and comparison, while the evaluation dimensions for entities belonging to services are service satisfaction. Based on parameter dimensions, special effects (quality dimension) and operational experience (functional dimension) can be evaluated separately, thereby achieving multi-emotional dimension separation and enabling multi-dimensional sentiment analysis.

[0222] E3: Generate third-party prompt information based on the third-party prompt template, entity relationship set, entity type set, and parameter dimensions.

[0223] The entity relationship set, entity type set, and parameter dimensions are populated into the corresponding positions in the third prompt template to obtain the third prompt information. For example, the third prompt information can be as follows:

[0224] You are an analyst at a mobile game forum. Please extract the relationships between entities from user comments and return them in JSON format:

[0225] Please analyze user reviews and complete the following tasks:

[0226] **Task 1: Entity Recognition**

[0227] -Type range: [Skins, Events, Merchandise, Bugs, Players]

[0228] - Special Note: The skin name must be complete (e.g., "XXX" not "X").

[0229] **Task 2: Relationship Extraction**

[0230] -Relationship types: [Purchase, Participation, Discussion, Comparison, Complaint]

[0231] Output format: "Entity A → Relationship → Entity B"

[0232] **Task 3: Sentiment Analysis**

[0233] -Score each entity across different parameter dimensions (the surrounding parameter dimensions are price: -1 to 1, quality: -1 to 1, and service: -1 to 1).

[0234] Example input:

[0235] "Although the B doll is of good quality, it is far inferior to the limited edition from the last event, and customer service does not solve the problem."

[0236] Example output:

[0237] {

[0238] "entity":[

[0239] {"Name":"B Doll","Entity Type":"Merchandise","Emotion":{"Quality":0.6,"Comparison":-0.8}},

[0240] {"Name":"Customer Service","Entity Type":"Service","Emotion":{"Service":-0.9}}

[0241] ],

[0242] "Entity Relationship": ["B Doll → Comparison → Limited Edition", "User → Complaint → Customer Service"]

[0243] }

[0244] E4: Based on the third prompt, instruct the large language model to match the object data to obtain the first entity, the second entity, the target entity relationship between the first entity and the second entity, the first entity type of the first entity, the second entity type of the second entity, the evaluation of the first entity based on the parameter dimension corresponding to the first entity type, and the evaluation of the second entity based on the parameter dimension corresponding to the second entity type.

[0245] The first entity is one of a set of entities, the second entity is one of a set of entities, the first entity is the entity being evaluated, the second entity is the entity being compared, the target entity relationship is one of a set of entity relationships, used to describe the entity relationship between the first entity and the second entity, the first entity type is one of a set of entity types, used to describe the type of the first entity, and the second entity type is one of a set of entity types, used to describe the type of the second entity.

[0246] The third prompt information guides the large language model to identify the first entity and the second entity included in the object data. The target entity relationship between the first entity and the second entity is matched from the entity relationship set. The type of the first entity and the type of the second entity are matched from the entity type set. The entity is evaluated according to the parameter dimensions corresponding to each entity type, thus obtaining the evaluation of the first entity and the evaluation of the second entity.

[0247] Therefore, by designing a structured third-prompt template, a set of entity relationships matching each application, a set of entity types, and parameter dimensions corresponding to each entity type, the third-prompt template can be automatically populated with the set of entity relationships matching the target application, the set of entity types, and the parameter dimensions corresponding to each entity type, generating third-prompt information. Subsequently, based on the third-prompt information, the large language model can be guided to output multiple entities included in the object data, the entity relationships between multiple entities, entity types, and parameter dimensions, realizing the automatic mining of logical chains between entities in the object data. Not only can the structured third-prompt information automatically identify logical chains between entities in object data, reducing manual intervention and improving efficiency, but it can also automatically supplement the corresponding set of entity relationships, the set of entity types, and the parameter dimensions corresponding to each entity type for different target applications, making it convenient, fast, and more accurate.

[0248] As one possible implementation, negative words complicate sentiment analysis. Therefore, to improve the accuracy of sentiment analysis, this application establishes a first mapping relationship, which describes the impact of multiple negative words on the evaluation.

[0249] For example, the presence of the negative word "unreasonable" can be directly interpreted as a negative emotion, lowering the evaluation score by 0.5 points. This establishes a primary mapping relationship between the negative word "unreasonable" and the 0.5-point reduction. Conversely, the presence of the negative word "don't think so" can mitigate subsequent negativity, potentially increasing or decreasing the score, with the magnitude of the increase or decrease reflecting the degree of impact. Furthermore, the presence of the negative word "not bad" can be directly interpreted as neutral to slightly positive, potentially increasing or not adjusting the score. Subsequently, third-party prompt information can be generated based on the third-party prompt template, the primary mapping relationship, the entity relationship set, the entity type set, and the parameter dimensions.

[0250] As one possible implementation, the first mapping relationship can be added as Task 4 to the third prompt template, such as:

[0251] **Task 4: Negation Word Processing**

[0252] Note the double negative structure and the steps for judging sentiment:

[0253] [First mapping relationship]

[0254] When filling in the third prompt template, you can add 1. "I don't think so" → weakening the subsequent negation; 2. "Unreasonable" → lowering the score by 0.5 to the first mapping relationship.

[0255] Therefore, by establishing a mapping relationship between multiple negative words and the degree of influence of their corresponding evaluations through the first mapping relationship, the influence of negative words on emotions can be clearly defined, thereby improving the accuracy of emotion recognition.

[0256] (4) Fourth prompt message.

[0257] In related technologies, statistical analysis of object data is generally based on preset fixed indicators (such as positive review rate, number of complaints, etc.). However, this fixed method cannot detect changes in object data, and therefore cannot accurately reflect the object data.

[0258] Based on this, embodiments of this application provide a fourth prompt message, which instructs a large language model to mine new metrics from comments in real time and automatically generate statistical formulas. Specifically, according to the fourth prompt message, the large language model is instructed to perform semantic understanding on the object data to obtain new metrics and the corresponding statistical methods for the new metrics.

[0259] For example, one form of the fourth prompt message is as follows:

[0260] prompt="""

[0261] Please identify new metrics that may impact business performance from the following game forum comments and provide their calculation formulas:

[0262] Example input: "Many players have mentioned that the new skin causes their phones to overheat."

[0263] Example output: {"metric":"Skin power consumption coefficient","formula":"Number of comments related to overheating / Total number of skin comments"}

[0264] enter:{}

[0265] "".format(comments[:1000])

[0266] Therefore, by introducing a fourth cue, the large language model can be instructed to mine new metrics from comments in real time and automatically generate statistical formulas. This allows for the real-time discovery of new changes in object data, enabling the target application to process them quickly and effectively, thereby improving the user experience.

[0267] (5) Fifth prompt message.

[0268] In related technologies, multiple data processing results are generally displayed using static charts (such as line charts, bar charts, etc.). However, due to the different use of various prompts, the data processing results may differ. If only a fixed static chart is used for display, it may not be able to highlight the different characteristics of the different data processing results, making it impossible to quickly view the data processing results.

[0269] Based on this, embodiments of this application provide a fifth prompt message, which can analyze data processing results, automatically select the best visualization format based on the analysis conclusions, and generate interpretive text. See F1-F3 for details.

[0270] F1: Get the fifth prompt message.

[0271] The fifth prompt is used to analyze the data processing results, automatically select the best visualization format based on the analysis conclusions, and generate interpretive text.

[0272] As one possible implementation, the fifth prompt message could be as follows:

[0273] query(f"""

[0274] Generate visualization code based on data characteristics:

[0275] Data source: {df.describe()}

[0276] Data pattern: Negative reviews focus on high-priced skins

[0277] Display Mode: Use Altair to create interactive charts that highlight the correlation between price and negative sentiment.

[0278] Based on the fifth prompt information, the data processing results of the data source can be analyzed to obtain the corresponding data pattern, thereby generating Altair / Vega-Lite code, rendering interactive charts, and generating graphic reports, etc.

[0279] F2: Based on the fifth prompt, instruct the large language model to determine the data patterns corresponding to the multiple data processing results, and determine the display mode based on the data patterns.

[0280] Based on the fifth prompt, the large language model is guided to analyze multiple data processing results and obtain the data patterns corresponding to each data processing result. These data patterns can indicate the characteristics of the corresponding data processing results. Therefore, the display pattern determined based on the data pattern can better reflect the characteristics of the corresponding data processing results.

[0281] F3: Displays multiple data processing results based on the display mode.

[0282] For example, multiple data processing results can be displayed in a targeted manner based on word clouds, time series graphs, cluster plots, line charts, bar charts, etc. Word clouds are used to display high-frequency topics, visually showing popular topics. Time series graphs can show the trend of topics changing over time, helping to identify dynamic changes. Cluster plots can display the clustering results of topics, showing the distribution and relationships of different topics.

[0283] Therefore, by analyzing the data models of multiple data processing results through the fifth prompt information, it is possible to determine the display mode that best highlights this data pattern. Based on this display mode, multiple data processing results can be displayed, better emphasizing their characteristics and facilitating user viewing. Combined with statistical analysis and visualization, this provides deeper business insights, helping to identify market trends and customer needs.

[0284] (6) Sixth prompt message.

[0285] During the analysis of object data, the object data may include a large number of informal expressions (such as "yyds", "critical hit rate yabi", etc.). Moreover, the general large language model does not understand professional terms (such as "A skin" and "jungle rhythm") well enough, which may lead to the low accuracy of the data processing results obtained by the large language model after executing the data processing flow.

[0286] Based on this, embodiments of this application provide a sixth prompt message, which includes multiple mapping relationships. Each mapping relationship is used to map between two expressions. For example, the second mapping relationship is between informal and formal expressions, and the third mapping relationship is between technical terms and terminology explanations. Specifically, according to the sixth prompt message, the large language model is instructed to escape the object data to obtain the required standard expression.

[0287] For example, if the sixth prompt includes a second or third mapping relationship, the standardization problem of unstructured text can be solved by combining the sixth prompt with a hybrid cleaning method that integrates a rule engine and a large model. One form of the sixth prompt is as follows:

[0288] slang_map = {"yyds":"Eternal God","yabi":"Yabi (Hero Abbreviation)"}

[0289] corrected = gpt4_query(f"Please convert the following game forum text to standard expression: {text}")

[0290] Therefore, by establishing a mapping relationship between words and adding them to the sixth prompt, data cleaning can be achieved through the sixth prompt, thus addressing the standardization problem of unstructured text.

[0291] (7) Seventh prompt message.

[0292] During the analysis of object data, the data may contain multiple languages, such as Chinese and English, which makes the skin feel very unpleasant. This lack of standardization in the object data can lead to lower accuracy in the data processing results.

[0293] Based on this, embodiments of this application provide a seventh prompt message, which can standardize object data. For example, one form of the seventh prompt message is as follows:

[0294] return gpt4_query(f"Convert the following mixed Chinese and English text to plain Chinese: {text}")

[0295] As one possible implementation, the seventh prompt could also instruct the large language model to first identify the language type, then determine the first language and the second language based on the number of texts included in the language category, where the first language has a larger number of texts and the second language has a smaller number of texts. Finally, it could instruct the large language model to convert the text in the second language into text in the first language.

[0296] Therefore, the seventh prompt can automatically normalize non-standard mixed inputs, improving the accuracy of subsequent data processing results.

[0297] (8) Eighth prompt message.

[0298] This application embodiment also provides an eighth prompt message, which can instruct the large language model to analyze the advantages and disadvantages of entities from object data in order to provide opinions.

[0299] For example, one form of the eighth prompt message is as follows:

[0300] If you are a game product manager, please summarize the advantages and disadvantages of [Skin A] based on the following comments:

[0301] Require:

[0302] 1. Each advantage / disadvantage should be accompanied by its frequency (percentage).

[0303] 2. The drawback needs to be distinguished between design issues (which can be improved) and subjective preferences (which do not need to be addressed).

[0304] Example input: "Special effects are amazing, but the feel is clunky; at 178 yuan, it's 50% more expensive than similar skins."

[0305] Output:

[0306] {

[0307] "advantage":[

[0308] {"Aspect":"Special Effects","Frequency":"82%","Example":"Explosive Special Effects"}

[0309] ],

[0310] "shortcoming":[

[0311] {"Aspect":"Price","Frequency":"65%","Type":"Design Issue"},

[0312] {"Aspect":"Feel","Frequency":"57%","Type":"Subjective Preference"} ]

[0314] }

[0315] For example, another form of the eighth prompt message is as follows:

[0316] You are a game product manager. Please generate three improvement suggestions based on player preferences for game A, providing feedback on both its strengths and weaknesses.

[0317] Therefore, the eighth prompt can instruct the large language model to analyze the entity, thereby obtaining the entity's advantages and disadvantages, and even the frequency of the advantages and disadvantages and the type of the disadvantages, to help users quickly analyze the entity.

[0318] To facilitate a further understanding of the technical solutions provided in the embodiments of this application, the following example uses a server as the execution subject of the data processing method provided in the embodiments of this application and a game application as the target application to provide an overall exemplary description of the data processing method.

[0319] The data processing system will be described below.

[0320] See Figure 5 The figure is a schematic diagram of a data processing system provided in an embodiment of this application.

[0321] (1) Obtain object data by calling the external application programming interface (API) through computer services.

[0322] The computer service can be Lambda, etc., and this application does not specifically limit it. Lambda is an ideal computing service, suitable for application scenarios that require rapid vertical scaling and can shrink to zero when not needed.

[0323] In other words, the aforementioned S302 is executed through a computer service, that is, object data is obtained from the target application according to a scheduled task.

[0324] (2) Use state machines and task management and scheduling computer services.

[0325] For example, Step Functions are used to manage and schedule Lambda tasks. Step Functions are based on state machines and tasks. In Step Functions, the state machine, called a workflow, is a series of event-driven steps. Each step in the workflow is called a state. For example, a task state represents a unit of work performed by another service.

[0326] Management and scheduling can involve setting up scheduled tasks for a target application, allowing the server to receive these scheduled tasks, or even multiple notification messages.

[0327] (3) Send the obtained object data to the distributed data stream processing platform.

[0328] A distributed data stream processing platform can publish, subscribe to, store, and process data streams in real time. It is designed to process data streams from multiple sources and deliver them to multiple consumers. The distributed data stream processing platform can be Kafka, and the processed data results are written to a primary storage device via tools such as Kafka Connector (a tool for data import and export).

[0329] (4) Execute data processing flow through computing cluster.

[0330] Data processing and analysis can be performed through computing clusters (such as EMR or EMR Serverless). For example, by executing the aforementioned S303, the large language model is instructed to perform different data processing procedures on the object data based on multiple prompts, and the data processing results corresponding to the multiple prompts are obtained.

[0331] (5) Store the obtained data processing results into the first memory.

[0332] In other words, after the data processing process is executed by the computing cluster, the resulting data processing results are stored in the first memory, i.e., the aforementioned S304 is executed, i.e., multiple data processing results are stored.

[0333] (6) Store the data processing results of some frequently queried data in the sub-secondary memory.

[0334] Specifically, multiple data processing results are divided into a first result set and a second result set. The first result set is stored in a first storage space, and the second result set is stored in a second storage space. By classifying and storing the data processing results in different storage spaces—that is, storing frequently queried data processing results in higher-cost storage spaces and less frequently queried data processing results in lower-cost storage spaces—storage costs can be reduced while improving the speed of subsequent data queries, thereby enhancing the user experience.

[0335] (7) Obtain data processing results from the second or first memory via API and cloud server.

[0336] In other words, users can input query content for the target application, thereby generating a query request and executing the aforementioned S305, that is, if a query request for the target application is obtained, the data processing result corresponding to the query request is searched from the multiple stored data processing results.

[0337] The following examples illustrate several ways to display the data processing results corresponding to different query requests.

[0338] Scenario 1.

[0339] See Figure 6 This figure is a schematic diagram of a target application dashboard analysis provided in an embodiment of this application. The target application dashboard analysis can display the overall distribution of the target application. The dashboard analysis allows users to first input a topic and date to perform a query, and then, after generating a query request, it can display the following content:

[0340] 1. Overview data on the number of articles published, total likes, and total comments on related topics.

[0341] 2. Keyword summary graphs and sentiment analysis of the target audience for the current search results.

[0342] 3. Distribution of article themes.

[0343] 4. Age distribution of community users.

[0344] By displaying multiple data processing results, these indicators can provide feedback on the overall situation of the target application, allowing the user to have a basic understanding of the current query content.

[0345] Scenario 2.

[0346] Click Figure 6 The details button (601) in the upper right corner also displays detailed information about the community overview. (See also...) Figure 7 This figure is a schematic diagram illustrating the details of an indicator provided in an embodiment of this application. Figure 7 The following details are shown in the image:

[0347] 1. The trend of changes in the subject's emotions during the current time period.

[0348] 2. Article rating trends.

[0349] 3. The article selects the opinions of the participants regarding the advantages and disadvantages of the game, and sorts them according to the percentage of advantages and disadvantages.

[0350] 4. Categorize the topics of the articles.

[0351] 5. A ranking of articles with high popularity.

[0352] It should be noted that points 3-5 above can be viewed for details, which will be explained below.

[0353] See Figure 8 This figure is a schematic diagram illustrating one advantage and disadvantage of an embodiment of this application. Figure 8 In the middle, is Figure 7The diagram shown is a further examination of point 3. The "Pros and Cons Details" section refers to the extraction of the game's strengths and weaknesses from articles using methods such as the eighth prompt and large language models mentioned earlier. This information is then summarized and analyzed to determine the percentage of each strength and weakness, and the specific reasons are analyzed. This information is listed in a table, and users can click on "Details" to access a page summarizing articles searched based on the current strengths and weaknesses.

[0354] See Figure 9 This figure is a schematic diagram of a subject matter provided in an embodiment of this application. Figure 9 In the middle, is Figure 7 The fourth point provides a further view of the displayed diagram. Here, "topic" refers to the topics extracted from the articles, which are then summarized. This interface displays the title of each topic, along with optimization suggestions and user sentiment analysis (both positive and negative) for the articles on that topic. It also shows the individual articles corresponding to that topic, along with their content, number of likes, number of replies, creation time, optimization suggestions, etc. Clicking "View Details" will take you to see the article's details. (Similar to the topic search interface in activity analysis).

[0355] See Figure 10 This figure is a schematic diagram of heat provided in an embodiment of this application. Figure 10 In the middle, is Figure 7 The diagram shown is a further look at point 5. The "Popularity Details" section filters articles based on likes and views, displaying their key information in a list. Clicking "View Details" allows you to see the article's details, similar to an event analysis / details interface.

[0356] See Figure 11 This figure is a schematic diagram of a topic activity analysis provided in an embodiment of this application. Figure 11 The interface displayed is the activity analysis interface. You can search by topic or activity link to get a list of related articles. The article list shows the main information of the searched content, such as the number of replies and likes.

[0357] See Figure 12 This figure is a schematic diagram of the details of a themed activity provided in an embodiment of this application. (Click...) Figure 11 The detailed control 1101 shown allows users to view the article's content, number of likes, number of replies, ratings of comments for the current article, optimization suggestions for the current article, analysis and refinement of comments for the current article, and summary of optimization opinions for the current comments.

[0358] In response to the data processing method described above, this application also provides a corresponding data processing apparatus so that the above data processing method can be applied and implemented in practice.

[0359] See Figure 13 This figure is a schematic diagram of the structure of a data processing device provided in an embodiment of this application. Figure 13 As shown, the data processing device 1300 includes: an acquisition unit 1301, an acquisition unit 1302, an analysis unit 1303, a storage unit 1304, and a query unit 1305;

[0360] The acquisition unit 1301 is used to acquire multiple prompt messages and timed tasks for the target application, and different prompt messages are used to instruct the execution of different data processing processes;

[0361] The acquisition unit 1302 is used to acquire object data from the target application according to the timed task, and the object data is used to characterize the object-related behavioral data generated based on the target application.

[0362] The analysis unit 1303 is used to instruct the large language model to perform different data processing procedures on the object data based on the multiple prompts, so as to obtain the data processing results corresponding to the multiple prompts respectively.

[0363] The storage unit 1304 is used to store multiple data processing results;

[0364] The query unit 1305 is used to search for the data processing result corresponding to the query request from among the stored multiple data processing results if a query request for the target application is obtained.

[0365] As can be seen from the above technical solution, by setting up scheduled tasks for the target application, object data can be retrieved from the target application at regular intervals based on the scheduled tasks, thus achieving automatic data acquisition. By setting multiple prompt messages, different prompt messages are used to instruct the large language model to execute different data processing flows. That is, based on the guidance of multiple prompt messages, the large language model can execute different data processing flows on the object data respectively, thereby obtaining multiple data processing results. This not only achieves automatic data analysis, but also enables the execution of multiple data processing flows through various prompt messages, which is convenient, fast, and has high recognition accuracy. The data processing results corresponding to the multiple prompt messages are stored. When a query request for the target application is obtained, the data processing result corresponding to the query request is found from the multiple data processing results, realizing data storage and retrieval. Therefore, through scheduled tasks, multiple prompt messages, a large language model, and storage, automatic data acquisition, analysis, storage, and retrieval are achieved. The automated process reduces manual intervention, quickly processes and analyzes large amounts of data, and improves data processing efficiency.

[0366] As one possible implementation, the acquisition unit 1302 is specifically used for:

[0367] Obtain source object data from the target application;

[0368] The source object data is stored in different coarse data storage tables, and the different coarse data storage tables are used to store source object data of different categories;

[0369] Based on multiple sub-timed tasks and multiple coarse data storage tables, multiple fine data storage tables are created, each of which is obtained based on one of the sub-timed tasks and at least one of the coarse data storage tables, and the timed tasks include the multiple sub-timed tasks.

[0370] The data stored in the detailed data storage table is determined as the object data.

[0371] As one possible implementation, if the source object data includes a first category, a second category, a third category, and a fourth category, where the first category is articles published for the target application, the second category is object behaviors performed for the target application, the third category is comments published for the target application, and the fourth category is topic activities published through the target application, then the acquisition unit 1302 is specifically used for:

[0372] If the sub-timed task is to periodically retrieve article detail data, then a first fine data storage table is created based on the first coarse data storage table and the second coarse data storage table. The first coarse data storage table is used to store source object data belonging to the first category, the second coarse data storage table is used to store source object data belonging to the second category, and the first fine data storage table is used to store article detail data related to each of the articles.

[0373] If the sub-timed task is to periodically obtain evaluations based on articles, then a second fine data storage table is created based on the first coarse data storage table and the third coarse data storage table. The third coarse data storage table is used to store source object data belonging to the third category, and the second fine data storage table is used to store comments published for each of the articles.

[0374] If the sub-timed task is to periodically obtain the theme activity details data, then a third fine data storage table is created based on the second coarse data storage table and the fourth coarse data storage table. The fourth coarse data storage table is used to store source object data belonging to the fourth category, and the third fine data storage table is used to store activity details data related to each of the theme activities.

[0375] If the sub-timed task is to periodically obtain evaluations based on a theme activity, then a fourth fine data storage table is created based on the third coarse data storage table and the fourth coarse data storage table. The fourth fine data storage table is used to store comments related to each of the theme activities.

[0376] As one possible implementation, the source object data may also include a fifth category, which is keywords generated based on the target application. In this case, the acquisition unit 1302 is specifically used for:

[0377] The source object data belonging to the fifth category is stored in the fifth coarse data storage table;

[0378] Search the target application for keywords stored in the fifth coarse data storage table, and create a sixth coarse data storage table, which is used to store associated data related to the keywords;

[0379] The step of determining the data stored in the detailed data storage table as the object data includes:

[0380] The data stored in the fine data storage table and the data stored in the sixth coarse data storage table are determined as the object data.

[0381] As one possible implementation, the acquisition unit 1302 is specifically used for:

[0382] If the query request for the target application includes a target keyword, then the data processing result corresponding to the target keyword is searched from the multiple data processing results;

[0383] If the query request for the target application includes a target topic activity, then the data processing result corresponding to the target topic activity is searched from the multiple data processing results.

[0384] As one possible implementation, the storage unit 1304 is specifically used for:

[0385] The data processing results are divided into a first result set and a second result set. The query frequency of the data processing results included in the first result set is less than or equal to a frequency threshold, and the query frequency of the data processing results included in the second result set is greater than the frequency threshold.

[0386] The first result set is stored in the first storage space, and the second result set is stored in the second storage space. The usage cost of the first storage space is less than the usage cost of the second storage space.

[0387] As one possible implementation, the analysis unit 1303 is specifically used for:

[0388] If the object data is of the stock type, then based on the multiple prompts, the first language model is instructed to perform different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts respectively;

[0389] If the object data is of an incremental type, then based on the multiple prompts, the second language model is instructed to perform different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts. The processing speed of the second language model is greater than that of the first language model.

[0390] As one possible implementation, for the first prompt message among the plurality of prompt messages, the analysis unit 1303 is specifically used for:

[0391] Based on the first prompt information, the data processing flow instructed to be performed by the large language model is as follows:

[0392] The object data is extracted to obtain multiple undetermined topics;

[0393] Feature extraction is performed on each of the multiple undetermined topics to obtain a topic vector corresponding to each undetermined topic;

[0394] Clustering the multiple topic vectors yields multiple vector clusters;

[0395] The topic vectors whose distance from the center of the target vector cluster is less than a distance threshold are determined as the target topic vectors corresponding to the target vector cluster, where the target vector cluster is one of the multiple vector clusters;

[0396] The undetermined topic corresponding to the target topic vector is determined as the topic corresponding to the target vector cluster;

[0397] By using the plurality of vector clusters as the target vector clusters, the topics corresponding to the plurality of vector clusters are obtained respectively;

[0398] Multiple of the aforementioned topics are identified as the topics of the object data.

[0399] As one possible implementation, for the second prompt message among the plurality of prompt messages, the analysis unit 1303 is specifically used for:

[0400] Get the second prompt template;

[0401] Based on the second prompt template, obtain a topic set, an emotion set, and an intent set. The topic set includes multiple topics related to the target application, the emotion set includes multiple emotion types, and the intent set includes multiple intent types.

[0402] The second prompt information is generated based on the second prompt template, the topic set, the emotion set, and the intention set;

[0403] Based on the second prompt information, the large language model is instructed to match the object data to obtain the target topic corresponding to the object data, the target sentiment type corresponding to the target topic, and the target intent type corresponding to the target topic. The target topic is one of the multiple topics included in the topic set, the target sentiment type is one of the multiple sentiment types included in the sentiment set, and the target intent type is one of the multiple intent types included in the intent set.

[0404] As one possible implementation, for the third prompt message among the multiple prompt messages, the analysis unit 1303 is specifically used for:

[0405] Get the third-party prompt template;

[0406] Based on the third prompt template, obtain the entity relationship set, the entity type set, and the parameter dimensions corresponding to each entity type. The parameter dimensions are used to describe the dimensions for evaluating the entity.

[0407] The third prompt information is generated based on the third prompt template, the entity relationship set, the entity type set, and the parameter dimension;

[0408] According to the third prompt information, the large language model is instructed to identify the object data to obtain a first entity, a second entity, a target entity relationship between the first entity and the second entity, a first entity type of the first entity, a second entity type of the second entity, an evaluation of the first entity based on the parameter dimension corresponding to the first entity type, and an evaluation of the second entity based on the parameter dimension corresponding to the second entity type. The target entity relationship is one of the multiple entity relationships included in the entity relationship set, the first entity type is one of the multiple entity types included in the entity type set, and the second entity type is one of the multiple entity types included in the entity type set.

[0409] As one possible implementation, the acquisition unit 1301 is further configured to:

[0410] A first mapping relationship is obtained based on the third prompt template. The first mapping relationship is used to describe the degree of influence of multiple negative words on the evaluation.

[0411] The analysis unit 1303 is specifically used for:

[0412] The third prompt information is generated based on the third prompt template, the first mapping relationship, the entity relationship set, the entity type set, and the parameter dimension.

[0413] As one possible implementation, for the fourth prompt message among the multiple prompt messages, the analysis unit 1303 is specifically used for:

[0414] Based on the fourth prompt, the large language model is instructed to perform semantic understanding on the object data to obtain new indicators and the corresponding statistical methods for the new indicators.

[0415] As one possible implementation, the acquisition unit 1301 is further configured to:

[0416] Get the fifth prompt message;

[0417] The analysis unit 1303 is specifically used for:

[0418] Based on the fifth prompt information, the large language model is instructed to determine the data patterns required for multiple data processing results, and a display mode is determined based on the data patterns.

[0419] The device further includes a display unit for:

[0420] Multiple data processing results are displayed based on the display mode.

[0421] This application also provides a computer device, which can be a server or a terminal device. The computer device provided in this application will be described below from a hardware implementation perspective. Figure 14 The diagram shown is a schematic of the server's structure. Figure 15 The diagram shown is a structural schematic of the terminal device.

[0422] See Figure 14This figure is a schematic diagram of a server structure provided in an embodiment of this application. The server 1400 can vary considerably due to different configurations or performance. It may include one or more processors 1422, such as a central processing unit (CPU), memory 1432, and one or more application programs 1442 or data storage media 1430 (e.g., one or more mass storage devices). The memory 1432 and storage media 1430 can be temporary or persistent storage. The program stored in the storage media 1430 may include one or more modules (not shown in the figure), each module may include a series of instruction operations on the server. Furthermore, the processor 1422 may be configured to communicate with the storage media 1430 and execute the series of instruction operations in the storage media 1430 on the server 1400.

[0423] Server 1400 may also include one or more power supplies 1426, one or more wired or wireless network interfaces 1450, one or more input / output interfaces 1458, and / or one or more operating systems 1441, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0424] The steps performed by the server in the above embodiments can be based on this Figure 14 The server structure shown.

[0425] The processor 1422 is used to perform the following steps:

[0426] Obtain multiple prompt messages and scheduled tasks for the target application, where different prompt messages are used to instruct the execution of different data processing flows;

[0427] According to the scheduled task, object data is obtained from the target application, and the object data is used to characterize object-related behavioral data generated based on the target application.

[0428] Based on the multiple prompts, the large language model performs different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts.

[0429] Store multiple data processing results;

[0430] If a query request for the target application is obtained, the data processing result corresponding to the query request is searched from the multiple stored data processing results.

[0431] Optionally, the processor 1422 may also execute method steps of any specific implementation of the data processing method in the embodiments of this application.

[0432] See Figure 15 This figure is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. The description will be based on a smartphone as an example. Figure 15 The diagram shown is a partial structural block diagram of the smartphone, which includes: a radio frequency (RF) circuit 1510, a memory 1520, an input unit 1530, a display unit 1540, a sensor 1550, an audio circuit 1560, a Wi-Fi module 1570, a processor 1580, and a power supply 1590, among other components. Those skilled in the art will understand that... Figure 15 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0433] The following is combined Figure 15 A detailed introduction to the various components of a smartphone:

[0434] The RF circuit 1510 can be used to receive and transmit signals during information transmission or calls. In particular, it receives downlink information from the base station and processes it with the processor 1580; in addition, it transmits uplink data to the base station.

[0435] The memory 1520 can be used to store software programs and modules, and the processor 1580 runs the software programs and modules stored in the memory 1520 to realize various functions and data processing of the smartphone.

[0436] Input unit 1530 can be used to receive input numeric or character information and generate key signal inputs related to user settings and function control of the smartphone. Specifically, input unit 1530 may include touch panel 1531 and other input devices 1532. Touch panel 1531, also known as a touch screen, can collect touch operations on or near the user and drive corresponding connected devices according to a pre-set program. In addition to touch panel 1531, input unit 1530 may also include other input devices 1532. Specifically, other input devices 1532 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0437] The display unit 1540 can be used to display information input by the user or information provided to the user, as well as various menus of the smartphone. The display unit 1540 may include a display panel 1541, which may optionally be configured as a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0438] Smartphones may also include at least one sensor 1550, such as a light sensor, a motion sensor, and other sensors. Other sensors that smartphones may also be equipped with, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be detailed here.

[0439] Audio circuit 1560, speaker 1561, and microphone 1562 provide an audio interface between the user and the smartphone. Audio circuit 1560 converts received audio data into electrical signals and transmits them to speaker 1561, where speaker 1561 converts them into sound signals for output. On the other hand, microphone 1562 converts collected sound signals into electrical signals, which are received by audio circuit 1560, converted into audio data, and then processed by processor 1580 before being transmitted via RF circuit 1510 to, for example, another smartphone, or the audio data can be output to memory 1520 for further processing.

[0440] The processor 1580 is the control center of the smartphone, connecting various parts of the smartphone through various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 1520, and by calling data stored in the memory 1520. Optionally, the processor 1580 may include one or more processing units.

[0441] The smartphone also includes a power supply 1590 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 1580 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0442] Although not shown, smartphones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0443] In this embodiment of the application, the memory 1520 included in the smartphone can store computer programs and transmit the computer programs to the processor.

[0444] The processor 1580 included in the smartphone can execute the data processing method provided in the above embodiments according to the instructions in the computer program.

[0445] This application also provides a computer-readable storage medium for storing a computer program that executes the data processing method provided in the above embodiments.

[0446] On the other hand, embodiments of this application provide a computer program product including a computer program, which, when run on a computer device, causes the computer device to perform the data processing methods provided in various optional implementations of the above aspects.

[0447] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium can be at least one of the following media: read-only memory (ROM), RAM, magnetic disk or optical disk, and other media that can store computer programs.

[0448] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0449] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0450] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0451] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Based on the implementation methods provided in the above aspects, this application can also be further combined to provide more implementation methods. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data processing method, characterized in that, The method includes: Obtain multiple prompt messages and scheduled tasks for the target application, where different prompt messages are used to instruct the execution of different data processing flows; According to the scheduled task, object data is obtained from the target application, and the object data is used to characterize object-related behavioral data generated based on the target application. Based on the multiple prompts, the large language model performs different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts. Store multiple data processing results; If a query request for the target application is obtained, the data processing result corresponding to the query request is searched from the multiple stored data processing results.

2. The method according to claim 1, characterized in that, The step of obtaining object data from the target application according to the scheduled task includes: Obtain source object data from the target application; The source object data is stored in different coarse data storage tables, and the different coarse data storage tables are used to store source object data of different categories; Based on multiple sub-timed tasks and multiple coarse data storage tables, multiple fine data storage tables are created. Each fine data storage table is obtained based on one of the sub-timed tasks and at least one of the coarse data storage tables. The timed tasks include the multiple sub-timed tasks. The data stored in the detailed data storage table is determined as the object data.

3. The method according to claim 2, characterized in that, If the source object data includes a first category, a second category, a third category, and a fourth category, where the first category is articles published for the target application, the second category is object behaviors performed for the target application, the third category is comments published for the target application, and the fourth category is topic activities published through the target application, then the creation of multiple fine data storage tables based on multiple sub-timed tasks and multiple coarse data storage tables includes: If the sub-timed task is to periodically retrieve article detail data, then a first fine data storage table is created based on the first coarse data storage table and the second coarse data storage table. The first coarse data storage table is used to store source object data belonging to the first category, the second coarse data storage table is used to store source object data belonging to the second category, and the first fine data storage table is used to store article detail data related to each of the articles. If the sub-timed task is to periodically obtain evaluations based on articles, then a second fine data storage table is created based on the first coarse data storage table and the third coarse data storage table. The third coarse data storage table is used to store source object data belonging to the third category, and the second fine data storage table is used to store comments published for each of the articles. If the sub-timed task is to periodically obtain the theme activity details data, then a third fine data storage table is created based on the second coarse data storage table and the fourth coarse data storage table. The fourth coarse data storage table is used to store source object data belonging to the fourth category, and the third fine data storage table is used to store activity details data related to each of the theme activities. If the sub-timed task is to periodically obtain evaluations based on a theme activity, then a fourth fine data storage table is created based on the third coarse data storage table and the fourth coarse data storage table. The fourth fine data storage table is used to store comments related to each of the theme activities.

4. The method according to claim 3, characterized in that, The source object data also includes a fifth category, which consists of keywords generated based on the target application. Therefore, storing the source object data in different coarse data storage tables includes: The source object data belonging to the fifth category is stored in the fifth coarse data storage table; Search the target application for keywords stored in the fifth coarse data storage table, and create a sixth coarse data storage table, which is used to store associated data related to the keywords; The step of determining the data stored in the detailed data storage table as the object data includes: The data stored in the fine data storage table and the data stored in the sixth coarse data storage table are determined as the object data.

5. The method according to claim 4, characterized in that, If a query request for the target application is obtained, then searching for the data processing result corresponding to the query request from among the multiple data processing results includes: If the query request for the target application includes a target keyword, then the data processing result corresponding to the target keyword is searched from the multiple data processing results; If the query request for the target application includes a target topic activity, then the data processing result corresponding to the target topic activity is searched from the multiple data processing results.

6. The method according to claim 1, characterized in that, The storage of multiple data processing results includes: The data processing results are divided into a first result set and a second result set. The query frequency of the data processing results included in the first result set is less than or equal to a frequency threshold, and the query frequency of the data processing results included in the second result set is greater than the frequency threshold. The first result set is stored in the first storage space, and the second result set is stored in the second storage space. The usage cost of the first storage space is less than the usage cost of the second storage space.

7. The method according to claim 1, characterized in that, The method involves instructing the large language model to perform different data processing procedures on the object data based on the multiple prompts, thereby obtaining the data processing results corresponding to the multiple prompts, including: If the object data is of the stock type, then based on the multiple prompts, the first language model is instructed to perform different data processing procedures on the object data to obtain the data processing results corresponding to the multiple prompts respectively; If the object data is of an incremental type, the second language model is instructed to perform different data processing procedures on the object data based on the multiple prompts, so as to obtain the data processing results corresponding to the multiple prompts respectively. The processing speed of the second language model is greater than that of the first language model.

8. The method according to claim 1, characterized in that, For the first prompt among the plurality of prompts, the step of instructing the large language model based on the plurality of prompts to perform different data processing flows on the object data to obtain the data processing results corresponding to the plurality of prompts respectively includes: Based on the first prompt information, the data processing flow instructed to be performed by the large language model is as follows: The object data is extracted to obtain multiple undetermined topics; Feature extraction is performed on each of the multiple undetermined topics to obtain a topic vector corresponding to each undetermined topic; Clustering the multiple topic vectors yields multiple vector clusters; The topic vectors whose distance from the center of the target vector cluster is less than a distance threshold are determined as the target topic vectors corresponding to the target vector cluster, where the target vector cluster is one of the multiple vector clusters; The undetermined topic corresponding to the target topic vector is determined as the topic corresponding to the target vector cluster; By using the plurality of vector clusters as the target vector clusters, the topics corresponding to the plurality of vector clusters are obtained respectively; Multiple of the aforementioned topics are identified as the topics of the object data.

9. The method according to claim 1, characterized in that, Regarding the second prompt among the plurality of prompts, the step of instructing the large language model based on the plurality of prompts to perform different data processing flows on the object data to obtain the data processing results corresponding to the plurality of prompts respectively includes: Get the second prompt template; Based on the second prompt template, obtain a topic set, an emotion set, and an intent set. The topic set includes multiple topics related to the target application, the emotion set includes multiple emotion types, and the intent set includes multiple intent types. The second prompt information is generated based on the second prompt template, the topic set, the emotion set, and the intention set; Based on the second prompt information, the large language model is instructed to match the object data to obtain the target topic corresponding to the object data, the target sentiment type corresponding to the target topic, and the target intent type corresponding to the target topic. The target topic is one of the multiple topics included in the topic set, the target sentiment type is one of the multiple sentiment types included in the sentiment set, and the target intent type is one of the multiple intent types included in the intent set.

10. The method according to claim 1, characterized in that, Regarding the third prompt among the plurality of prompts, the step of instructing the large language model based on the plurality of prompts to perform different data processing flows on the object data to obtain the data processing results corresponding to the plurality of prompts respectively includes: Get the third-party prompt template; Based on the third prompt template, obtain the entity relationship set, the entity type set, and the parameter dimensions corresponding to each entity type. The parameter dimensions are used to describe the dimensions for evaluating the entity. The third prompt information is generated based on the third prompt template, the entity relationship set, the entity type set, and the parameter dimension; According to the third prompt information, the large language model is instructed to identify the object data to obtain a first entity, a second entity, a target entity relationship between the first entity and the second entity, a first entity type of the first entity, a second entity type of the second entity, an evaluation of the first entity based on the parameter dimension corresponding to the first entity type, and an evaluation of the second entity based on the parameter dimension corresponding to the second entity type. The target entity relationship is one of the multiple entity relationships included in the entity relationship set, the first entity type is one of the multiple entity types included in the entity type set, and the second entity type is one of the multiple entity types included in the entity type set.

11. The method according to claim 10, characterized in that, The method further includes: A first mapping relationship is obtained based on the third prompt template. The first mapping relationship is used to describe the degree of influence of multiple negative words on the evaluation. The step of generating third prompt information based on the third prompt template, the entity relationship set, the entity type set, and the parameter dimension includes: The third prompt information is generated based on the third prompt template, the first mapping relationship, the entity relationship set, the entity type set, and the parameter dimension.

12. The method according to claim 1, characterized in that, Regarding the fourth prompt among the plurality of prompts, the step of instructing the large language model based on the plurality of prompts to perform different data processing flows on the object data to obtain the data processing results corresponding to the plurality of prompts respectively includes: Based on the fourth prompt, the large language model is instructed to perform semantic understanding on the object data to obtain new indicators and the corresponding statistical methods for the new indicators.

13. The method according to claim 1, characterized in that, The method further includes: Get the fifth prompt message; Based on the fifth prompt information, the large language model is instructed to determine the data patterns required for multiple data processing results, and a display mode is determined based on the data patterns. Multiple data processing results are displayed based on the display mode.

14. A data processing apparatus, characterized in that, The device includes: an acquisition unit, an analysis unit, a storage unit, and a query unit; The acquisition unit is used to acquire multiple prompt messages and timed tasks for the target application, and different prompt messages are used to instruct the execution of different data processing flows; The acquisition unit is used to acquire object data from the target application according to the scheduled task, and the object data is used to characterize object-related behavioral data generated based on the target application. The analysis unit is used to instruct the large language model to perform different data processing procedures on the object data based on the multiple prompts, so as to obtain the data processing results corresponding to the multiple prompts respectively. The storage unit is used to store multiple data processing results; The query unit is configured to, if a query request for the target application is obtained, search for the data processing result corresponding to the query request from a plurality of stored data processing results.

15. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs and to transfer the computer programs to the processor; The processor is configured to perform the method according to any one of claims 1-13 according to the computer program.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method according to any one of claims 1-13.

17. A computer program product comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method described in any one of claims 1-13.

Images