A report generation method, device, equipment and medium based on digital employees

By obtaining user requests and data, using large language models to generate personalized prompt words and prediction models, and adjusting reports with employee feedback, it solves the problem that digital clones can't understand employees' unique needs, improves the personalization and quality of reports, and achieves innovation Sexual and forward-looking reports are generated.

CN118627487BActive Publication Date: 2024-11-15FUJIAN DIANSHI NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410882773.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-15
Estimated Expiration
2044-07-03

AI Technical Summary

Technical Problem

The existing digital clones have difficulty fully understanding each employee’s unique work style and needs, resulting in the lack of personalization of the generated reports, over-reliance on historical data, and lack of innovation and forward-lookingness.

Method used

By obtaining user requests, obtaining report templates and user data, using large language models to obtain personalized prompt words and historical data, combining the prediction model to generate a report draft, and adjusting it through real employee feedback, and finally generating a personalized and innovative report.

Benefits of technology

Improve the personalization and quality of the report, reduce employees' investment in repetitive text work, making the report not only based on historical data, but also innovative and forward-looking, ensuring the accuracy and practicality of the report.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627487B_ABST
    Figure CN118627487B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention provides a report generation method, device, equipment and medium based on digital employees, and relates to the field of AI digital human technology. The method includes S1, obtaining a request for generating a report. S2, according to the request, obtaining a report template and extracting work overview related data and work plan related data from user data. S3, according to the request, obtaining personalized prompt words through the first language model. S4, inputting the request, work overview related data and personalized prompt words into the second language model to obtain the work overview of the preset time in the past. S5, according to the report cycle, the work plan related data is sorted into time series data, and prediction is performed to obtain prediction data. S6, inputting the request, the prediction data of the next time series and the personalized prompt words into the second language model to obtain the work plan of the preset time in the future. S7, generating a draft report based on the work overview and work plan. S8, adjusting the draft report to obtain the final report.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of AI digital human technology, and in particular to a report generation method, device, equipment and medium based on digital employees. Background Art

[0002] Big language models are a breakthrough technology in the field of artificial intelligence. Based on the principles of deep learning, they build models on large amounts of text data through training, enabling them to understand and generate natural language. These models typically use complex neural network architectures, such as transformers, and use attention mechanisms to handle language serialization issues. Big language models can perform a variety of language tasks, including but not limited to language translation, text summarization, sentiment analysis, question answering, and content creation.

[0003] Building a digital twin / digital assistant based on a large language model to assist in completing some textual tasks can greatly improve employee work efficiency. For example, in scenarios where standard format documents (weekly reports, daily reports, etc.) are generated. Digital twins can effectively reduce repetitive textual work, allowing employees to devote more energy to tasks that require creativity and strategic thinking.

[0004] However, existing digital twins have difficulty fully understanding each employee’s unique work style and needs, resulting in a lack of personalization in the generated content. In addition, the generated reports are overly dependent on historical data, resulting in a lack of innovation and foresight.

[0005] In view of this, the applicant filed this application after studying the existing technology. Summary of the invention

[0006] The present invention provides a report generation method, apparatus, device and medium based on digital employees to improve at least one of the above technical problems.

[0007] In a first aspect, an embodiment of the present invention provides a report generation method based on digital employees, which comprises steps S1 to S8.

[0008] S1. Obtain a user's request for generating a report.

[0009] S2. According to the request, a report template is obtained, and work overview related data and work plan related data are extracted from the user data according to the request and the report template. The report template includes an employee basic information part, a work overview part, and a work plan part. The user data includes the employee's communication data on the company's communication platform, work file records, and historical reports.

[0010] S3. According to the request, a personalized prompt word is obtained through the first language model.

[0011] S4. Input the request, the work summary related data and the personalized prompt word into the second largest language model to obtain the work summary of the preset time in the past.

[0012] S5. Arrange the work plan-related data into time series data according to the report cycle, and predict the time series data through a prediction model to obtain prediction data for the next time series.

[0013] S6. Input the request, the predicted data of the next time series and the personalized prompt word into the second largest language model to obtain a future work plan of a preset duration.

[0014] S7. Generate a draft report based on the work overview of the past preset time and the work plan of the future preset time.

[0015] S8. Adjust the draft report based on the feedback from real employees to obtain the final report.

[0016] In an optional embodiment, the work file record includes a record of changes in the file content and a record of changes in the file storage location. The file storage location is set according to the workflow, and at least includes in progress and completed. Preferably, step S2 specifically includes S21 to S29.

[0017] S21. According to the request, obtain a report template and a historical report corresponding to the report template, wherein the report template is a daily report, a weekly report or a monthly report.

[0018] S22. According to the report template, obtain a time range of the report.

[0019] S23. Extracting the record contents within the time range from the communication data and work file records.

[0020] S24, extract the first planning content of the work planning part of the last historical report, use the first text vector similarity to perform similarity matching in the record content, and obtain the historical planning related content. The calculation model of the first text vector similarity is , where is the first text vector, is the vector in the database, yes The Euclidean norm of yes The Euclidean norm of .

[0021] S25. Extract a second keyword from the work summary part of the report template, and map the second keyword to a pre-built work-related knowledge graph to obtain work summary-related entities.

[0022] S26. Search the record content according to the work summary related entities to obtain template related content.

[0023] S27. Use a general large language model to summarize the unmatched parts of the record content and the change records of the file storage location to obtain a summary.

[0024] S28. Acquire work overview related data according to the historical planning related content, template related content and the summary.

[0025] S29. Extract the overview contents of the work overview part and the planning contents of the work planning part in multiple historical reports in chronological order to obtain work planning related data.

[0026] In an optional embodiment, step S3 is specifically used to input the request into a first large language model, and the first large language model outputs a personalized prompt word.

[0027] Preferably, the first language model is trained according to the following steps:

[0028] According to the user's communication data, the user's word usage habits are obtained.

[0029] Get the user's report topic trend based on the user's historical reports.

[0030] According to the word usage habits and the report topic trends, the general large language model is fine-tuned to obtain the first large language model.

[0031] In an optional embodiment, step S5 specifically includes steps S51 to S54.

[0032] S51. Adjust the summary content in the work planning related data and the work summary of the preset time in the past into a first periodic text sequence according to the report cycle, and then predict the work summary of the next cycle through a pre-constructed first prediction model.

[0033] S52: Adjust the change record of the file storage location into a second periodic text sequence according to the report period, and then predict the change record of the next period by using the second prediction model.

[0034] S53. Adjust the planning content in the work planning related data into a third periodic text sequence according to the report period, and then predict the planning content of the current report through the third prediction model.

[0035] S54: Acquire forecast data for the next time series according to the work overview of the next cycle, the change record of the next cycle and the planning content of the current report.

[0036] In an optional embodiment, the first prediction model and the third prediction model include an input layer, an LSTM layer, a Dropout layer, a bidirectional LSTM layer, multiple fully connected layers and an output layer connected in sequence. Among them, the input layer is used to receive the vectorized representation of periodic text. The LSTM layer is used to process the output of the embedding layer to capture short-term dependencies and contains 128 units. The Dropout layer is used to reduce overfitting, and the Dropout ratio is 0.2 or 0.5. The bidirectional LSTM layer is used to process past and future information simultaneously to enhance the model's understanding of time series data. The fully connected layer is used to learn complex nonlinear relationships, and each fully connected layer is connected to an output layer to correspond to an event. The output layer uses a softmax activation function.

[0037] In an optional embodiment, the second prediction model includes an input layer, an embedding layer, a GRU layer, a Dropout layer, a sequence expansion and an output layer connected in sequence. The input layer is used to receive the original sequence data. The embedding layer is used to convert the discrete data of the input layer into a continuous word vector. The GRU layer is composed of a plurality of GRU units, each of which includes an update gate and a reset gate to control the flow and forgetting of information. The Dropout layer is used to reduce overfitting. The sequence is expanded so that at each time step, the output of the GRU layer will be used as the input of the next time step until the sequence ends. The output layer is a parallel output layer, which is used to set an independent output layer for each event.

[0038] In an optional embodiment, step S8 specifically includes: the employee adjusts the request for generating a report according to the draft report or directly modifies the draft report to obtain a final report.

[0039] In an optional embodiment, the digital employee can also be used to answer query requests from other employees.

[0040] The digital worker conversational communication steps include:

[0041] A1. Get query requests from other employees.

[0042] A2. According to the query request, obtain a second personalized prompt word through the first language model.

[0043] A3. According to the query request, relevant text is retrieved from the user data corresponding to the digital employee through vector similarity.

[0044] A4. Input the second personalized prompt word and the related text into the second largest language model to obtain an answer to the query request.

[0045] In a second aspect, an embodiment of the present invention provides a report generation device based on digital employees, which includes a report request module, a template acquisition module, a personalized prompt word module, a work overview module, a prediction module, a work plan module, a report draft module and a final report module.

[0046] The report request module is used to obtain user requests for generating reports.

[0047] The template acquisition module is used to acquire a report template according to the request, and extract work overview related data and work plan related data from the user data according to the request and the report template. The report template includes an employee basic information part, a work overview part, and a work plan part. The user data includes the employee's communication data on the company's communication platform, work file records, and historical reports.

[0048] The personalized prompt word module is used to obtain the personalized prompt word through the first language model according to the request.

[0049] The work summary module is used to input the request, the work summary related data and the personalized prompt word into the second language model to obtain the work summary of the preset time in the past.

[0050] The prediction module is used to organize the work plan related data into time series data according to the report cycle, and predict the time series data through the prediction model to obtain the prediction data of the next time series.

[0051] The work plan module is used to input the request, the prediction data of the next time series and the personalized prompt word into the second language model to obtain a work plan of a preset time in the future.

[0052] The report draft module is used to generate a report draft based on the work overview of the past preset time and the work plan of the future preset time.

[0053] The final report module is used to adjust the draft report according to the feedback of real employees to obtain the final report.

[0054] In a third aspect, an embodiment of the present invention provides a report generation device based on digital employees, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement the report generation method based on digital employees as described in any paragraph of the first aspect.

[0055] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the report generation method based on digital employees as described in any paragraph of the first aspect.

[0056] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0057] A report generation method based on digital employees in an embodiment of the present invention combines artificial intelligence technology with personalized services, aiming to improve work efficiency, enhance the personalization and quality of reports, and promote employee participation and satisfaction. Improve work efficiency by automating the report generation process, reduce employees' investment in repetitive text work, and enable employees to devote more energy to tasks that require creative and strategic thinking. Improve the quality of reports: By combining historical data analysis and future forecasts, the report is not only based on historical data, but also innovative and forward-looking, which improves the overall quality of the report. Ensure the accuracy of the report: Employees can participate in the report generation process through feedback to ensure the accuracy and practicality of the report. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0059] Figure 1 is a flow chart of a digital employee-based report generation method.

[0060] Figure 2 is a logical block diagram of a report generation method based on a digital employee.

[0061] Figure 3 This is a logical diagram of the training process of the first language model. DETAILED DESCRIPTION

[0062] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] Example 1, please refer to Figure 1 and Figure 2 The first embodiment of the present invention provides a method for generating a report based on a digital employee, which can be performed by a report generating device based on a digital employee (hereinafter referred to as a report generating device based on a digital employee). In particular, the method is performed by one or more processors in the report generating device based on a digital employee to implement steps S1 to S8.

[0064] S1. Obtain a user's request for generating a report.

[0065] Specifically, the request can be a simple natural language such as "generate a weekly report", or a sentence with certain requirements on the report content, which is not specifically limited in the present invention. The user can input the request for generating a report in natural language by text input or voice input.

[0066] S2. According to the request, a report template is obtained, and according to the request and the report template, work overview related data and work planning related data are extracted from the user data.

[0067] The report template includes employee basic information, work overview, and work planning. The user data includes employee communication data on the company communication platform, work file records, and historical reports.

[0068] Specifically, user data includes 1. Conversation between employees (through conversation, the problem of insufficient stylization is compensated). 2. Employees' work files (through file changes, the problem of inaccurate report content is compensated), and work file records. 3. Weekly report template (including things completed in the past week and things planned to be completed in the next week). 4. Historical weekly report (through things completed in the past, predictions are made to compensate for the lack of foresight). Preferably, work file records include changes in file content and changes in file storage location. File storage locations are set according to the workflow, including at least in progress and completed.

[0069] Based on the above embodiment, in an optional embodiment of the present invention, step S2 specifically includes S21 to S29.

[0070] S21. According to the request, obtain a report template and a historical report corresponding to the report template.

[0071] S22. According to the report template, obtain a time range of the report.

[0072] Specifically, the report can be a daily report, a weekly report, a monthly report, or a report of other periods, which is not specifically limited in the present invention. Different reports are targeted at different objects, and periodic reports (eg, weekly reports) have a time range.

[0073] S23. Extracting the record contents within the time range from the communication data and work file records.

[0074] Specifically, relevant data within the time range of the report is obtained, the amount of data processing is reduced, and the accuracy of the data is increased.

[0075] S24, extract the first planning content of the work planning part of the last historical report, use the first text vector similarity to perform similarity matching in the record content, and obtain the historical planning related content. The calculation model of the first text vector similarity is , where is the first text vector, is the vector in the database, yes The Euclidean norm of yes The Euclidean norm of .

[0076] Specifically, the first text vector similarity measures the direction relationship between vectors rather than their size, which can effectively capture the semantic similarity of texts and is very suitable for text data. It is concerned with the direction of the text vector rather than its modulus.

[0077] S25. Extract a second keyword from the work summary part of the report template, and map the second keyword to a pre-built work-related knowledge graph to obtain work summary-related entities.

[0078] Specifically, the work-related knowledge graph is built based on various types of work-related data, such as employee information, project documents, historical reports, company policies, etc., as well as some professional books in this field (to make up for the lack of knowledge in a specific field). It contains work-related entities and relationships.

[0079] S26. Search the record content according to the work summary related entities to obtain template related content.

[0080] S27. Use a general large language model to summarize the unmatched parts of the record content and the change records of the file storage location to obtain a summary.

[0081] S28. Acquire work overview related data according to the historical planning related content, template related content and the summary.

[0082] Specifically, the content data of the planning part in the previous report and the clearly related content in the weekly report are strongly related content. The relevant data are screened out through vector similarity retrieval and text search respectively, and directly used as reference materials for subsequent steps, which greatly improves the quality of the draft report and has great practical significance. As for the data other than the strongly related data, the summary is used as the reference material for the subsequent steps, which can avoid omissions and reduce the amount of data processing, which has great practical significance.

[0083] S29. Extract the overview contents of the work overview part and the planning contents of the work planning part in multiple historical reports in chronological order to obtain work planning related data.

[0084] S3. According to the request, a personalized prompt word is obtained through the first language model. Specifically, the request is input into the first language model, and the first language model outputs the personalized prompt word.

[0085] Preferably, Figure 3 As shown, the first language model is trained according to the following steps:

[0086] B1. Obtain the user's word usage habits based on the user's communication data.

[0087] B2. Obtain the user's report topic trend based on the user's historical report.

[0088] B3. Fine-tune the general large language model according to the word usage habits and the report topic trends to obtain the first large language model.

[0089] Specifically, a new subsystem is developed for personalized prompt words, which is used to automatically generate prompt words based on user interactions and historical report content. This subsystem is used to analyze user query habits, keyword usage frequency, and report topic trends, and dynamically generate high-quality prompt words by combining the capabilities of large language models (such as ChatGPT, etc.). Through the described task interface, such as the interface for generating reports, the large model is pre-called to pre-generate appropriate prompt words for the large model that actually generates reports.

[0090] S4. Input the request, the work summary related data and the personalized prompt word into the second largest language model to obtain the work summary of the preset time in the past.

[0091] Specifically, the large language model automatically organizes the work content within a certain period of time in the past based on the input content.

[0092] S5, organizing the work plan related data into time series data according to the report cycle, and predicting the time series data through the prediction model to obtain the prediction data of the next time series. Preferably, step S5 specifically includes steps S51 to S54.

[0093] S51. Adjust the summary content in the work planning related data and the work summary of the preset time in the past into a first periodic text sequence according to the report cycle, and then predict the work summary of the next cycle through a pre-constructed first prediction model.

[0094] S52: Adjust the change record of the file storage location into a second periodic text sequence according to the report period, and then predict the change record of the next period by using the second prediction model.

[0095] S53. Adjust the planning content in the work planning related data into a third periodic text sequence according to the report period, and then predict the planning content of the current report through the third prediction model.

[0096] S54: Acquire forecast data for the next time series according to the work overview of the next cycle, the change record of the next cycle and the planning content of the current report.

[0097] On the basis of the above embodiments, in an optional embodiment of the present invention, the first prediction model and the third prediction model use a long short-term memory network, including an input layer, an LSTM layer, a Dropout layer, a bidirectional LSTM layer, a plurality of fully connected layers and an output layer connected in sequence. Among them, the input layer is used to receive the vectorized representation of periodic text. The LSTM layer is used to process the output of the embedding layer to capture short-term dependencies and includes 128 units. The Dropout layer is used to reduce overfitting, and the Dropout ratio is 0.2 or 0.5. The bidirectional LSTM layer is used to process past and future information simultaneously to enhance the model's understanding of time series data. The fully connected layer is used to learn complex nonlinear relationships, and each fully connected layer is connected to an output layer to correspond to an event. The output layer uses a softmax activation function.

[0098] Specifically, the Long Short-Term Memory (LSTM) network can effectively maintain long-term state information and can make predictions for more complex text content.

[0099] On the basis of the above embodiments, in an optional embodiment of the present invention, the second prediction model uses a gated recurrent unit (GRU) including an input layer, an embedding layer, a GRU layer, a Dropout layer, a sequence expansion and an output layer connected in sequence. Among them, the input layer is used to receive the original sequence data. The embedding layer is used to convert the discrete data of the input layer into a continuous word vector. The GRU layer is composed of a plurality of GRU units, each of which includes an update gate and a reset gate to control the flow and forgetting of information. The Dropout layer is used to reduce overfitting. The sequence is expanded so that at each time step, the output of the GRU layer will be used as the input of the next time step until the end of the sequence. The output layer is a parallel output layer, which is used to set an independent output layer for each event.

[0100] Specifically, the changes in the file storage location are scattered and the content is brief. Therefore, an embedding layer is set to convert it. The gated recurrent unit is used for prediction. The structure of the gated recurrent unit is simpler, which greatly reduces the amount of calculation.

[0101] S6: Input the request, the next time series prediction data and the personalized prompt word into the second language model to obtain a future work plan of a preset duration. Specifically, the large language model automatically sorts out the work content of a certain duration in the future according to the input content as the work plan.

[0102] S7: Generate a draft report based on the work summary of the preset time in the past and the work plan of the preset time in the future. Specifically, step S7 also includes: obtaining basic information of employees and adding it to the draft report.

[0103] S8. Adjust the draft report according to the feedback from the real employees to obtain the final report. Preferably, step S8 specifically includes: the employee adjusts the request for generating the report according to the draft report or directly modifies the draft report to obtain the final report.

[0104] A report generation method based on digital employees in an embodiment of the present invention combines artificial intelligence technology with personalized services, aiming to improve work efficiency, enhance the personalization and quality of reports, and promote employee participation and satisfaction. Improve work efficiency by automating the report generation process, reduce employees' investment in repetitive text work, and enable employees to devote more energy to tasks that require creative and strategic thinking. Improve the quality of reports: By combining historical data analysis and future forecasts, the report is not only based on historical data, but also innovative and forward-looking, which improves the overall quality of the report. Ensure the accuracy of the report: Employees can participate in the report generation process through feedback to ensure the accuracy and practicality of the report.

[0105] Based on the above embodiment, in an optional embodiment of the present invention, the digital employee can also be used to answer query requests from other employees.

[0106] The steps of the dialogue communication of the digital employee include step A1 to step A4.

[0107] A1. Get query requests from other employees.

[0108] A2. According to the query request, obtain a second personalized prompt word through the first language model.

[0109] A3. According to the query request, relevant text is retrieved from the user data corresponding to the digital employee through vector similarity.

[0110] A4. Input the second personalized prompt word and the related text into the second largest language model to obtain an answer to the query request.

[0111] Specifically, a report generation method based on digital employees in an embodiment of the present invention can not only automatically generate weekly reports based on the employee's knowledge base, but also provide personalized answers to questions based on the employee's knowledge base. Digital employees can be directly questioned during non-working hours to avoid disturbing real employees, which has great practical significance.

[0112] Embodiment 2: The embodiment of the present invention provides a report generation device based on digital employees, which includes a report request module, a template acquisition module, a personalized prompt word module, a work overview module, a prediction module, a work plan module, a report draft module and a final report module.

[0113] The report request module is used to obtain user requests for generating reports.

[0114] The template acquisition module is used to acquire a report template according to the request, and extract work overview related data and work plan related data from the user data according to the request and the report template. The report template includes an employee basic information part, a work overview part, and a work plan part. The user data includes the employee's communication data on the company's communication platform, work file records, and historical reports.

[0115] The personalized prompt word module is used to obtain the personalized prompt word through the first language model according to the request.

[0116] The work summary module is used to input the request, the work summary related data and the personalized prompt word into the second language model to obtain the work summary of the preset time in the past.

[0117] The prediction module is used to organize the work plan related data into time series data according to the report cycle, and predict the time series data through the prediction model to obtain the prediction data of the next time series.

[0118] The work plan module is used to input the request, the prediction data of the next time series and the personalized prompt word into the second language model to obtain a work plan of a preset time in the future.

[0119] The report draft module is used to generate a report draft based on the work overview of the past preset time and the work plan of the future preset time.

[0120] The final report module is used to adjust the draft report according to the feedback of real employees to obtain the final report.

[0121] Embodiment 3: The embodiment of the present invention provides a report generation device based on digital employees, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement the report generation method based on digital employees as described in any paragraph of Embodiment 1.

[0122] Embodiment 4: The embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the report generation method based on digital employees as described in any paragraph of Embodiment 1.

[0123] It is understandable that the digital employee-based report generation device may be an electronic device with computing capabilities, such as a portable notebook computer, a desktop computer, a server, a smart phone, or a tablet computer.

[0124] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0125] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0126] If the function is implemented in the form of a software function module 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 the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0127] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0128] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0129] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0130] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0131] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A report generation method based on digital employees, characterized in that: Include: Get the user's request to generate a report; According to the request, a report template is obtained, and work overview-related data and work plan-related data are extracted from the user data according to the request and the report template; wherein the report template includes an employee basic information part, a work overview part, and a work plan part; and the user data includes the employee's communication data on the company's communication platform, work file records, and historical reports; According to the request, obtaining personalized prompt words through the first language model; Inputting the request, the work summary related data and the personalized prompt word into a second language model to obtain a work summary of a preset time in the past; Arrange the work plan-related data into time series data according to the reporting cycle, and predict the time series data through a prediction model to obtain the prediction data of the next time series; Input the request, the next time series prediction data and the personalized prompt word into the second largest language model to obtain a future work plan of a preset duration; Generate a draft report based on the work overview of the past preset time and the work plan of the future preset time; Adjust said draft report based on feedback from real employees to obtain a final report; The work planning related data is sorted into time series data according to the report cycle, and the time series data is predicted through the prediction model to obtain the prediction data of the next time series, which specifically includes: Adjusting the summary content in the work planning related data and the work summary of the preset duration in the past into a first periodic text sequence according to the report cycle, and then predicting the work summary of the next cycle through a pre-built first prediction model; Adjusting the change record of the file storage location into a second periodic text sequence according to the report period, and then predicting the change record of the next period by using a second prediction model; Adjusting the planning content in the work planning related data into a third periodic text sequence according to the report cycle, and then predicting the planning content of the current report through the third prediction model; According to the work overview of the next cycle, the change record of the next cycle and the planning content of the current report, the forecast data of the next time series is obtained.

2. The method for generating a report based on digital employees according to claim 1, characterized in that: Work file records include changes in file content and file storage location; file storage locations are set according to the workflow, including at least in progress and completed; According to the request, a report template is obtained, and work overview related data and work planning related data are extracted from the user data according to the request and the report template, specifically including: According to the request, a report template and a historical report corresponding to the report template are obtained; wherein the report template is a daily report, a weekly report or a monthly report; According to the report template, obtain the time range of the report; Extract the records within the time range from the communication data and work file records; Extract the first planning content of the work planning part of the last historical report, use the first text vector similarity to perform similarity matching in the record content, and obtain the historical planning related content; wherein the calculation model of the first text vector similarity is , where is the first text vector, is the vector in the database, yes The Euclidean norm of yes The Euclidean norm of ; Extracting a second keyword from the work summary part of the report template, and mapping the second keyword to a pre-built work-related knowledge graph to obtain work summary-related entities; Searching the record content according to entities related to the work overview to obtain template-related content; Use a general large language model to summarize the unmatched parts of the record content and the change records of the file storage location to obtain a summary; Acquire work overview-related data based on the historical planning-related content, the template-related content and the summary; The overview contents of the work overview part and the planning contents of the work planning part of multiple historical reports are extracted in chronological order to obtain data related to the work planning.

3. The method for generating a report based on digital employees according to claim 1, characterized in that: According to the request, a personalized prompt word is obtained through the first language model, specifically including: Inputting the request into a first language model, and the first language model outputting a personalized prompt word; The first language model is trained according to the following steps: Obtain the user's word usage habits based on the user's communication data; Obtain the user's report topic trend based on the user's historical report; According to the word usage habits and the report topic trends, the general large language model is fine-tuned to obtain the first large language model.

4. The method for generating a report based on digital employees according to claim 1, characterized in that: The first prediction model and the third prediction model include an input layer, an LSTM layer, a Dropout layer, a bidirectional LSTM layer, multiple fully connected layers and an output layer connected in sequence; wherein, the input layer is used to receive the vectorized representation of periodic text; the LSTM layer is used to process the output of the embedding layer to capture short-term dependencies, and contains 128 units; the Dropout layer is used to reduce overfitting, and the Dropout ratio is 0.2 or 0.5; the bidirectional LSTM layer is used to process past and future information at the same time to enhance the model's understanding of time series data; the fully connected layer is used to learn complex nonlinear relationships, and each fully connected layer is connected to an output layer to correspond to an event; the output layer uses a softmax activation function; The second prediction model includes an input layer, an embedding layer, a GRU layer, a Dropout layer, a sequence expansion and an output layer connected in sequence; wherein, the input layer is used to receive the original sequence data; the embedding layer is used to convert the discrete data of the input layer into a continuous word vector; the GRU layer is composed of multiple GRU units, each of which contains an update gate and a reset gate to control the flow and forgetting of information; the Dropout layer is used to reduce overfitting; the sequence is expanded so that at each time step, the output of the GRU layer will be used as the input of the next time step until the end of the sequence; the output layer is a parallel output layer, which is used to set an independent output layer for each event.

5. The method for generating a report based on digital employees according to any one of claims 1 to 4, characterized in that: Adjust the draft report based on real employee feedback to obtain a final report that includes: The employee adjusts the request for generating the report according to the draft report or directly modifies the draft report to obtain the final report.

6. The method for generating a report based on digital employees according to any one of claims 1 to 4, characterized in that: Digital employees can also be used to answer queries from other employees; The steps of the digital employee's dialogue communication include: Get query requests from other employees; According to the query request, obtaining a second personalized prompt word through the first language model; According to the query request, relevant text is retrieved from user data corresponding to the digital employee by vector similarity; The second personalized prompt word and the related text are input into a second language model to obtain an answer to the query request.

7. A report generation device based on digital employees, characterized in that: Include: A report request module is used to obtain a user's request for generating a report; A template acquisition module, used to acquire a report template according to the request, and extract work overview related data and work plan related data from user data according to the request and the report template; wherein the report template includes an employee basic information part, a work overview part, and a work plan part; and the user data includes employee communication data on the company communication platform, work file records, and historical reports; A personalized prompt word module, used to obtain a personalized prompt word through a first language model according to the request; A work summary module, used for inputting the request, the work summary related data and the personalized prompt word into a second language model to obtain a work summary of a preset time in the past; A prediction module is used to organize the work plan related data into time series data according to the report cycle, and predict the time series data through a prediction model to obtain the prediction data of the next time series; A work plan module, used for inputting the request, the next time series prediction data and the personalized prompt word into a second language model to obtain a future work plan of a preset duration; A report draft module, used for generating a report draft according to the work summary of the past preset time and the work plan of the future preset time; A final report module for adjusting the draft report based on feedback from real employees to obtain a final report; The prediction module is specifically used for: Adjusting the summary content in the work planning related data and the work summary of the preset duration in the past into a first periodic text sequence according to the report cycle, and then predicting the work summary of the next cycle through a pre-built first prediction model; Adjusting the change record of the file storage location into a second periodic text sequence according to the report period, and then predicting the change record of the next period by using a second prediction model; Adjusting the planning content in the work planning related data into a third periodic text sequence according to the report cycle, and then predicting the planning content of the current report through the third prediction model; According to the work overview of the next cycle, the change record of the next cycle and the planning content of the current report, the forecast data of the next time series is obtained.

8. A report generation device based on digital employees, characterized in that: The system comprises a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement the report generation method based on digital employees as claimed in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the report generation method based on digital employees according to any one of claims 1 to 6.