Report summary text generation method and device and nonvolatile storage medium

Abstract

The invention discloses a report summary text generation method and device and a nonvolatile storage medium. The method comprises the following steps: determining a request type of a report summary text request; when the request type is a non-calculation request and the number of tokens of the target report exceeds a preset number of tokens, performing compressed sampling processing on data in the target report, and calling a semantic model to generate a first report summary text according to a compressed sampling result and the report summary text request; under the condition that the request type is a calculation request, calling a semantic model to generate an executable code according to the report summary text request and the execution code cue word, and calling the semantic model to execute the executable code to process data in the target report to obtain a second report summary text. According to the report summarization method and device, the technical problem that the accuracy and efficiency are low when reports are summarized by adopting a semantic model due to the fact that the same large model processing mode is adopted for different types of summarization tasks in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a report summary text generation method and device and a nonvolatile storage medium. BACKGROUND

[0002] In related technologies, the same processing method is usually used for different types of summary tasks when a large model is used to summarize a report. However, the correct rate of certain specific types of tasks is low under the general processing method, or the general processing method cannot be used to process the report due to the length of the report. This results in low correct rate and efficiency when a large model is used to summarize a report in related technologies.

[0003] At present, no effective solution has been proposed to solve the above problems. SUMMARY

[0004] The embodiments of the present application provide a report summary text generation method, device and nonvolatile storage medium, to at least solve the technical problem of low correct rate and efficiency when a semantic model is used to summarize a report due to the same large model processing method for different types of summary tasks in related technologies.

[0005] According to an aspect of an embodiment of the present application, a report summary text generation method is provided, including: determining a request type of a report summary text request, wherein the request type includes a calculation request and a non-calculation request; in the case of the request type being a non-calculation request, determining a target report indicated by the report summary text request, and in the case of the number of tokens of the target report exceeding a preset token number, performing compression sampling processing on the data in the target report, and calling a semantic model to generate a first report summary text according to the compression sampling result and the report summary text request; in the case of the request type being a calculation request, calling the semantic model to generate executable code according to the report summary text request and an execution code prompt word, and calling the semantic model to execute the executable code to process the data in the target report to obtain a second report summary text.

[0006] Optionally, the compression sampling processing on the data in the target report includes: determining to-be-compressed data in the target report; performing sampling processing on the to-be-compressed data to obtain a plurality of sampling data points; determining an importance evaluation value of the sampling data points, and screening the sampling data points according to the importance evaluation value to obtain a compression sampling result.

[0007] Optionally, the determining the importance evaluation value of the sampling data point comprises: sorting the sampling data point according to the position information of the sampling data point in the data to be compressed, to obtain a sequence of the sampling data point; determining the average value of the adjacent sampling data point of each sampling data point in the sequence of the sampling data point; determining the difference value between the sampling data point and the average value; and determining the importance evaluation value of the sampling data point according to the difference value, wherein the greater the difference value is, the greater the importance evaluation value is.

[0008] Optionally, the calling the semantic model to execute the executable code to process the data in the target report comprises: in the case that the executable code fails to execute, obtaining error information of the executable code; updating the execution code prompt word according to the error information; calling the semantic model to update the executable code according to the updated execution code prompt word; and calling the semantic model to execute the updated executable code to process the data in the target report.

[0009] Optionally, before the calling the semantic model to generate the executable code according to the report summary text request and the execution code prompt word, the method further comprises: performing sampling processing on the data in the target report to obtain sampling data; and inputting the sampling data into the semantic model, so that the semantic model learns the data format of the data in the target report.

[0010] Optionally, the method for generating the report summary text further comprises: creating a report template, and determining an application program interface corresponding to the report template; binding a target component in the report template to a corresponding data registration center interface through the application program interface, and obtaining the data required by the target component through the data registration center; and filling the data required by the component into the report template to obtain the target report.

[0011] Optionally, the creating the report template comprises: in response to an operation instruction of a target object, determining a target component from a set of alternative components, and placing the target component in a position indicated by the operation instruction; and in response to a debugging instruction of the target object, adjusting a component attribute of the target component.

[0012] According to another aspect of the embodiments of the present application, a report summary text generation apparatus is further provided, comprising: a first processing module configured to determine a request type of a report summary text request, wherein the request type comprises a calculation request and a non-calculation request; a second processing module configured to, in the case that the request type is the non-calculation request, determine a target report indicated by the report summary text request, and in the case that a token number of the target report exceeds a preset token number, perform compression sampling processing on data in the target report, and call a semantic model to generate a first report summary text according to a compression sampling result and the report summary text request; and a third processing module configured to, in the case that the request type is the calculation request, call the semantic model to generate an executable code according to a report summary text request and an execution code prompt word, and call the semantic model to execute the executable code to process the data in the target report, to obtain a second report summary text.

[0013] According to another aspect of the embodiments of the present application, a nonvolatile storage medium is also provided, which stores a program, and when the program is executed, the device in which the nonvolatile storage medium is located is controlled to perform the report summary text generation method.

[0014] According to another aspect of the embodiments of the present application, an electronic device is also provided, which includes a memory and a processor, and the processor is configured to execute a program stored in the memory, and when the program is executed, the report summary text generation method is performed.

[0015] According to another aspect of the embodiments of the present application, a computer program product is also provided, which includes a computer program, and when the computer program is executed by a processor, the report summary text generation method is implemented.

[0016] In the embodiments of the present application, the request type of the report summary text request is determined, wherein the request type includes a calculation request and a non-calculation request; in the case of the request type being a non-calculation request, the target report indicated by the report summary text request is determined, and in the case of the token number of the target report exceeding a preset token number, the data in the target report is compressed and sampled, and a semantic model is called to generate a first report summary text according to the compressed and sampled result and the report summary text request; in the case of the request type being a calculation request, an executable code is generated by calling the semantic model according to the report summary text request and an execution code prompt word, and the data in the target report is processed by calling the semantic model to execute the executable code to obtain a second report summary text. In this way, by determining the request type and using different processing methods for different request types, the semantic model can correctly and efficiently process various summary requests, thereby achieving the technical effect of efficiently obtaining an accurate report summary text, and further solving the technical problem that the accuracy and efficiency are both low when the semantic model is used to summarize a report due to the same large model processing method used by related technologies for different types of summary tasks. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the present application, form a part of the present application and illustrate the illustrative embodiments of the present application and the explanation of the present application, and do not constitute improper limitations on the present application. In the drawings:

[0018] Figure 1 is a structural schematic diagram of a computer terminal (mobile device) provided according to an embodiment of the present application;

[0019] Figure 2 is a flowchart of a report summary text generation method provided according to an embodiment of the present application;

[0020] Figure 3is a flowchart of a report summary text generation process according to an embodiment of the present application;

[0021] Figure 4 is an architecture diagram of a report generation system according to an embodiment of the present application;

[0022] Figure 5 is a structural diagram of a report summary text generation device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0024] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0026] AI (Artificial Intelligence, Artificial Intelligence): refers to the ability of a computer system to simulate, extend or expand human intelligence, including learning, reasoning, perception, decision making, etc. Simply put, AI can enable machines to solve problems, adapt to environments or complete complex tasks like humans.

[0027] Large language model (Large Language Model, Large Language Model): a deep learning model specially designed to process and generate human language, with a large parameter size, trained on a large amount of data, capable of capturing complex data patterns, and therefore has strong language understanding and generation capabilities.

[0028] Prompt: Text entered by the user to guide the model in generating a response.

[0029] Data sampling: Data sampling is a statistical technique used to select a subset of individuals, items, or observations from a large population.

[0030] vue-draggable: A front-end drag-and-drop component. It provides highly customizable drag-and-drop sorting capabilities, suitable for creating interactive user interface elements such as sortable lists and grid layouts. It is suitable for application scenarios that require dynamically adjustable layouts, such as dashboards and editor interfaces.

[0031] ECharts: A JavaScript-based visualization library that can be used to create interactive charts and data visualizations.

[0032] Canvas: An element in Hypertext Markup Language that provides an area for drawing graphics using JavaScript.

[0033] html2canvas: A JavaScript library that allows developers to convert HTML elements in a webpage into Canvas elements, which can then be exported as images. It is suitable for scenarios that require taking screenshots, generating preview images, or printing webpage content.

[0034] CKEditor: A rich text editor used in various web applications to meet the needs of creating and editing complex content.

[0035] OSS (Object Storage Service): A cloud storage solution that provides high availability and durability. It allows users to upload, download, and manage unstructured data, such as images, videos, and backup files, over a network.

[0036] In the cloud network operations field, a large amount of data is generated daily. Reports, as a key tool for managing data, bear the responsibility of analyzing historical data, extracting valuable information, and providing decision support for operations personnel and managers through intuitive graphic and textual presentations. Reports are not only an important carrier for displaying and recording work results, but also play important roles in process control, data storage, and information exchange. However, automated report generation / analysis tools in related technologies have the following problems:

[0037] (1) Complex indicator data and error-prone manual report generation: In the report generation process of related technologies, various types of data are scattered in different systems or devices in different formats and structures, making information extraction and summarization complex. The task of batch generating similar reports is tedious and inefficient, and frequent manual copying and pasting is prone to errors, reducing reliability. Once a parameter changes, the entire report needs to be rewritten, which is inefficient and prone to errors.

[0038] (2) Manual report analysis is time-consuming and labor-intensive, and relying solely on large models for engineering analysis is prone to errors: When analyzing data in reports, senior data analysts typically need to conduct in-depth analysis and summarization based on experience, resulting in low efficiency and high costs. However, using large models to provide AI-based summarization and analysis of data charts and text paragraphs within reports presents several limitations. First, the large model's character limit prevents it from summarizing and analyzing longer reports. Second, for computational analysis problems, the large model's insufficient computational power fails to meet the demand for accurate analysis. Furthermore, large models in related technologies have low accuracy in summarizing computational analysis problems, leading to inaccuracies in calculating the total number and average of tables, and inability to accurately statistically analyze trend chart values ​​exceeding a certain threshold. Additionally, trend chart components in reports are generally dense with excessive redundant information; large language models have strict length limitations when processing input data. Excessively long input data may exceed the model's context window (token limit) and could cause catastrophic forgetting, resulting in the loss of key information or a decline in model performance. These technical limitations make directly inputting the raw trend chart data into large models for summary analysis impractical.

[0039] To address the aforementioned issues, this application provides relevant solutions, which are detailed below.

[0040] According to an embodiment of this application, a method embodiment for generating report summary text is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0041] The methods and embodiments provided in this application can be executed on mobile terminals, computer terminals, or similar computing devices. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method to generate report summary text is shown. Figure 1As shown, the computer terminal 10 (or mobile device 10) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0042] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0043] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the report summary text generation method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned report summary text generation method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0044] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0045] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0046] Under the aforementioned operating environment, this application provides a method for generating report summary text, such as... Figure 2 As shown, the method includes the following steps:

[0047] Step S202: Determine the request type for the report summary text request, wherein the request type includes calculation requests and non-computation requests;

[0048] In some embodiments of this application, the report summary prompt template can be published as an API interface, with the input parameters being the variable parts of the report summary prompt template content. The published API interfaces are divided into three types: intent recognition API, code analysis API, and report summary API. Furthermore, the intent recognition API, code analysis API, and report summary API can all utilize existing semantic models (i.e., large language models) as their base models. Additionally, since the intent recognition API, code analysis API, and report summary API require different types of capabilities, they can choose different semantic models as their base models. For example, the intent recognition API and report summary API can choose a large language model with good instruction compliance capabilities as their base model, while the code analysis API can use a large language model with strong divergent thinking and a certain level of reasoning ability as its base model.

[0049] In some embodiments of this application, the types of prompt words in the prompt word template may include abilities and roles, background information, instructions, thought chains, and examples, which assist the model in reasoning and thinking so as to better analyze the report.

[0050] In some embodiments of this application, the trend charts and tables in the report are prone to long and inaccurate calculations. Therefore, when summary requests involve trend charts and tables, an intent recognition API can be used to classify user questions into calculation-based and non-calculation-based requests. For example, "Please help me analyze how many regions have a maintenance work order completion time rate below 99%" is a calculation-based request, while "Please help me summarize the changes in ITV user numbers over different time periods" is a non-calculation-based request.

[0051] Step S204: If the request type is a non-computation request, determine the target report indicated by the report summary text request, and if the number of tokens in the target report exceeds the preset number of tokens, perform compression sampling on the data in the target report, and call the semantic model to generate the first report summary text based on the compression sampling result and the report summary text request.

[0052] The semantic model described above can be any large language model.

[0053] In the technical solution provided in step S204, the step of compressing and sampling the data in the target report includes: determining the data to be compressed in the target report; sampling the data to be compressed to obtain multiple sampling data points; determining the importance evaluation value of the sampling data points, and filtering the sampling data points according to the importance evaluation value to obtain the compressed sampling result.

[0054] In some embodiments of this application, the step of determining the importance evaluation value of a sampled data point includes: sorting the sampled data points according to their position information in the data to be compressed to obtain a sampled data point sequence; determining the average value of adjacent sampled data points in the sampled data point sequence; determining the difference between a sampled data point and the average value; and determining the importance evaluation value of the sampled data point based on the difference, wherein the larger the difference, the greater the importance evaluation value. The aforementioned position information may include the horizontal and vertical coordinates of each sampled data point in a coordinate system. During sorting, the data points can be sorted according to the order of their horizontal coordinates.

[0055] In some embodiments of this application, for non-computational requests, when the length of the target report content (i.e., the number of tokens) input into the semantic model exceeds a preset number of tokens (e.g., 30,000), taking the trend chart component as an example, it is necessary to perform compression, sampling, and data preprocessing. For example, for the compressed time point length "2024 / 09 / 23 04:30:20", the year is only retained once, and the year is removed at other times to shorten the time point length, thereby reducing the number of tokens input into the semantic model.

[0056] In some embodiments of this application, to further reduce tokens, the trend of the data can be preserved while irrelevant points are ignored. For example, after obtaining each sampled data point, these sampled data points can be sorted in ascending order according to the magnitude of their corresponding x-coordinates. Then, the difference between the average value of a point and the average values ​​of its two neighbors can be calculated. For example, for three sampled data points 1, 2, and 3, the average value is (1+3) / 2 = 2, and the difference is 2-2 = 0. The smaller the difference, the lower the importance. The average importance of each curve is calculated, and then all points are sorted. Important points are then retained based on importance scores. For example, for a sequence [1,2,3,4,5,6,7,8,9,10], a sampling method based on importance scores can remove less important data points in the middle, retaining only [1,10], without affecting the overall trend of the sequence. After data preprocessing, the number of tokens required to input a large model for non-computational requests, such as "summarizing the changes in the number of ITV users over different time periods," is greatly reduced.

[0057] In some embodiments of this application, a method is provided as follows: Figure 3 The report summary text generation process is shown below. It is important to note that... Figure 3 The specific figures mentioned are for illustrative purposes only and do not represent a limitation on the relevant schemes in this application. Figure 3 As can be seen, the first step is to use AI (large model) to divide the data to be analyzed into tabular data, trend charts, and other components, such as pie charts, bar charts, percentage charts, and completion rate charts. For tabular data and trend charts, further analysis can be conducted to determine whether the task type is computational or non-computational. For other components, it directly checks whether the character count of these components exceeds a preset character limit. If it does, a message is displayed indicating that the current data exceeds the large model's character limit and analysis is not supported. If it does not exceed the limit, the general large model prompts are used to perform the summary task.

[0058] In some embodiments of this application, for summary tasks involving tables, if the preset character limit is exceeded, a message can be directly displayed indicating that the current data exceeds the character limit of the large model and analysis is not supported. For trend charts, the data can be compressed and sampled before analysis continues. If compression sampling fails, a message can be displayed indicating that the current data exceeds the character limit of the large model and analysis is not supported.

[0059] For computational tasks, if the number of calculation failures is too high, a summary text can be generated directly from the calculation results. Additionally, if the calculated data is too long, the response can be truncated from the beginning of the result. For example, if limited to 1000 tokens, the calculation results for the first 1000 tokens can be truncated, and a message will be displayed indicating that the calculation content is too long and affects the report display, therefore it has been truncated.

[0060] In some embodiments of this application, since there may be different analysis tasks in the same report, the summary texts of these different analysis tasks can also be uniformly summarized and polished.

[0061] Step S206: If the request type is a calculation request, the semantic model is invoked to generate executable code based on the report summary text request and the execution code prompt, and the semantic model is invoked to execute the executable code to process the data in the target report and obtain the second report summary text.

[0062] In the technical solution provided in step S206, the step of calling the semantic model to execute executable code to process the data in the target report includes: obtaining the error information of the executable code in the case of execution failure; updating the execution code prompt word according to the error information; calling the semantic model to update the executable code according to the updated execution code prompt word; and calling the semantic model to execute the updated executable code to process the data in the target report.

[0063] In some embodiments of this application, before calling the semantic model to generate executable code based on the report summary text request and execution code prompts, the method further includes: sampling the data in the target report to obtain sampled data; and inputting the sampled data into the semantic model so that the semantic model can learn the data format of the data in the target report.

[0064] In some embodiments of this application, taking data to be analyzed and processed, including tabular data or trend chart component data, a portion of the data can be sampled first to allow the semantic model to understand the format of the data input into the model. Then, user questions and prompts for executing code are input into the large model to generate executable code. This executable code is then used to execute the full dataset and output the calculation results, addressing the issue of low accuracy when directly inputting data into the large model for computational requests. After obtaining the calculation results from the code, the large model generates a second report summary text based on the calculation results, thus providing a summary that the user can understand.

[0065] In some embodiments of this application, an error retry mechanism is also provided. If code execution fails, error information can be appended to the code analysis API prompts, allowing the large model to obtain the error information and thus guide it to update its code, improving the accuracy of code generation. Furthermore, code generation has a fallback mechanism: when the number of error retries exceeds a preset number, a summary text can be generated based on the calculation result of the last execution, noting that the calculation result may contain errors, thus preventing errors in the report summary generation.

[0066] As an alternative implementation, the analysis results of error information from a large model can be written into the code generation prompts to increase the probability of generating correct code.

[0067] In some embodiments of this application, the process of creating a target report includes the following steps: creating a report template and determining the application programming interface (API) corresponding to the report template; binding the target component in the report template to the corresponding data registry interface through the API, and obtaining the data required by the target component through the data registry; filling the data required by the component into the report template to obtain the target report.

[0068] In some embodiments of this application, the step of creating a report template includes: responding to an operation instruction from the target object, determining a target component from a set of candidate components, and placing the target component at the location indicated by the operation instruction; responding to a debugging instruction from the target object, adjusting the component properties of the target component. The target object may be a user or a terminal device controlled by the user, etc.

[0069] In some embodiments of this application, low-code front-end technology can be used to quickly build report templates through a component repository and grid orchestration. The target user can construct the report template by dragging and dropping components, configuring properties, and setting logic using a graphical user interface. Components are selected from the left component selection area, dragged into the middle preview area, and placed into appropriate container blocks. Component properties are then debugged in the right property editing area. This process is repeated until the template is built. After the template is built, when generating the report, the front-end requests back-end data based on the template configuration and displays the results on the template, ultimately outputting it in a preset format (e.g., base64).

[0070] In some embodiments of this application, the aforementioned set of alternative components can be a component repository. This component repository, specifically a report-style component repository, enables the graphical display of standard format data, greatly enriching the visualization methods of reports. The repository contains various types of components commonly used in reports, such as pie charts to intuitively show the proportion of data, trend charts to clearly present data trends, bar charts for easy comparison of different data, tables suitable for displaying detailed data lists, and rich text components for adding explanatory text. This diverse component configuration meets the needs of report content display in different business scenarios.

[0071] Furthermore, the graphical interface in this embodiment employs a grid arrangement method, enabling flexible arrangement of different components. When generating a report template, a responsive grid layout can be created simply by dragging and dropping, and the required components can be configured within the grid. Additionally, the width, height, position, and other attributes of each grid item can be configured. This grid arrangement method not only achieves reasonable layout of report content but also enhances the report's aesthetics, improving its overall quality and readability.

[0072] In some embodiments of this application, considering that the data sources and formats of reports may vary when the report template is generated, it is necessary to centrally extract and process the data required for the report from different systems or devices, and parse it into the data required for the report. Therefore, embodiments of this application also construct a backend data registry center to acquire and parse the data, where the data registry center is also a low-code platform. The data registry center can extract and process datasets scattered across different systems or devices, and encapsulate the ability to acquire report data into atoms using Python scripts, where the output parameters of the atoms correspond to the output parameter formats of different report components. The Python scripts can acquire data from different sources such as databases, Elasticsearch, and external interfaces, and perform data transformation. The data registry center also undertakes the data foundation work for report generation, and can encapsulate the ability to process data into API interfaces. The published APIs are bound to different types of components, and the input parameters of the atomic APIs will be used as part of the input parameters of the report template. When the report is generated subsequently, the data that meets the report requirements can be obtained by adjusting the input parameters of the API interface.

[0073] In some embodiments of this application, taking network maintenance as an example, operations and maintenance personnel need to understand the network's operational status in a timely manner and quickly detect anomalies, such as sudden changes in the number of online users, alarm counts exceeding the average, and significant imbalances in inbound and outbound traffic. Therefore, operations and maintenance personnel can use the methods provided in the embodiments of this application to configure relevant data analysis requirements in the report. Then, the analysis requirements can be intentionally identified to determine the task type. For example, if the analysis requirement is a calculation-type request such as "Please help me analyze whether today's traffic fluctuates significantly compared to yesterday," then the report summary text generation method corresponding to the calculation-type request can be executed to obtain the analysis results, which can then be embedded into the report and periodically pushed to the operations and maintenance personnel.

[0074] In some embodiments of this application, an automatic report generation process is also provided, including the following steps:

[0075] The first step is to aggregate the raw data into the data registry. The data registry supports writing program scripts to obtain raw data from databases, external APIs, etc., and to convert the raw data into the data format of the component repository by writing program code.

[0076] The second step is to create a report template, which is composed of components registered in the data registry. Set the name of the report template, select components from the left-hand component selection area, drag them into the middle preview area, and place them in appropriate container blocks. Then, adjust the component properties in the right-hand property editing area. Repeat this process until the template is complete. The report template supports adjustments to the size, position, and background style of components. Furthermore, report summary capabilities can also be embedded as a special type of component into the report template.

[0077] The third step is to publish the report template as an API interface after it is created. This API interface can concurrently request the data registry interface bound to all components in the template, thereby obtaining all the data required for the report at once and generating a report screenshot. The screenshot is then converted into a string using Base64 encoding and output. The input parameters of the report template API include the input parameters of the atomic API bound to each element in the template, and these parameters are used to request the atomic API, obtain data, and render the report.

[0078] The fourth step is to use the scheduled task module to periodically call the report template API to generate reports, and then send the generated reports to the corresponding business personnel in a specified manner.

[0079] The fifth step is to categorize the sent reports according to the template, store them via OSS backup, and provide them to business personnel for review.

[0080] In some embodiments of this application, a configuration process for a report summary text generation system is also provided, including the following steps:

[0081] The first step is to deploy a general-purpose large language model to understand and process natural language input and output formatted text content according to the input requirements. You can choose to deploy various open-source large language models on the market for private deployment, or you can directly use the large language model interface.

[0082] The second step is to deploy the data registry module as the data base for automatic report generation. This module is developed based on the Flask framework and can support the execution of internal Python scripts. Through Python scripts, real-time or historical data can be obtained from different systems and the data can be processed.

[0083] The third step is to deploy the component repository module. This module is based on a deeply customized version of ECharts components, including various styles for bar charts, pie charts, and trend charts; optimized display of X and Y axis coordinates; optimized display of maximum and minimum values ​​for trend charts; and adjustable text such as icon titles. It is used to transform data into visualization components.

[0084] The fourth step is to deploy the report template editor module, which can be developed based on vue-draggable + html2canvas, and is used to arrange different components and report summaries into report templates;

[0085] The fifth step is to deploy the prompt module of the large language model, which is used to receive raw component data and analysis and summary rules, and output a report summary.

[0086] Step 6: Deploy the timer module to generate reports periodically;

[0087] Step 7: Connect and interact with the report sending interface through the open interface of the enterprise-level chat tool or the enterprise email, mainly to enable the push of reports.

[0088] According to an embodiment of this application, a method such as... is also provided. Figure 4 The report generation system shown includes a data source, a data registry center of a backend low-code platform, a frontend low-code platform, and an output terminal. The frontend low-code platform includes a component repository, where report templates can be obtained by selecting components from the repository. Data processed by the data registry center can then be written into the report template via API, and the report template is periodically pushed to the output terminal.

[0089] By employing a method that determines the request type of the report summary text request (including computational and non-computational requests), and in the case of a non-computational request, identifies the target report indicated by the report summary text request, and if the number of tokens in the target report exceeds a preset number, performs compressed sampling on the data in the target report, and calls a semantic model to generate a first report summary text based on the compressed sampling result and the report summary text request, and in the case of a computational request, calls a semantic model to generate executable code based on the report summary text request and execution code prompts, and calls a semantic model to execute the executable code to process the data in the target report to obtain a second report summary text, this approach achieves the goal of enabling the semantic model to correctly and efficiently process various summary requests by determining the request type and using different processing methods for different request types. This achieves the technical effect of efficiently obtaining accurate report summary texts, and solves the technical problem of low accuracy and efficiency when using a semantic model to summarize reports, which is caused by related technologies using the same large model processing method for different types of summary tasks.

[0090] This application provides a report summary text generation device. Figure 5 This is a schematic diagram of the device. From Figure 5 As can be seen from the diagram, the device includes: a first processing module 50, used to determine the request type of the report summary text request, wherein the request type includes a calculation request and a non-computation request; a second processing module 52, used to determine the target report indicated by the report summary text request when the request type is a non-computation request, and to perform compressed sampling processing on the data in the target report when the number of tokens in the target report exceeds a preset number of tokens, and to call a semantic model to generate a first report summary text based on the compressed sampling result and the report summary text request; and a third processing module 54, used to call a semantic model to generate executable code based on the report summary text request and the execution code prompt when the request type is a calculation request, and to call a semantic model to execute the executable code to process the data in the target report to obtain a second report summary text.

[0091] In some embodiments of this application, the second processing module 52 performs compression sampling processing on the data in the target report, including: determining the data to be compressed in the target report; performing sampling processing on the data to be compressed to obtain multiple sampling data points; determining the importance evaluation value of the sampling data points, and filtering the sampling data points according to the importance evaluation value to obtain the compression sampling result.

[0092] In some embodiments of this application, the step of the second processing module 52 in determining the importance evaluation value of the sampled data points includes: sorting the sampled data points according to their position information in the data to be compressed to obtain a sequence of sampled data points; determining the average value of adjacent sampled data points in the sequence of sampled data points; determining the difference between the sampled data points and the average value; and determining the importance evaluation value of the sampled data points based on the difference, wherein the larger the difference, the greater the importance evaluation value.

[0093] In some embodiments of this application, the step of the third processing module 54 calling the semantic model to execute executable code to process the data in the target report includes: obtaining error information of the executable code in the event that the executable code fails to execute; updating the execution code prompt word according to the error information; calling the semantic model to update the executable code according to the updated execution code prompt word; and calling the semantic model to execute the updated executable code to process the data in the target report.

[0094] In some embodiments of this application, before the semantic model generates executable code based on the report summary text request and execution code prompts, the third processing module 54 is further configured to: sample the data in the target report to obtain sampled data; and input the sampled data into the semantic model so that the semantic model can learn the data format of the data in the target report.

[0095] In some embodiments of this application, the report summary text device is further configured to: create a report template and determine the application programming interface (API) corresponding to the report template; bind the target component in the report template to the corresponding data registry interface through the API, and obtain the data required by the target component through the data registry; fill the data required by the component into the report template to obtain the target report.

[0096] In some embodiments of this application, the steps of the report summary text device in creating a report template include: responding to an operation instruction from a target object, determining a target component from a set of candidate components, and placing the target component at the location indicated by the operation instruction; and responding to a debugging instruction from the target object, adjusting the component properties of the target component.

[0097] It should be noted that the modules in the above-mentioned report summary text generation device can be program modules (e.g., a set of program instructions to implement a certain function) or hardware modules. For the latter, they can be represented in the following forms, but are not limited to these: each of the above modules is represented by a processor, or the functions of each of the above modules are implemented by a processor.

[0098] According to an embodiment of this application, a non-volatile storage medium is also provided, which stores a program. During program execution, the device containing the non-volatile storage medium executes the following report summary text generation method: determining the request type of the report summary text request, wherein the request type includes a computational request and a non-computational request; if the request type is a non-computational request, determining the target report indicated by the report summary text request, and if the number of tokens in the target report exceeds a preset number of tokens, performing compression sampling on the data in the target report, and calling a semantic model to generate a first report summary text based on the compression sampling result and the report summary text request; if the request type is a computational request, calling a semantic model to generate executable code based on the report summary text request and execution code prompts, and calling a semantic model to execute the executable code to process the data in the target report, thereby obtaining a second report summary text.

[0099] According to an embodiment of this application, an electronic device is also provided, including a memory and a processor. The processor is used to run a program stored in the memory, wherein the program executes the following report summary text generation method during runtime: determining the request type of the report summary text request, wherein the request type includes a computational request and a non-computational request; when the request type is a non-computational request, determining the target report indicated by the report summary text request, and when the number of tokens in the target report exceeds a preset number of tokens, performing compression sampling processing on the data in the target report, and calling a semantic model to generate a first report summary text based on the compression sampling result and the report summary text request; when the request type is a computational request, calling a semantic model to generate executable code based on the report summary text request and the execution code prompt, and calling a semantic model to execute the executable code to process the data in the target report to obtain a second report summary text.

[0100] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the following method for generating report summary text: determining the request type of the report summary text request, wherein the request type includes a computational request and a non-computational request; when the request type is a non-computational request, determining the target report indicated by the report summary text request, and when the number of tokens in the target report exceeds a preset number of tokens, performing compression sampling processing on the data in the target report, and calling a semantic model to generate a first report summary text based on the compression sampling result and the report summary text request; when the request type is a computational request, calling a semantic model to generate executable code based on the report summary text request and an execution code prompt, and calling a semantic model to execute the executable code to process the data in the target report, thereby obtaining a second report summary text.

[0101] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0102] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0103] The units described as separate components may or may not be physically separate. 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 units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0104] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0105] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0106] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for generating report summary text, characterized in that, include: Determine the request type for the report summary text request, wherein the request type includes computational requests and non-computational requests; If the request type is a non-computation request, determine the target report indicated by the report summary text request, and if the number of tokens in the target report exceeds a preset number of tokens, perform compression sampling on the data in the target report, and call the semantic model to generate a first report summary text based on the compression sampling result and the report summary text request; When the request type is a calculation request, the semantic model is invoked to generate executable code based on the report summary text request and the execution code prompt, and the semantic model is invoked to execute the executable code to process the data in the target report and obtain the second report summary text.

2. The report summary text generation method according to claim 1, characterized in that, The data compression and sampling process in the target report includes: Identify the data to be compressed in the target report; The data to be compressed is sampled to obtain multiple sampled data points; The importance evaluation value of the sampled data points is determined, and the sampled data points are filtered according to the importance evaluation value to obtain the compressed sampling result.

3. The report summary text generation method according to claim 2, characterized in that, Determining the importance evaluation value of the sampled data points includes: According to the position information of the sampled data points in the data to be compressed, the sampled data points are sorted to obtain a sequence of sampled data points; In the sequence of sampled data points, the average value of the adjacent sampled data points of each sampled data point is determined; Determine the difference between the sampled data points and the average value; Based on the difference, the importance evaluation value of the sampled data point is determined, wherein the larger the difference, the larger the importance evaluation value.

4. The report summary text generation method according to claim 1, characterized in that, Invoking the semantic model to execute the executable code to process the data in the target report includes: If the executable code fails to execute, obtain the error information of the executable code; Update the execution code prompt based on the error message; The semantic model is invoked to update the executable code based on the updated execution code prompt; The updated executable code is executed by invoking the semantic model to process the data in the target report.

5. The report summary text generation method according to claim 1, characterized in that, Before invoking the semantic model to generate executable code based on the report summary text request and execution code prompts, the method further includes: The data in the target report are sampled to obtain sampled data; The sampled data is input into the semantic model, thereby allowing the semantic model to learn the data format of the data in the target report.

6. The report summary text generation method according to claim 1, characterized in that, The report summary text method also includes: Create a report template and determine the corresponding application programming interface (API) for the report template; The target component in the report template is bound to the corresponding data registry interface through the application programming interface, and the data required by the target component is obtained through the data registry. The target report is obtained by filling the data required by the component into the report template.

7. The report summary text generation method according to claim 1, characterized in that, Creating a report template includes: In response to an operation command from the target object, a target component is determined from a set of candidate components, and the target component is placed at the position indicated by the operation command; In response to the debugging instructions of the target object, adjust the component properties of the target component.

8. A report summary text generation device, characterized in that, include: The first processing module is used to determine the request type of the report summary text request, wherein the request type includes calculation requests and non-computation requests; The second processing module, when the request type is a non-computation request, determines the target report indicated by the report summary text request, and when the number of tokens in the target report exceeds a preset number of tokens, performs compression sampling on the data in the target report, and calls the semantic model to generate a first report summary text based on the compression sampling result and the report summary text request; The third processing module, when the request type is a calculation request, calls the semantic model to generate executable code based on the report summary text request and the execution code prompt, and calls the semantic model to execute the executable code to process the data in the target report to obtain the second report summary text.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores a program, wherein when the program is executed, it controls the device where the non-volatile storage medium is located to execute the report summary text generation method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: A memory and a processor, the processor being configured to run a program stored in the memory, wherein the program, when running, executes the report summary text generation method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, The method includes a computer program that, when executed by a processor, implements the report summary text generation method according to any one of claims 1 to 7.

Images