Method and system for generating operation report based on enterprise index assets

Abstract

The invention provides a method and a system for generating an operation report based on enterprise index assets, electronic equipment and a storage medium, and aims to solve the problems that the analysis of the operation report is time-consuming and incomplete, and the index cannot be selected autonomously, the method comprises the following steps: carrying out index storage on business index data of an enterprise through a graph database, the incidence relation between the indexes is stored; identifying a user intention and recommending a diagnosis direction and a diagnosis index according to the user intention in an intelligent question and answer interaction mode; and outputting a diagnosis conclusion by utilizing LLM according to the recommended diagnosis indexes, and generating a visual diagnosis report by adopting an AIGC technology and index graphical assets. The diagnosis indexes can be quickly determined according to the user intention, and the conclusion can be diagnosed through the LLM, so that business personnel can perform comprehensive diagnosis based on the indexes more flexibly, and the diagnosis report can be generated more efficiently.

Description

Technical Field

[0001] The present disclosure relates to the field of information data technology, and in particular to a method for generating an operating report based on enterprise indicator assets, an FTTH project network access acceptance system based on AI duty, an electronic device, and a computer-readable storage medium. Background Art

[0002] In the digital age, the status of data analysis is becoming increasingly prominent. It has not only become an indispensable assistant for corporate decision makers, but also the key to promoting innovation and improving market competitiveness. The collection, integration, analysis and research of various types of data play a very important role in the development and decision-making of enterprises. However, in the current field of enterprise management and data analysis, when business personnel use indicators to conduct all-round business diagnosis to find deficiencies and generate diagnostic reports for leadership business decisions, the following problems still exist:

[0003] The traditional method of discovering indicator association models is time-consuming and incomplete: Enterprise indicators are complex and numerous. In the traditional way, indicators stored in relational databases take more than 3 hours for business personnel to quickly find the relationship between indicators, and the search is not comprehensive.

[0004] Limitations of real-time interaction and self-selection of indicators: Business personnel cannot independently select diagnostic directions and indicators and can only obtain diagnostic results through traditional methods of raising requirements and developing technologies, which takes more than 3 days.

[0005] Inefficient generation of visual reports: Visual reports of diagnostic results are manually developed and put online in a system that takes more than seven days. There is a lack of intelligent means to quickly generate reports for business personnel in a short period of time. Summary of the invention

[0006] In order to at least solve the problems in the prior art that the business report is analyzed based on business experience, which is time-consuming and incomplete, and the indicators cannot be selected independently, and the development process has many long cycles. The present invention provides a method for generating a business report based on enterprise indicator assets, a system for generating a business report based on enterprise indicator assets, an electronic device, and a computer-readable storage medium, which can quickly determine the diagnostic indicators according to the user's intention and diagnose the conclusion through LLM (Large Language Model), so that business personnel can perform all-round diagnosis based on indicators more flexibly and generate diagnostic reports more efficiently.

[0007] In a first aspect, the present disclosure provides a method for generating an operating report based on enterprise index assets, the method comprising:

[0008] The business indicator data of the enterprise is stored in the graph database, and the relationship between the indicators is stored;

[0009] Through intelligent question-answering interaction, it identifies user intent and recommends diagnostic directions and indicators based on user intent.

[0010] LLM is used to output diagnostic conclusions based on the recommended diagnostic indicators, and AIGC (Artificial Intelligence Generated Content) technology and indicator graphical assets are used to generate a visual diagnostic report.

[0011] Furthermore, the method further comprises:

[0012] For the business indicators of an enterprise, derivative indicators are created through the basic indicator interface operation, and the parent-child relationship between indicators is recorded. Indicators are put on and taken off the shelves according to business needs, and all business indicators are stored in the graph database.

[0013] Furthermore, the interactive method of intelligent question and answer is used to identify the user's intention and recommend the diagnosis direction and diagnosis indicators according to the user's intention, including:

[0014] Obtain the diagnostic requirements input by users through the question-and-answer interface and identify user intent through a large model;

[0015] The user intention is processed through the BERT (Bidirectional Encoder Representations from Transformers) error correction model, sensitive words are filtered, indicator association relationship queries are called, and recommended diagnostic indicators are output.

[0016] Furthermore, the calling of indicator association relationship query and output of recommended diagnostic indicators includes:

[0017] When you need to query the correlation indicators of the corresponding indicators of user intent, use the Neo4j traversal framework to calculate the route, including:

[0018] Filter the path based on time, given the path from the starting node of the traversal to the current node, and return the connection that can be traversed further down, input parameters, the path extending from the starting node of the traversal to the current traversal node and the branch status provided by the client, and output diagnostic indicators.

[0019] Furthermore, the outputting of the diagnosis conclusion using LLM according to the recommended diagnosis index includes:

[0020] According to the recommended diagnostic indicators, feature data is input, and feature profiling, correlation analysis, causal analysis, and model weight fitting are performed through feature clustering and Shapely Value expert model. In combination with prompt engineering and model fine-tuning, multiple results are finally summarized, diagnostic conclusions are inferred, and the diagnostic conclusions are output through LLM.

[0021] In a second aspect, the present disclosure provides a system for generating an operating report based on enterprise index assets, the system comprising:

[0022] A storage module is configured to store the business indicator data of the enterprise through a graph database and store the association relationship between the indicators;

[0023] A diagnosis module, which is configured to identify user intent in an interactive manner of intelligent question and answer and recommend diagnosis directions and diagnostic indicators based on the user intent;

[0024] The generation module is configured to output the diagnosis conclusion using LLM according to the recommended diagnosis indicators, and to generate a visual diagnosis report using AIGC technology and indicator graphical assets.

[0025] Furthermore, the system also includes a creation module;

[0026] The creation module is set to create derived indicators for the business indicators of the enterprise through basic indicator interface operations, and record the parent-child relationship between indicators. The indicators are put on and taken off the shelves according to business needs, and all business indicators are stored in the graph database.

[0027] Furthermore, the diagnosis module is specifically configured as follows:

[0028] Obtain the diagnostic requirements input by the user through the question-answering interface and identify the user's intention through the big model; and,

[0029] User intent is processed through the BERT error correction model, sensitive words are filtered, indicator association queries are called, and recommended diagnostic indicators are output.

[0030] In a third aspect, the present disclosure provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for generating an operating report based on enterprise indicator assets as described in any one of the first aspects.

[0031] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for generating an operating report based on enterprise indicator assets as described in any one of the first aspects above is implemented.

[0032] Beneficial effects:

[0033] The present disclosure provides a method for generating an operating report based on enterprise indicator assets, a system, an electronic device, and a storage medium for generating an operating report based on enterprise indicator assets. By managing indicator assets through Graph Database, using real-time interactive dialogue to accurately capture user intentions for rapid diagnosis and LLM diagnosis summary, business personnel can make all-round diagnosis based on indicators more flexible, generate diagnostic reports more efficiently, and make decisions more quickly based on the market's flexibility and changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a method for generating an operating report based on enterprise index assets provided in the first embodiment of the present disclosure;

[0035] Figure 2 An architecture diagram of a system for generating an operating report based on enterprise index assets provided in Embodiment 2 of the present disclosure;

[0036] Figure 3 A flowchart of a method for generating an operating report based on enterprise index assets provided in Embodiment 2 of the present disclosure;

[0037] Figure 4 A schematic diagram of the effect of querying the correlation index of the main network migration income index provided in the second embodiment of the present disclosure;

[0038] Figure 5 A schematic diagram of a process for generating a diagnosis conclusion provided in the second embodiment of the present disclosure;

[0039] Figure 6 This is an architecture diagram of a system for generating an operating report based on enterprise index assets provided in Embodiment 3 of the present disclosure;

[0040] Figure 7 This is an architecture diagram of an electronic device provided in Embodiment 4 of the present disclosure. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, the present disclosure is further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention, rather than to limit the present invention.

[0042] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence; and, in the absence of conflict, the embodiments in the present disclosure and the features in the embodiments can be arbitrarily combined with each other.

[0043] The terms used in the embodiments of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. The singular forms of "a", "said" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings.

[0044] In the subsequent description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present disclosure, and have no specific meanings. Therefore, "module", "component" or "unit" can be used in a mixed manner.

[0045] In the current field of enterprise management and data analysis, business personnel still face the following problems when using indicators to conduct all-round business diagnosis to find deficiencies and generate diagnostic reports for leadership business decision-making:

[0046] 1. Index assets are stored in relational databases, which cannot deeply explore the correlation model between enterprise business indicators. Business personnel can only analyze based on business experience, which is time-consuming and incomplete.

[0047] 2. There is a lack of a method for business personnel to interact in real time and independently select indicators to generate diagnostic results. They need to communicate with technical personnel on requirements, development, and obtaining diagnostic results, which involves many steps and a long cycle.

[0048] 3. Lack of methods to quickly generate visual reports based on the results of diagnosis in real-time business needs

[0049] To sum up, the present disclosure aims to propose a method for generating business diagnostic reports based on enterprise indicator assets, so as to overcome the shortcomings of the existing technology, improve the efficiency of business personnel in generating reports for comprehensive business diagnosis based on enterprise indicators, and provide strong support for enterprise decision-making.

[0050] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned problems existing in the prior art are described in detail below with specific embodiments. It is understandable that in the embodiments of the present application, the execution subject can perform some or all of the steps in the embodiments of the present application, and these steps or operations are only examples. The embodiments of the present application can also perform other operations or deformations of various operations. In addition, each step can be performed in a different order presented in the embodiments of the present application, and it is possible not to perform all the operations in the embodiments of the present application. And, the following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0051] Figure 1 A flowchart of a method for generating an operating report based on enterprise index assets provided in the first embodiment of the present disclosure is shown in FIG. Figure 1 As shown, the method includes:

[0052] Step S101: The business indicator data of the enterprise is stored in a graph database, and the relationship between the indicators is stored;

[0053] Step S102: identifying the user's intention in an interactive manner of intelligent question and answer and recommending a diagnosis direction and a diagnosis indicator according to the user's intention;

[0054] Step S103: Use LLM to output the diagnostic conclusion according to the recommended diagnostic indicators, and use AIGC technology and indicator graphical assets to generate a visual diagnostic report.

[0055] In order to overcome the shortcomings of the prior art, improve the efficiency of business personnel in generating reports based on all-round business diagnosis based on enterprise indicators, and provide strong support for enterprise decision-making, the embodiment of the present disclosure provides a method for generating business reports based on enterprise indicator assets. First, for each business indicator of the enterprise, it is stored through a graph database mode. The graph database is suitable for processing complex relationship networks and has key features such as nodes, edges, and property graphs. Mainstream graph databases include Neo4j, OrientDB, etc. Graph databases are a type of database used to store relationships between entities. It uses a graph structure to represent data, in which entities are represented as nodes (Node) and the relationship between entities is represented as edges (Edge). The relationship between indicators can be easily stored through a graph database.

[0056] When a user uses a certain indicator to query, the associated indicator can be quickly found. For each service indicator, the nature and level of each indicator can be set. An exemplary indicator association service of an operator is shown in Table 1:

[0057] Table 1: Schematic list of indicator association business

[0058]

[0059] After creating a business schematic list of indicator association relationships, the indicators are stored through the Graph Database, and the listing and delisting of indicators can be managed.

[0060] When business personnel need to produce a diagnostic report for their business, the report can be used through an interactive intelligent question-and-answer method to identify user intent and recommend diagnostic directions and indicators.

[0061] For example, if the user inputs "help me generate a branch company revenue management diagnosis report", the big model identifies the user's intention, the organization is "a branch company", and the recommended diagnosis direction is "revenue", and the key indicator is "main mobile network revenue". And through the association relationship, it finds the related indicators such as "actual revenue increase, 100 yuan equivalent, and number of 5G development users".

[0062] After obtaining the data of each diagnostic indicator, the current situation analysis and problem analysis diagnostic conclusions can be generated in various ways, such as outputting diagnostic conclusions through a pre-trained model, using LLM to output diagnostic conclusions, and using AIGC technology and indicator graphical assets to generate a visual diagnostic report.

[0063] The disclosed embodiment manages indicator assets through Graph Database, uses real-time interactive dialogue to accurately capture user intentions for rapid diagnosis and summarizes through LLM diagnosis, so that business personnel can conduct all-round diagnosis based on indicators more flexibly, generate diagnostic reports more efficiently, and make decisions more quickly based on flexible and changeable market conditions.

[0064] Furthermore, the method further comprises:

[0065] For the business indicators of an enterprise, derivative indicators are created through the basic indicator interface operation, and the parent-child relationship between indicators is recorded. Indicators are put on and taken off the shelves according to business needs, and all business indicators are stored in the graph database.

[0066] According to actual needs, corresponding derivative indicators can be created and the relationship between the derivative indicators and other indicators can be determined to enrich the business indicator data in the Graph Database, dynamically discover the indicator relationship, provide corresponding reference data for each specific business diagnosis direction, and enrich the system performance.

[0067] The disclosed embodiment stores indicators in the Graph Database mode, and also utilizes BiLSTM (Bi-directional Long Short-Term Memory) + CRF (Conditional Random Field) deep learning and RNN (Recurrent Neural Network) recurrent neural network to realize the vectorization and multi-tuple of enterprise indicators, so as to solve the problems of insufficient accuracy and low efficiency of traditional methods when processing complex and nonlinear enterprise indicator data.

[0068] First, the enterprise indicator data were preprocessed, including data cleaning, normalization and other steps to ensure the quality and consistency of the data.

[0069] Then, a BiLSTM network is constructed to capture the temporal dependencies in the data. Through the bidirectional LSTM (Long Short-Term Memory, long short-term memory artificial neural network) unit, the model can consider both past and future information at the same time, so as to understand the data more accurately. Next, the output of BiLSTM is used as the input of CRF, and CRF is used to optimize the sequence labeling to obtain a more accurate vectorized representation. Including transfer matrix initialization, decoding algorithm, label constraint conditions to complete the CRF layer processing. Specifically including:

[0070] (1) Transfer matrix initialization: Use uniform distribution or initialization method based on prior knowledge to ensure the stability of model training. For example, the "Table 1: Business Schematic List of Indicator Association Relationships" in this article is used as the prior knowledge of enterprise indicators for training.

[0071] (2) Decoding algorithm: The Viterbi algorithm is used for dynamic programming decoding to find the globally optimal label sequence.

[0072] (3) Label constraints: Label constraints are introduced in the CRF layer (e.g., label1, label2, label3 in “B-label1 I-label2 I-label3…” must be of the same entity category) to further improve the accuracy of sequence labeling.

[0073] At the same time, a special RNN structure is designed for the multi-group data in corporate indicators. By introducing additional gating mechanisms and attention mechanisms, the model can better handle such complex data.

[0074] In the process of implementation, technical difficulties such as data imbalance and model overfitting were effectively solved by introducing regularization technology, adopting cross-validation, and adjusting model parameters, thus improving the performance and stability of the model.

[0075] Experimental results show that compared with the existing technology, this method has significant improvements in key indicators such as accuracy, recall rate and F1 score.

[0076] Furthermore, the interactive method of intelligent question and answer is used to identify the user's intention and recommend the diagnosis direction and diagnosis indicators according to the user's intention, including:

[0077] Obtain the diagnostic requirements input by users through the question-and-answer interface and identify user intent through a large model;

[0078] User intent is processed through the BERT error correction model, sensitive words are filtered, indicator association queries are called, and recommended diagnostic indicators are output.

[0079] When performing user identification, various existing large models can be used for identification, and the large models will be trained on intent based on the corpus of business personnel during use.

[0080] The diagnostic needs of the acquired users are identified through a large model to obtain specific key information, such as related subjects and key indicators. The information is then processed through the BERT error correction model and sensitive words are filtered. BERT uses the Transformer encoder structure to learn contextual information from both the left and right directions of the text, enabling the model to better understand the semantics of each word in the sentence. Then, the relevant diagnostic indicators are obtained based on the correlation between the indicators in the Graph Database.

[0081] Furthermore, the calling of indicator association relationship query and output of recommended diagnostic indicators includes:

[0082] When you need to query the correlation indicators of the corresponding indicators of user intent, use the Neo4j traversal framework to calculate the route, including:

[0083] Filter the path based on time, given the path from the starting node of the traversal to the current node, and return the connection that can be traversed further down, input parameters, the path extending from the starting node of the traversal to the current traversal node and the branch status provided by the client, and output diagnostic indicators.

[0084] Graph Database abstracts indicators as vertices and the relationships between indicators as edges, so that deep chain query and analysis of indicator relationships can be completed. Taking the "main mobile network income" indicator as an example, when it is necessary to query the relationship indicators of the main mobile network income indicator, the traversal framework of Neo4j is used to calculate the route:

[0085] Through the Interval Path Expander class, based on time filtering paths, given the path from the starting node of the traversal (main mobile network income) to the current node, the expand() method returns the connections that can be traversed further. Input parameters, the path extending from the starting node of the traversal to the current node of the traversal and the deliveryInterval branch state provided by the client, output Iterable <relationship>; The ParceRouteCalculator class is used to calculate all the logic required for the "main mobile network income" indicator from entering the network to reaching other business indicator routes, and finally quickly find the related indicators "actual revenue increase, hundred yuan equivalent, and number of 5G development users".

[0086] Furthermore, the outputting of the diagnosis conclusion using LLM according to the recommended diagnosis index includes:

[0087] According to the recommended diagnostic indicators, feature data is input, and feature profiling, correlation analysis, causal analysis, and model weight fitting are performed through feature clustering and Shapely Value expert model. In combination with prompt engineering and model fine-tuning, multiple results are finally summarized, diagnostic conclusions are inferred, and the diagnostic conclusions are output through LLM.

[0088] LLM is used to output diagnostic conclusions, such as "the branch is downstream, and its diagnosed operating problems mainly lie in the "main business mobile network income" and "100 yuan equivalent" being poor"; and AIGC technology is used to quickly generate a visual report on the diagnostic content. At the same time, users can download the diagnostic report to the local computer in the form of an image for use in work reporting and other scenarios.

[0089] When conducting diagnostic conclusion analysis, after obtaining indicator data, various expert models can be used to infer diagnostic conclusions. Specifically, the process of generating a visual diagnostic report includes:

[0090] 1. Input feature data processing

[0091] Data preprocessing: First, clean the input indicators and the characteristic data of the indicator daily or monthly account period data, including removing missing values, detecting and processing outliers, converting data types, etc., to ensure data quality. Then, perform feature standardization or normalization to eliminate the influence of different feature dimensions and improve the efficiency and accuracy of model training.

[0092] For example, in Table 2, the index value of the main operating income is a missing value, and the index value of the income from new development work orders of the mobile network is an abnormal value, which requires data preprocessing when inputting feature data.

[0093] Table 2: Obtained indicator data

[0094] Indicator Code Indicator name 2024-01 001 New development work order income from mobile network 19000987 Yuan 002 Broadband new development work order income 123 yuan 003 5G new development work order income 233 yuan 004 Main revenue

[0095] Feature selection: Use statistical methods (such as variance selection and correlation coefficient analysis) or machine learning algorithms (such as random forest importance scoring) to perform preliminary feature screening, retain features that are highly relevant to the diagnostic target, reduce redundant information, and improve model performance.

[0096] 2. Feature Clustering

[0097] K-means, DBSCAN, hierarchical clustering and other algorithms are used to cluster the preprocessed features, and similar or related features are grouped into the same group for subsequent analysis. In the clustering process, it is necessary to select appropriate distance metrics (such as Euclidean distance, Manhattan distance) and clustering parameters according to the data distribution characteristics.

[0098] 3. Shapely Value calculation

[0099] The importance of features is calculated using efficient algorithms such as TreeSHAP and KernelSHAP. The marginal contribution of each feature in different combinations is iteratively calculated to finally obtain the Shapley Value of each feature.

[0100] 4. Feature profiling and correlation analysis

[0101] Feature portrait construction: Based on the clustering results and Shapley Value, a detailed portrait is constructed for each feature, including the importance of the feature, the degree of correlation with other features, and the way it contributes to the model.

[0102] Correlation analysis: Use methods such as Pearson correlation coefficient and Spearman rank correlation coefficient to analyze the linear or nonlinear relationship between features, identify potential collinearity problems, and provide guidance for model construction.

[0103] 5. Fitting model weights

[0104] Select a suitable machine learning model (such as logistic regression, support vector machine, neural network, etc.) according to the nature of the problem, and use cross-validation, grid search and other techniques to optimize model parameters to ensure the generalization ability of the model.

[0105] 6. Tip Engineering and Model Fine-tuning

[0106] Hint design: Based on feature profiling and correlation analysis, design targeted hints or rules to guide the model to focus on key features and improve diagnostic accuracy.

[0107] For example: when the completion rate of the selected indicator ranks lower than the top 3 in the same group, the problem description is: [Indicator name] + "in" + [branch type] + "relatively backward"; when the completion rate of all indicators ranks 1st, 2nd, and 3rd in the same group, the problem description is: [Indicator name 1, 2,...] + in [branch type] + "overall good".

[0108] Model fine-tuning: Based on feedback from prompt engineering and actual business users, fine-tune the model, such as adjusting the learning rate, adding regularization terms, etc., to further optimize model performance.

[0109] 7. Summarize multiple results and infer diagnostic conclusions

[0110] Based on the above analysis, logical reasoning and expert experience are used to summarize diagnostic conclusions, identify the root causes of the problem and provide solution recommendations.

[0111] 8.LLM (Large Language Model) outputs diagnostic conclusions

[0112] The summarized diagnostic conclusions are input into a pre-trained large-scale language model (such as Tongyi Qianwen, BERT, etc.), and its natural language generation capabilities are used to output the final diagnostic report in clear and accurate language.

[0113] Report optimization: According to the first draft output by LLM, language polishing and format adjustment are carried out to ensure the professionalism and readability of the report and meet user needs. For example, "the branch company is in the downstream, and its diagnosed operating problems are mainly that the "main mobile network income, 100 yuan equivalent" is poor.

[0114] A diagnostic report is generated based on the user's diagnostic direction, and images can be downloaded, solving the problem of long waiting periods for business personnel to customize pages using traditional development models. The efficiency is significantly improved from 15 days to 5 minutes.

[0115] Furthermore, the method further comprises:

[0116] Based on the user's questions and answers, we generate the current situation analysis and problem analysis diagnosis structure, identify the indicators of the revenue topic recommended by Graph Database for the user, and assemble the accumulated indicator visualization assets to generate the visualization results of the current situation analysis, including the key indicator cards of the revenue category. At the same time, based on the multi-faceted problem analysis and diagnosis direction, we generate corresponding chart cards for the indicators selected by the user according to different business types. The key point is to generate a structure that includes the user's selected structure and the indicators under each hierarchical structure, and quickly generate cards that meet the business presentation according to the indicator type. For example, for completion probability indicators, we generate ring charts, bar charts and other style cards, and for more complex multi-indicators, we generate rich text indicator cards.

[0117] Through the above method, by managing indicator assets through Graph Database, using real-time interactive dialogue to accurately capture user intentions for rapid diagnosis and summarizing through LLM diagnosis, business personnel can conduct all-round diagnosis based on indicators more flexibly, generate diagnostic reports more efficiently, and make decisions faster in a flexible and ever-changing market.

[0118] The disclosed embodiment is based on the indicator assets accumulated by the enterprise, and quickly finds related indicators through Graph Database; using NLP technology, business personnel recommend diagnostic directions and diagnostic indicators to users through dialogue, combined with the indicator asset model of Graph Database; using AIGC technology, a diagnostic report is generated based on the user's diagnostic direction, and pictures can be downloaded, which solves the problem of long waiting period for business personnel to customize pages using traditional development models, and reduces the generation of diagnostic reports from 15 days to 5 minutes, and the efficiency is improved by more than 15 times.

[0119] The second embodiment of the present disclosure also provides a method for generating an operating report based on an enterprise index asset, and a system implementation for generating an operating report based on an enterprise index asset, such as Figure 2 As shown, the system includes three modules: full life cycle management of indicator assets, real-time interactive recommendation indicators for users, and diagnostic reports.

[0120] The indicator asset full life cycle management module is used to create and manage the enterprise's indicators, store indicators through the Graph Database mode, and use BiLSTM+CRF deep learning and RNN recurrent neural network to realize the vectorization and multi-tuple of enterprise indicators; the user real-time interactive recommendation indicator module is used for users to independently select diagnostic directions and indicators through intelligent question-and-answer format; the diagnostic report module is used to dynamically generate diagnostic directions and results in the real-time interactive recommendation indicator module, and use feature data analysis, expert model, correlation analysis, fitting model weights, model fine-tuning and LLM for diagnostic summary, quickly generate visual diagnostic reports for users and download pictures. The effect is: through GraphDatabase to manage indicator assets, use real-time interactive dialogue to accurately capture user intentions for rapid diagnosis and use LLM diagnostic summary, so that business personnel can make all-round diagnosis based on indicators more flexible, generate diagnostic reports more efficiently, and make decisions faster in a flexible and changeable market.

[0121] Functionally, Figure 2 As shown, the indicator asset full life cycle management module is used for intuitive storage of enterprise indicator graph association relationships, and includes three functions: indicator market, indicator creation and listing, and indicator graph relationship model.

[0122] Specifically, users create derived indicators by dragging and dropping through the basic indicator interface, and record the parent-child relationship between indicators, and put indicators on and off the shelves according to business needs. All business indicators are stored in the graph database. When users use a certain indicator to query, they can quickly find related indicators. For details, please refer to the business indicator association diagram in Table 1.

[0123] Table 1: Schematic list of indicator association business

[0124]

[0125]

[0126] The user real-time interactive recommendation indicator module is used to identify the user's intention to recommend diagnostic directions and indicators in an interactive manner of intelligent question and answer.

[0127] Specifically, users input diagnostic requirements through the question-and-answer interface, such as "Help me generate a branch company revenue and operation diagnostic report". Through corpus training, user intent recognition and slot extraction, the organizational structure is "a certain branch company". First, the BERT error correction model, sensitive word filtering, thinking, calling indicator correlation relationships, and outputting recommended diagnostic indicators are used.

[0128] The diagnostic report module is used to output a diagnostic summary using LLM according to the diagnostic indicators selected by the user, and to generate a visual diagnostic report using AIGC technology and indicator cards.

[0129] Specifically, the diagnosis summary module generates results for the interactive recommendation indicators, and the report generation module quickly generates a visual report for the diagnosis content. At the same time, users can download the diagnosis report in the form of pictures to the local computer for use in scenarios such as work reports.

[0130] Reference Figure 3 As shown, the method includes:

[0131] S101 manages the business indicators of the enterprise through the indicator asset full life cycle management module. It stores indicators, as well as parent-child and business-related indicators, through the GraphDatabase. As the business changes and the number of indicators increases, it dynamically discovers the indicator association relationship, creates derivative indicators from basic indicators, and manages the listing and delisting of indicators.

[0132] Through BiLSTM+CRF deep learning and RNN recurrent neural network, enterprise indicators are vectorized and multi-tuples are used to solve the problems of insufficient accuracy and low efficiency of traditional methods when processing complex and nonlinear enterprise indicator data.

[0133] First, we preprocessed the enterprise indicator data, including data cleaning, normalization and other steps to ensure the quality and consistency of the data.

[0134] Then, we build a BiLSTM network to capture the temporal dependencies in the data. Through the bidirectional LSTM unit, the model can consider both past and future information at the same time, so as to understand the data more accurately. Next, we use the output of BiLSTM as the input of CRF, and use CRF to optimize the sequence labeling to obtain a more accurate vectorized representation. Including transfer matrix initialization, decoding algorithm, label constraint conditions to complete the CRF layer processing. Specifically including:

[0135] Transfer matrix initialization: Use uniform distribution or initialization method based on prior knowledge to ensure the stability of model training. For example, "Table 1: Business Schematic List of Indicator Association Relationships" in this article is used as prior knowledge of enterprise indicators for training.

[0136] Decoding algorithm: The Viterbi algorithm is used for dynamic programming decoding to find the globally optimal label sequence.

[0137] Label constraints: Label constraints are introduced in the CRF layer (such as label1, label2, label3 must be of the same entity category in "B-label1I-label2 I-label3...") to further improve the accuracy of sequence labeling.

[0138] At the same time, for the multi-group data in corporate indicators, we designed a special RNN structure and introduced additional gating mechanisms and attention mechanisms to enable the model to better handle such complex data.

[0139] In the process of implementation, technical difficulties such as data imbalance and model overfitting were effectively solved by introducing regularization technology, adopting cross-validation, and adjusting model parameters, thus improving the performance and stability of the model.

[0140] Experimental results show that compared with the existing technology, our method has significant improvements in key indicators such as accuracy, recall and F1 score.

[0141] S102, refer to Figure 4 , through the user real-time interactive recommendation indicator module, taking the "main mobile network income" indicator as an example, by abstracting the indicator as a vertex and the relationship between indicators as an edge, the deep chain query and analysis of the indicator relationship can be completed. When it is necessary to query the relationship indicator of the main mobile network income indicator, the traversal framework of Neo4j is used to calculate the route;

[0142] The ParceRouteCalculator class is used to calculate the "main mobile network income" indicator, from entering the network to knowing all the necessary logic for reaching other business indicator routes, and finally quickly find the related indicators "actual revenue increase, hundred yuan equivalent, and number of 5G development users."

[0143] S103, refer to Figure 5 Through the diagnosis report module, according to the indicators selected by the user, the feature data is input, and expert models such as feature clustering and Shapely Value are used to perform feature profiling, correlation analysis, and fitting model weights. Combined with prompt engineering and model fine-tuning, multiple results are finally summarized and diagnostic conclusions are inferred. The diagnostic conclusions are output through LLM, such as "the branch is downstream, and its diagnosed operating problems are mainly "main mobile network income, 100 yuan equivalent" is poor" (this diagnostic conclusion is an example).

[0144] (1) Input feature data processing

[0145] Data preprocessing: First, clean the input indicators and the characteristic data of the indicator daily or monthly account period data, including removing missing values, detecting and processing outliers, converting data types, etc., to ensure data quality. Then, perform feature standardization or normalization to eliminate the influence of different feature dimensions and improve the efficiency and accuracy of model training.

[0146] For example, in Table 2, the index value of the main operating income is a missing value, and the index value of the income from new development work orders of the mobile network is an abnormal value, which requires data preprocessing when inputting feature data.

[0147] Table 2: Obtained indicator data

[0148] Indicator Code Indicator name 2024-01 001 New development work order income from mobile network 19000987 Yuan 002 Broadband new development work order income 123 yuan 003 5G new development work order income 233 yuan 004 Main revenue

[0149] Feature selection: Use statistical methods (such as variance selection and correlation coefficient analysis) or machine learning algorithms (such as random forest importance scoring) to perform preliminary feature screening, retain features that are highly relevant to the diagnostic target, reduce redundant information, and improve model performance.

[0150] (2) Feature clustering

[0151] K-means, DBSCAN, hierarchical clustering and other algorithms are used to cluster the preprocessed features, and similar or related features are grouped into the same group for subsequent analysis. In the clustering process, it is necessary to select appropriate distance metrics (such as Euclidean distance, Manhattan distance) and clustering parameters according to the data distribution characteristics.

[0152] (3) Shapely Value Calculation

[0153] The importance of features is calculated using efficient algorithms such as TreeSHAP and KernelSHAP. The marginal contribution of each feature in different combinations is iteratively calculated to finally obtain the Shapley Value of each feature.

[0154] 4. Feature profiling and correlation analysis

[0155] Feature portrait construction: Based on the clustering results and Shapley Value, a detailed portrait is constructed for each feature, including the importance of the feature, the degree of correlation with other features, and the way it contributes to the model.

[0156] Correlation analysis: Use methods such as Pearson correlation coefficient and Spearman rank correlation coefficient to analyze the linear or nonlinear relationship between features, identify potential collinearity problems, and provide guidance for model construction.

[0157] 5. Fitting model weights

[0158] Select a suitable machine learning model (such as logistic regression, support vector machine, neural network, etc.) according to the nature of the problem, and use cross-validation, grid search and other techniques to optimize model parameters to ensure the generalization ability of the model.

[0159] 6. Tip Engineering and Model Fine-tuning

[0160] Hint design: Based on feature profiling and correlation analysis, design targeted hints or rules to guide the model to focus on key features and improve diagnostic accuracy.

[0161] For example: when the completion rate of the selected indicator ranks lower than the top 3 in the same group, the problem description is: [Indicator name] + "in" + [branch type] + "relatively backward"; when the completion rate of all indicators ranks 1st, 2nd, and 3rd in the same group, the problem description is: [Indicator name 1, 2,...] + in [branch type] + "overall good".

[0162] Model fine-tuning: Based on feedback from prompt engineering and actual business users, fine-tune the model, such as adjusting the learning rate, adding regularization terms, etc., to further optimize model performance.

[0163] 7. Summarize multiple results and infer diagnostic conclusions

[0164] Based on the above analysis, logical reasoning and expert experience are used to summarize diagnostic conclusions, identify the root causes of the problem and provide solution recommendations.

[0165] 8.LLM (Large Language Model) outputs diagnostic conclusions

[0166] The summarized diagnostic conclusions are input into a pre-trained large-scale language model (such as Tongyi Qianwen, BERT, etc.), and its natural language generation capabilities are used to output the final diagnostic report in clear and accurate language.

[0167] Report optimization: According to the first draft output by LLM, language polishing and format adjustment are carried out to ensure the professionalism and readability of the report and meet user needs. For example, "the branch company is in the downstream, and its diagnosed operating problems are mainly that the "main mobile network income, 100 yuan equivalent" is poor.

[0168] Furthermore, the method also includes generating a current situation analysis, a problem analysis and a diagnostic direction structure based on the user's questions and answers, identifying the indicators of the income topic recommended by the Graph Database for the user, and generating a visualization result of the current situation analysis, including a key indicator card of the income category.

[0169] Through the above method, by managing indicator assets through Graph Database, using real-time interactive dialogue to accurately capture user intentions for rapid diagnosis and summarizing through LLM diagnosis, business personnel can conduct all-round diagnosis based on indicators more flexibly, generate diagnostic reports more efficiently, and make decisions faster in a flexible and ever-changing market.

[0170] The third embodiment of the present disclosure also provides a system for generating an operating report based on enterprise index assets, such as Figure 6 As shown, the system comprises:

[0171] A storage module 11 is configured to store the business indicator data of the enterprise through a graph database and store the association relationship between the indicators;

[0172] The diagnosis module 12 is configured to identify the user's intention in an interactive manner of intelligent question and answer and recommend a diagnosis direction and a diagnosis indicator according to the user's intention;

[0173] The generation module 13 is configured to output the diagnosis conclusion using LLM according to the recommended diagnosis indicators, and to generate a visual diagnosis report using AIGC technology and indicator graphical assets.

[0174] Furthermore, the system also includes a creation module 14;

[0175] The creation module 14 is configured to create derived indicators for the business indicators of the enterprise through basic indicator interface operations, and record the parent-child relationship between indicators, and put indicators on and off the shelves according to business needs, and store all business indicators in the graph database.

[0176] Furthermore, the diagnosis module 12 is specifically configured as follows:

[0177] Obtain the diagnostic requirements input by the user through the question-answering interface and identify the user's intention through the big model; and,

[0178] User intent is processed through the BERT error correction model, sensitive words are filtered, indicator association queries are called, and recommended diagnostic indicators are output.

[0179] Furthermore, the calling of indicator association relationship query and output of recommended diagnostic indicators includes:

[0180] When you need to query the correlation indicators of the corresponding indicators of user intent, use the Neo4j traversal framework to calculate the route, including:

[0181] Filter the path based on time, given the path from the starting node of the traversal to the current node, and return the connection that can be traversed further down, input parameters, the path extending from the starting node of the traversal to the current traversal node and the branch status provided by the client, and output diagnostic indicators.

[0182] Furthermore, the generating module 13 is specifically configured as follows:

[0183] According to the recommended diagnostic indicators, feature data is input, and feature profiling, correlation analysis, causal analysis, and model weight fitting are performed through feature clustering and Shapely Value expert model. In combination with prompt engineering and model fine-tuning, multiple results are finally summarized, diagnostic conclusions are inferred, and the diagnostic conclusions are output through LLM.

[0184] The system for generating operating reports based on enterprise indicator assets in the embodiment of the present disclosure is used to implement the method for generating operating reports based on enterprise indicator assets in method embodiment 1 and embodiment 2, so the description is relatively simple. For details, please refer to the relevant description in the previous method embodiment, which will not be repeated here.

[0185] In addition, if Figure 7 As shown, the fourth embodiment of the present disclosure further provides an electronic device, including a memory 100 and a processor 200, wherein the memory 100 stores a computer program, and when the processor 200 runs the computer program stored in the memory 100, the processor 200 executes the above-mentioned various possible methods.

[0186] The memory 100 is connected to the processor 200. The memory 100 may be a flash memory, a read-only memory or other memory, and the processor 200 may be a central processing unit or a single-chip microcomputer.

[0187] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the above-mentioned various possible methods.

[0188] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), Digital Video Disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0189] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.< / relationship>

Claims

1. A method for generating an operating report based on enterprise indicator assets, characterized in that: The method comprises: The business indicator data of the enterprise is stored in the graph database, and the relationship between the indicators is stored; Through intelligent question-answering interaction, it identifies user intent and recommends diagnostic directions and indicators based on user intent. According to the recommended diagnostic indicators, the large language model LLM is used to output the diagnostic conclusion, and the artificial intelligence content generation AIGC technology and indicator graphical assets are used to generate a visual diagnostic report.

2. The method according to claim 1, characterized in that The method further comprises: For the business indicators of an enterprise, derivative indicators are created through the basic indicator interface operation, and the parent-child relationship between indicators is recorded. Indicators are put on and taken off the shelves according to business needs, and all business indicators are stored in the graph database.

3. The method according to claim 1, characterized in that The interactive method of intelligent question and answer is used to identify the user's intention and recommend the diagnosis direction and diagnosis indicators according to the user's intention, including: Obtain the diagnostic requirements input by users through the question-and-answer interface and identify user intent through a large model; The user intention is processed through the bidirectional encoder representation method BERT error correction model, sensitive words are filtered, indicator association relationship query is called, and recommended diagnostic indicators are output.

4. The method according to claim 1, characterized in that The calling indicator association relationship query and outputting the recommended diagnostic indicators include: When you need to query the correlation indicators of the corresponding indicators of user intent, use the Neo4j traversal framework to calculate the route, including: Filter the path based on time, given the path from the starting node of the traversal to the current node, and return the connection that can be traversed further down, input parameters, the path extending from the starting node of the traversal to the current traversal node and the branch status provided by the client, and output diagnostic indicators.

5. The method according to claim 1, characterized in that The method of outputting a diagnosis conclusion using LLM according to the recommended diagnosis indicators includes: According to the recommended diagnostic indicators, feature data is input, and feature profiling, correlation analysis, causal analysis, and model weight fitting are performed through feature clustering and Shapely Value expert model. In combination with prompt engineering and model fine-tuning, multiple results are finally summarized, diagnostic conclusions are inferred, and the diagnostic conclusions are output through LLM.

6. A system for generating operating reports based on enterprise index assets, characterized in that: The system comprises: A storage module is configured to store the business indicator data of the enterprise through a graph database and store the association relationship between the indicators; A diagnosis module, which is configured to identify user intent in an interactive manner of intelligent question and answer and recommend diagnosis directions and diagnostic indicators based on the user intent; The generation module is configured to output the diagnosis conclusion using LLM according to the recommended diagnosis indicators, and to generate a visual diagnosis report using AIGC technology and indicator graphical assets.

7. The system according to claim 6, characterized in that The system also includes a creation module; The creation module is set to create derived indicators for the business indicators of the enterprise through basic indicator interface operations, and record the parent-child relationship between indicators. The indicators are put on and taken off the shelves according to business needs, and all business indicators are stored in the graph database.

8. The method according to claim 6, characterized in that The diagnostic module is specifically configured as follows: Obtain the diagnostic requirements input by the user through the question-answering interface and identify the user's intention through the big model; and, User intent is processed through the BERT error correction model, sensitive words are filtered, indicator association queries are called, and recommended diagnostic indicators are output.

9. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method for generating an operating report based on enterprise indicator assets as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for generating an operating report based on enterprise indicator assets according to any one of claims 1 to 6 is implemented.

Images