Report generation method and device, equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of data processing, in particular to a report generation method and device, equipment and a storage medium, and aims to quickly generate a work report corresponding to each work scene. The method comprises the steps that scene data of a target working scene are received through a pre-trained scene recognition model, a working report template corresponding to the target working scene is obtained, and the scene data comprise scene feature data, scene task data and scene demand data; performing content filling on the work report template according to a preset rule through a decision tree classifier to obtain a first work report; performing task arrangement on the work tasks corresponding to the first work report to obtain a second work report; and performing technical supplement on the second work report according to cross-domain expert suggestions to obtain a third work report.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and in particular, to a report generation method, apparatus, device and storage medium. Background Art

[0002] In many current work fields, such as the telecommunications field, business analysis, scenario analysis and technical assessment are required when working on various work scenarios. In the prior art, technical personnel are required to analyze and assess each work scenario.

[0003] In the existing technology, when technical personnel analyze and evaluate various work scenarios, due to the difficulty in transferring and updating expert resources and professional knowledge, and the difficulty in cross-professional coordination and cooperation, a large amount of manual intervention is required to make subjective choices and judgments, which consumes a lot of manpower and time costs. Summary of the invention

[0004] The embodiments of the present application provide a report generation method, apparatus, device and storage medium, which are intended to quickly generate work reports corresponding to various work scenarios.

[0005] A first aspect of an embodiment of the present application provides a report generation method, the method comprising: Receive scene data of a target work scene through a pre-trained scene recognition model, and obtain a work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; Filling the work report template with content according to preset rules through a decision tree classifier to obtain a first work report; Arrange tasks corresponding to the first work report to obtain a second work report; Based on the opinions of cross-domain experts, the second working report was technically supplemented to obtain the third working report.

[0006] Optionally, the method further comprises: Receiving analysis opinions corresponding to the third work report; Based on the analysis opinions, parameters of the models and algorithms used in the work report generation process are optimized.

[0007] Optionally, the training step of the scene recognition model includes: Collect work scenario data; Adding corresponding scene labels to the work scene data to obtain a training data set; The training data set is input into the model to be trained, and the model is trained to obtain the scene recognition model.

[0008] Optionally, receiving scene data of a target work scene through a pre-trained scene recognition model to obtain a work report template corresponding to the target work scene includes: Analyzing the scene data through the scene recognition model to determine the scene type and target task corresponding to the target work scene; Generate the corresponding work report template according to the scenario type and the target task.

[0009] Optionally, the filling of the work report template with content according to preset rules by a decision tree classifier to obtain a first work report includes: Classifying the scene features corresponding to the target work scene by the decision tree classifier to obtain a corresponding decision chain; According to the decision chain, a support plan corresponding to the target work scenario is generated; The work report template is filled in using the support solution to obtain the first work report.

[0010] Optionally, the arranging the work tasks corresponding to the first work report to obtain the second work report includes: According to the expert opinions corresponding to the first work report, an importance assessment is performed on each task in the first work report to obtain an importance weight and a task time corresponding to each task; According to the importance weight and the task time, each of the work tasks is scheduled to obtain the second work report.

[0011] Optionally, the second working report is technically supplemented according to the cross-domain expert opinions to obtain a third working report, including: Obtaining the cross-domain expert opinion on the second work report; Inputting the cross-domain expert opinions into a vector model to obtain a vector corresponding to the cross-domain expert opinions; Determining a technical opinion for each of the tasks according to the vectors corresponding to the cross-domain expert opinions; The technical opinions are added to the second working report to obtain the third working report.

[0012] A second aspect of an embodiment of the present application provides a report generating device, the device comprising: A scene recognition module, used to receive scene data of a target work scene through a pre-trained scene recognition model, and obtain a work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; A decision classification module, used to fill the work report template with content according to preset rules through a decision tree classifier to obtain a first work report; A task scheduling module, used to schedule the work tasks corresponding to the first work report to obtain a second work report; The technical supplement module is used to provide technical supplements to the second working report based on cross-domain expert opinions to obtain a third working report.

[0013] Optionally, the device further comprises: An opinion receiving module, used for receiving analysis opinions corresponding to the third work report; The parameter optimization module is used to optimize the parameters of the model and algorithm used in the work report generation process according to the analysis opinions.

[0014] Optionally, the training step of the scene recognition model includes: Collect work scenario data; Adding corresponding scene labels to the work scene data to obtain a training data set; The training data set is input into the model to be trained, and the model is trained to obtain the scene recognition model.

[0015] Optionally, the scene recognition module includes: A type determination submodule, used to analyze the scene data through the scene recognition model to determine the scene type and target task corresponding to the target work scene; The template generation module is used to generate the corresponding work report template according to the scenario type and the target task.

[0016] Optionally, the decision classification module includes: A decision chain acquisition submodule, used to classify the scene features corresponding to the target work scene through the decision tree classifier to obtain a corresponding decision chain; A support solution acquisition submodule, used to generate a support solution corresponding to the target work scenario according to the decision chain; The first work report acquisition submodule is used to fill the work report template with the support solution to obtain the first work report.

[0017] Optionally, the task scheduling module includes: A task evaluation submodule, configured to evaluate the importance of each task in the first work report according to the expert opinions corresponding to the first work report, and obtain the importance weight and task time corresponding to each task; The second work report acquisition submodule is used to schedule each of the work tasks according to the importance weight and the task time to obtain the second work report.

[0018] Optionally, the technical supplement module includes: A cross-domain expert opinion acquisition submodule, used to obtain the cross-domain expert opinion for the second work report; A vector acquisition submodule, used for inputting the cross-domain expert opinion into a vector model to obtain a vector corresponding to the cross-domain expert opinion; A technical opinion determination submodule, used to determine the technical opinion for each of the tasks according to the vectors corresponding to the cross-domain expert opinions; The third working report obtaining submodule is used to add the technical opinions into the second working report to obtain the third working report.

[0019] A third aspect of an embodiment of the present application provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps in the method described in the first aspect of the present application are implemented.

[0020] A fourth aspect of an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the method described in the first aspect of the present application are implemented.

[0021] The report generation method provided by the present application is adopted to receive the scene data of the target work scene through the pre-trained scene recognition model, and obtain the work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; through the decision tree classifier, the work report template is filled with content according to preset rules to obtain a first work report; the work tasks corresponding to the first work report are arranged to obtain a second work report; Based on the opinions of cross-domain experts, the second working report was technically supplemented to obtain the third working report.

[0022] In this method, when generating a work report for a target scenario, the target scenario is first identified through a scene recognition model to generate a corresponding work report template. The scene recognition model can identify the scene type of the target work scenario, and then generate a corresponding work report template based on the scene type, thereby accurately identifying the work scenario. The decision tree classifier is then used to fill in the content of the work report template according to preset rules to obtain a basic work report, and the work tasks in the first work report are task-organized to obtain a second work report, thereby automatically arranging the work tasks. Finally, the second work report is technically supplemented based on cross-domain expert opinions to obtain a third work report. Using the above method, a corresponding work report is automatically generated for the target work scenario, so that the generation process of the work report is standardized in the region, saving a lot of manpower and material resources, and improving the generation efficiency of the work report, thereby efficiently guiding the development and progress of the work. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0024] Figure 1 is a flow chart of a report generation method proposed in an embodiment of the present application; Figure 2 It is a schematic diagram of a report generation method proposed in an embodiment of the present application; Figure 3 It is a business application matching diagram proposed in an embodiment of the present application; Figure 4 It is a schematic diagram of the overall flow of the working method proposed in one embodiment of the present application; Figure 5 is a schematic diagram of a report generation method and device proposed in an embodiment of the present application; Figure 6 It is a schematic diagram of an electronic device proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] refer to Figure 1 , Figure 1: is a flow chart of a report generation method proposed in an embodiment of the present application. Figure 1 As shown, the method comprises the following steps: S11: Receive scene data of a target work scene through a pre-trained scene recognition model, and obtain a work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data.

[0027] In this embodiment, the scene recognition model is trained by a deep learning model and is used to identify the scene type of the target work scene. The target work scene is a work scene that requires technicians to perform a series of tasks. The work report is a report issued when performing work tasks for the target scene, which specifies which specific tasks need to be performed for the target scene, which equipment is required to perform the tasks, which technologies need to be used, etc., and has a guiding and planning role in performing work tasks. The work report template is the basic framework of a work report, which specifies various items and the content that needs to be filled in.

[0028] Scenario feature data is data that characterizes the characteristics of the target work scenario, scenario task data is data on which specific tasks need to be performed in the target work scenario, and scenario requirement data is data on the specific effects that need to be achieved by performing work tasks in the target work scenario.

[0029] In this embodiment, when generating a work report for a target scene, it is first necessary to receive the scene data of the target work scene through a pre-trained scene recognition model. After receiving the scene data, the scene recognition model identifies the scene data, and then determines the scene type and scene task of the target work scene, and then generates a corresponding work report template.

[0030] In this embodiment, the specific steps of receiving the scene data of the target work scene through the pre-trained scene recognition model and obtaining the work report template corresponding to the target work scene include: S11-1: Analyze the scene data through the scene recognition model to determine the scene type and target task corresponding to the target work scene.

[0031] In this embodiment, the scene data is first analyzed by the scene recognition model to determine the scene type and target task corresponding to the target work scene. The scene recognition model is a model trained based on scene characteristics, task requirements, and previous work cases, and has the ability to recognize and perceive work scenes. When the scene recognition model receives the scene data, the corresponding scene type and target task can be obtained based on the scene data.

[0032] For example, the scenario type may be a telecommunications work scenario, and the target task may be a fault handling task, a resource change task, and the like.

[0033] S11-2: Generate the corresponding work report template according to the scenario type and the target task.

[0034] In this embodiment, after determining the type of scenario and the target task, a basic work report template is determined based on previous work cases, and the template will specify the work tasks that need to be performed for the target work scenario.

[0035] In another embodiment of the present application, the corresponding template can be searched in the template library according to the scene type and the ready-made template can be used directly.

[0036] In another embodiment of the present application, the training step of the scene recognition model includes: S21: Collect work scenario data.

[0037] In this embodiment, when training the scene recognition model, work scene data is first collected. The work scene data includes the characteristics of each scene, the task data performed in each scene, and previous work report cases of the work scene.

[0038] S22: Add corresponding scene labels to the work scene data to obtain a training data set.

[0039] In this embodiment, a corresponding scene label is added to each item in the work scene data to obtain the corresponding training data.

[0040] S23: Inputting the training data set into the model to be trained, training the model, and obtaining the scene recognition model.

[0041] In this embodiment, the collected training data set is input into the model to be trained, and the model is trained to obtain a scene recognition model.

[0042] In this embodiment, based on the characteristics, tasks and requirements of the work scene and previous work cases, a deep learning model, such as a recurrent neural network (RNN), is used to train the collected work scene data, thereby realizing the perception and recognition of different work environments. The deep learning model can process complex, high-dimensional data, automatically extract and learn scene characteristics, and thus form a basic model for supporting work scenes.

[0043] In this embodiment, if there is a set of training data ,in The input work scenario characteristics (such as information communication, technical exchange, program presentation, project order signing, solution formulation, project delivery services, etc.), is the scene label. Using deep learning model Train it, are model parameters. The goal is to minimize the loss function: (1) in, is the loss function, which means that when choosing parameters The average loss of the model on the training data when . The goal is to find Smallest . is a normalization factor that averages the losses of all samples. Usually represents a single sample So we need to average them to get the average loss of the whole dataset. Is a summation symbol, indicating that for all training samples Perform the summation. is the corresponding true label. The goal is to find the best parameters through learning , so that the model best fit to the data.

[0044] Is a single sample The loss function is The model is parameterized Next pair input The predicted output is It’s a real label. The function is used to calculate the difference or loss between the predicted output and the true label. The basic idea of ​​this formula is to minimize the loss function To find the best parameters This process is usually achieved through gradient descent or other optimization algorithms. During the optimization process, to reduce the value of , until a satisfactory solution is reached.

[0045] In this embodiment, in the training process of the deep learning model, a recurrent neural network (RNN) can be used as the model structure. First, a training data set is required. ,in is the input data of the working scenario characteristics, is the corresponding scene label. Then, define the model structure , you can choose RNN variants such as LSTM (Long Short-Term Memory Network) or GRU (Gated Recurrent Unit).

[0046] In this embodiment, the loss function is used The model is trained using commonly used optimization algorithms, including stochastic gradient descent (SGD), Adam, etc. In each training iteration, the model parameters are updated through the back propagation algorithm. , so that the loss function is minimized. Cross entropy can be used as the loss function during training. , and regularization methods can be used to prevent overfitting.

[0047] In this embodiment, the training data set Including input data and the corresponding labels .in, is the input data of the working scenario characteristics, is the corresponding scene label. Assuming the input data is a vector of dimension , label is a category label.

[0048] In this embodiment, a long short-term memory network (LSTM) is used as a variant of RNN. LSTM has good memory capacity and is suitable for processing time series data. Model structure It can be represented as a deep neural network containing multiple layers of LSTM units, where are model parameters.

[0049] In this embodiment, during the training process, it is necessary to define a loss function To measure the difference between the model prediction output and the true label. The commonly used loss function is the cross entropy loss function, and its mathematical expression is: = (2) is an average operation over all samples. For each sample i, its loss is calculated, then the losses of all samples are added together and finally divided by the number of samples n. :When the true label The loss of this part is This means that when the model predicts the probability When it is close to 1, the loss will be small; When it approaches 0, the loss will be huge. :When the true label The loss of this part is This means that when the model predicts the probability When it is close to 0, the loss will be small; When it approaches 1, the loss will be large.

[0050] In this embodiment, during the training process, an optimization algorithm is used to minimize the loss function. Common optimization algorithms include stochastic gradient descent (SGD) and Adam. The gradient is calculated and the model parameters are updated through the back propagation algorithm. , so that the loss function is minimized.

[0051] In this embodiment, stochastic gradient descent (SGD) is used in training: SGD is a basic optimization algorithm that uses the gradient of a single sample to update the model parameters in each iteration. Its parameter update rule is: (3) in, is the learning rate, is the loss function with respect to the parameters gradient.

[0052] Adam optimization algorithm: Adam is an adaptive learning rate optimization algorithm that combines the characteristics of momentum gradient descent and adaptive learning rate adjustment. It uses the first-order moment estimation and second-order moment estimation of the gradient to adjust the learning rate. Its parameter update rule is: (4) (5) (6) (7) (8) in, and are the first-order moment estimate and the second-order moment estimate of the gradient, and is the decay rate, is the learning rate, is a small constant added for numerical stability, is the number of training iterations.

[0053] In this embodiment, when training a deep learning model, in order to prevent overfitting, a regularization method can be used to limit the size of the model parameters. One of the common regularization methods is L2 regularization. Its purpose is to penalize model complexity by adding the sum of squares of parameter weights to the loss function. The form of the L2 regularization term is: = (9) in, is the original loss function, is the regularization parameter, which is used to control the weight of the regularization term. Represents the parameters of the model.

[0054] In the above formula, is a regularization term that penalizes the model parameters so that the parameter values ​​are not too large, thereby preventing the model from overfitting the training data. The value of is usually a small positive number, and a suitable value can be selected by cross-validation and other methods. During the training process, the loss function with regularization term is minimized. , the model parameters can be updated by an optimization algorithm such as stochastic gradient descent or Adam , so as to minimize the loss function. Through L2 regularization, the complexity of the model can be effectively controlled and the generalization ability of the model can be improved, thereby improving the performance of the model while preventing overfitting.

[0055] In this embodiment, through the above training process, a scene recognition model is obtained to effectively perceive and recognize the target working scene.

[0056] S12: Filling the work report template with content according to preset rules through a decision tree classifier to obtain a first work report.

[0057] In this embodiment, the decision tree classifier is a classification tool based on the decision tree machine learning classification algorithm, which can select the most appropriate support scheme for the target work scenario according to predefined rules or rules obtained through learning. The preset rules are pre-made decision and classification rules, and the corresponding rules can be adjusted according to different work scenarios.

[0058] In this embodiment, the work report template is filled with content according to preset rules through a decision tree classifier, and the corresponding support plan is filled into the work report template. The work report template after content filling is called the first work report. After identifying the work scenario, the system uses a decision tree machine learning classification algorithm to select the most appropriate support plan for the current work scenario according to predefined rules or rules obtained through learning. For example, according to the scenario type of telecommunications work: information communication, technical exchange, program presentation, project delivery service, solution formulation, etc., the system can automatically adjust and optimize the operating parameters, thereby filling in the basic template of the work items, adapting the specific details of the attributes, continuously performing content analysis and work scenario adaptation, and then gradually filling in the content of the basic model, thereby improving the work analysis report.

[0059] In this embodiment, the specific steps of filling the work report template with content according to preset rules by using a decision tree classifier to obtain the first work report include: S12-1: Classify the scene features corresponding to the target work scene through the decision tree classifier to obtain a corresponding decision chain.

[0060] In this embodiment, the decision chain is a directed acyclic graph, each node in the decision chain represents a decision, and each edge represents the relationship between decisions. By optimizing the decision chain, the optimal solution to the problem can be obtained.

[0061] In this embodiment, the scene features in the target work scene are input into the decision input classifier, and the decision tree classifier classifies the scene features according to preset rules to obtain a corresponding decision chain. The decision chain includes the corresponding decision made for each task in the target scene.

[0062] S12-2: Generate a support plan corresponding to the target work scenario based on the decision chain.

[0063] In this embodiment, after obtaining the decision chain corresponding to the target work scenario, a support plan corresponding to the target work scenario is generated, and the support plan is the optimal decision for each work task in the decision chain.

[0064] S12-3: Fill the work report template with the support solution to obtain the first work report.

[0065] In this embodiment, after obtaining the support plan for each work task in the target work scenario, the work report template is filled to obtain the first work report.

[0066] In this embodiment, when analyzing and scheduling the work scene, a decision tree classifier is used to select the most appropriate support solution for the current work scene. It is a decision tree classifier, which is a function that takes the input features To predict the results ,in is a set of all possible categories or support solutions. Decision trees split nodes to find the best decision rule by maximizing information gain.

[0067] Where F(X) is a decision tree classifier, based on the given input features , predict the result .

[0068] Y is the set of all possible support solutions, that is, the output category of the classifier. Information gain: When splitting a node, select the feature that can maximize the information gain for splitting.

[0069] In this embodiment, the process of building a decision tree generally includes steps such as feature selection, node splitting and pruning.

[0070] Feature selection can use information gain or Gini coefficient to evaluate the importance of features and determine the best split features. The node splitting process selects the best split point by maximizing information gain or other indicators to construct the nodes of the decision tree.

[0071] Pruning is to prevent overfitting. The decision tree can be pruned by pre-pruning or post-pruning to improve the generalization ability of the model.

[0072] The final decision tree classifier can predict the corresponding support plan based on the input feature X.

[0073] In this embodiment, the construction process of the decision tree classifier involves the steps of feature selection, node splitting and pruning: Feature selection is a key step in decision tree construction. Its purpose is to evaluate the importance of each feature to classification in order to determine the best classification feature. Commonly used indicators include information gain and Gini coefficient.

[0074] Information gain: Information gain evaluates the importance of a feature by calculating the difference between the entropy of the parent node and the weighted average entropy of the child nodes. The information gain calculation formula is: [{Information Gain} = {Entropy of the parent node} - {Weighted average entropy of the child nodes}] Gini coefficient: The Gini coefficient is an indicator to measure the purity of data. The smaller the Gini coefficient, the higher the purity of the data. The Gini coefficient of a feature can be obtained by weighted calculation of the Gini coefficients of the child nodes.

[0075] The node splitting process is to select the best splitting point to construct the nodes of the decision tree based on the feature selection index.

[0076] Information gain splitting: By calculating the information gain of each feature, the feature with the largest information gain is selected as the splitting feature of the node.

[0077] Gini coefficient splitting: By calculating the Gini coefficient of each feature, the feature with the smallest Gini coefficient is selected as the splitting feature of the node.

[0078] Pruning is to prevent overfitting. The decision tree can be pruned by pre-pruning or post-pruning.

[0079] Pre-pruning: When splitting a node, you can set a threshold to limit the conditions for node splitting to prevent the decision tree from overgrowing.

[0080] Post-pruning: After the decision tree is built, you can decide whether to prune the node by calculating the model performance before and after pruning.

[0081] The final decision tree classifier can be constructed based on the input features. Predict the corresponding support plan, perform feature judgment along the branches of the decision tree, and finally reach the leaf node to obtain the classification result.

[0082] The decision tree classifier can be reorganized according to the current scene resources to obtain the best solution for work support.

[0083] S13: Arrange the work tasks corresponding to the first work report to obtain a second work report.

[0084] In this embodiment, after obtaining the first work report, the work tasks in the first work report are arranged. During the arrangement, the first work report is analyzed and the overall arrangement is performed in combination with the expert evaluation opinions on each task to obtain the second work report.

[0085] In this embodiment, the specific steps of arranging the work tasks corresponding to the first work report to obtain the second work report include: S13-1: Based on the expert opinions corresponding to the first work report, evaluate the importance of each task in the first work report to obtain the importance weight and task time corresponding to each task.

[0086] In this embodiment, firstly, the importance of each task in the first work report is evaluated according to the expert opinions corresponding to the first work report, and the importance weight and task time corresponding to each task are obtained.

[0087] S13-2: Schedule each of the work tasks according to the importance weight and the task time to obtain the second work report.

[0088] In this embodiment, an overall task chain is formed according to the obtained importance weights and task times, and then the tasks are arranged to obtain a second work report.

[0089] In this embodiment, the matters are analyzed and arranged to form the entire task chain. The analysis and arrangement of matters is mainly the arrangement and collation of the overall implementation process of the work analysis report, which integrates matters and experts to form an overall arrangement, combines the professional skills, successful cases, working hours of experts, and the analysis and discussion of the work, relies on natural language processing technology (NLP) to understand the communication between experts, analyzes the content, relationship and priority of tasks for supplementation, and uses optimization algorithms (such as linear programming or genetic algorithms) to sort the experts' professional field knowledge, cases, skills and experience, as well as the correlation information between experts and work scenarios, and the experts' work saturation evaluation, etc., to achieve a highly efficient task arrangement plan for the entire line, thereby ensuring the effective support effect of the work and enriching the execution link in the work analysis report.

[0090] In this example, assume there are m tasks, and each task i has an importance weight and completion time . Forced sorting problems can usually be optimized by solving the following linear programming problem: (10) : The total number of tasks.

[0091] : The importance weight of task i.

[0092] : The time required to complete task i.

[0093] : The given time limit.

[0094] : Binary variable indicating whether task i is selected.

[0095] Through the above steps, an orderly work arrangement template is formed, experts are dispatched and supported, work is coordinated and connected through the approval process between the business system, and the effectiveness of expert support is incentivized through paid points acquisition and capability marking, thereby ensuring effective support for the work. Through the arrangement of matters, the work process is sorted out to facilitate the smooth development of work and open up the execution link of the work report.

[0096] S14: Based on the cross-domain expert opinions, the second working report is technically supplemented to obtain a third working report.

[0097] In this embodiment, the cross-domain expert opinion is a technical opinion on the work report put forward by experts in a work field different from the target work scenario.

[0098] In this embodiment, the opinions of cross-domain experts are encoded, represented in the form of vectors, added to the model, and the second working report is technically supplemented to obtain a third working report, which is the final working report.

[0099] In this embodiment, the specific steps of technically supplementing the second working report based on the cross-domain expert opinions to obtain the third working report include: S14-1: Obtain the cross-domain expert opinions on the second work report.

[0100] In this embodiment, cross-domain expert opinions on the second work report are first obtained.

[0101] In this embodiment, a cross-domain expert online session may be automatically created, the second work report may be sent to the device of the cross-domain expert, and the cross-domain expert may reply or discuss opinions on the second work report online.

[0102] S14-2: Input the cross-domain expert opinion into a vector model to obtain a vector corresponding to the cross-domain expert opinion.

[0103] In this embodiment, the vector model is used to encode natural language into a language that can be recognized by a machine.

[0104] In this embodiment, the received cross-domain expert opinions are input into a vector model to obtain vectors corresponding to the cross-domain expert opinions.

[0105] S14-3: Determine the technical opinion for each of the tasks according to the vectors corresponding to the cross-domain expert opinions.

[0106] In this embodiment, the system analyzes the vectors corresponding to the cross-domain expert opinions, and then determines the cross-domain expert's technical opinions on each task.

[0107] S14-4: Add the technical opinions to the second working report to obtain the third working report.

[0108] In this embodiment, the summarized technical opinions are added to the second working report to obtain a third working report.

[0109] In this embodiment, the work report is comprehensively analyzed through cross-domain analysis collaboration to determine the reliability and feasibility of the model. With the help of the cloud platform, when cross-domain expert support is needed, a collaborative working environment scenario is provided for different experts in various fields to promote the fluidity of technology and business in the entire field, and realize the interconnection of information such as technology sharing and expert sharing, thereby ensuring the feasibility of the work analysis report.

[0110] In this embodiment, an intranet environment can be used to establish a communication voucher, and the information data of the expert can be interoperated with the local application in the form of an interface. The interface is processed by interface authentication. During the data transmission process, the data can be encrypted and transmitted, and after reaching the system, it is decrypted. The operating users of the method system are divided into different rights and domains for level-by-level processing, that is, each user has different processing permissions for the work report, such as viewing permissions, modification permissions, arrangement permissions, and permissions to make comments. The user account involves identity card verification to ensure the safety and reliability of the user. More complex work will involve experts in different fields to support. The method is to form a complete work report by calculating and analyzing the fields in which experts are good at and the case library of previous work through the identification, perception, and adaptation of the work scene; if the original accumulated scene template of the system has a very low adaptability, it is necessary to construct a new work scene template. The method can actively recommend and select the recommended solutions of experts corresponding to complex work, apply the matter arrangement capability, pull experts from various domains to carry out this work through process approval, and incorporate the support solution into the new work scene template library for future work case reference.

[0111] In this embodiment, in a cross-domain expert collaboration environment, the word embedding model BERT is used to encode the text communication of experts. , it will be converted into a high-dimensional vector These encoded representations can then be used for various tasks such as classification, entity recognition, etc.

[0112] : Input text, such as expert communication content.

[0113] : High-dimensional vector, which is the BERT model's input text The encoding result.

[0114] In this embodiment, during the expert matter arrangement and cross-domain expert collaboration stages, natural language processing technology (NLP) can be applied to understand the communication between experts and extract important information. Among them, BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained deep learning model that can be used for text encoding and understanding. By using the BERT model to encode the text communication of experts, the text can be converted into a high-dimensional vector representation to capture the semantics and key information in the text.

[0115] In this embodiment, when the BERT model is applied to encode expert text communication, the following steps are generally included: Processing of input text: First, the text communication between experts needs to be preprocessed, including word segmentation, tokenization, and conversion to the input format of the BERT model.

[0116] Loading the BERT model: Next, you need to load the pre-trained BERT model. You can use the interface provided by the existing deep learning framework (such as TensorFlow, PyTorch) to load the BERT model.

[0117] Decoding of encoded output: Input the text communication into the BERT model, and after the model's encoding process, a high-dimensional vector representation of the text is obtained. These vectors can be used for various NLP tasks, such as text classification, entity recognition, sentiment analysis, etc.

[0118] The specific mathematical formula of the BERT model involves the self-attention mechanism (Self-Attention) and multi-head attention (Multi-Head Attention) in the Transformer architecture. The details are as follows: Self-attention mechanism: given an input sequence , the output representation of each position is calculated through the self-attention mechanism The calculation formula of the self-attention mechanism is: (11) Among them, Q, K, and V are the matrices of query, key, and value respectively. Is the dimension of the query or key.

[0119] Multi-head attention: The BERT model uses multiple attention heads to calculate attention representations of different aspects, and then concatenates these representations together. (12) in, = is the calculation result of the i-th attention head, , , is the corresponding weight matrix, is the output matrix.

[0120] In this embodiment, through cross-domain analysis collaboration, the technical opinions corresponding to the cross-domain expert opinions are supplemented into the second working report to obtain the third working report, that is, the final working report.

[0121] In another embodiment of the present application, the method further includes: S31: Receive analysis opinions corresponding to the third work report.

[0122] In this embodiment, after the third work report is formed, during use, analysis opinions on the third work report are received from technicians or experts.

[0123] S32: Based on the analysis opinions, the parameters of the model and algorithm used in the work report generation process are optimized.

[0124] In this embodiment, based on the received analysis opinions, parameters of the models and algorithms used in the work report generation process are optimized, such as optimizing the scene recognition model, optimizing the algorithm corresponding to the decision tree classifier, optimizing the work task scheduling related algorithms, and optimizing the encoding algorithm of cross-domain expert opinions.

[0125] In this embodiment, the opinions of each technician can be marked to screen out high-quality technical opinions, thereby motivating the technicians to more actively put forward technical opinions.

[0126] In this embodiment, the basic work report template is determined through work scenario recognition and perception, and then the report is supplemented with content through scenario analysis and adaptation. The implementation process is completed through the arrangement of matters, and then the analysis is integrated through cross-domain expert support analysis. The expert opinions are integrated into the work report based on data analysis and optimization algorithms to obtain the final work report. The work report includes environmental analysis, data analysis, talent screening, and case analysis. In this method, as the number of work scenarios increases, the basic scenario types will be continuously expanded, the system will be continuously trained to adjust the performance of the model and algorithm, and the accuracy of the generated report will be continuously improved. The work analysis report formed by the above method can greatly reduce the ex ante human investment cost, including the scope definition and content analysis of scenario matters, the data analysis and solutions for the difficulties of specific work, the specific arrangement of work matters, and the recommendation of experts who need support at each link. The resulting analysis report, when put into use in the early stage, needs to undergo an internal final report review and revision to form a formal work analysis report. A summary and feedback of this work will be conducted afterwards to revise the applicability of the method, continuously optimize and iterate, continuously promote the expansion and upgrading of scenarios, and promote the completeness of the work analysis report method, thereby realizing the transition of the method from modeling to standardization and ultimately achieving efficient work support.

[0127] In this embodiment, when generating a report, active analysis and identification are first performed, and scene data is annotated and environment adapted in combination with the scene, and preliminary data collection is performed: For example, the scene annotation is as follows: Table 1

[0128] The environment adaptation is as follows: Table 2

[0129] According to the above annotations, a work report template is formed, which includes various work tasks and operations to be performed, as well as required work environment information and the like.

[0130] In this embodiment, when performing decision algorithm analysis, the expert information and work cases involved in the work tasks are integrated, and the decision tree classifier is classified according to different dimensions to form a decision chain, and the high matching content in the serial work cases is recombined, that is, the relationship between the businesses is taken into account during the arrangement, and the high adaptation functions of the case requirements and detailed designs are extracted sufficiently, and the feasibility of each data is classified and decided to form the supporting content based on each task. In combination with the arrangement of work items and cross-domain analysis and collaboration, the experts in this domain are adapted, and the data analysis of cross-domain experts based on cases and communication of talent cloud is accurately adapted and data annotated, and then the business verification indicators of the analysis report are formed. The expert data annotation is as follows: Table 3

[0131] Combined with the overall evaluation of the environment, scenario cases, expert strength and other data required for the work items, an objective analysis table is comprehensively evaluated and the fitness ratio of the overall analysis report is given, as shown in the following table: Table 4

[0132] refer to Figure 2 , Figure 2 is a schematic diagram of a report generation method proposed in an embodiment of the present application, such as Figure 2 As shown, first, the work scenario data is collected and input into the deep learning model. After perception and recognition by the deep learning model, the optimal solution is obtained through the decision tree analyzer, natural language processing, and cloud computing platform to form the final work report.

[0133] In another embodiment of the present application, the report generation method is further described in combination with the communication service working scenario: First, we identify and perceive the work scenes and collect work scenes in the communication field, such as information communication, technical exchanges, program presentations, project order signing, solution customization, etc. We use the deep learning model recurrent neural network to train these data, where the input Xi is the scene feature and Yi is the corresponding scene label. By minimizing the loss function for training, we get a model that can automatically identify and perceive different work environments.

[0134] The decision tree classifier is then used to select the most appropriate support solution for the current work scenario. For example, when the "information communication" scenario is identified, the system may select the support solution of "providing a large-screen display and high-quality audio equipment"; while in the "plan presentation" scenario, the system may select the support solution of "providing projection equipment and demonstration materials".

[0135] Through the analysis and arrangement of matters, the work forms an orderly work task execution template, combined with (natural language processing) NLP technology to understand the communication between experts, analyze the task content, relationship and priority for supplementation. Then, the linear programming algorithm is used to sort the experts' professional field knowledge, work saturation, skills and experience, as well as the related information between experts and work scenarios, to achieve a highly efficient task arrangement plan for the entire line, thereby ensuring the effective support effect of the work.

[0136] Through the cloud computing platform, a collaborative working environment is provided for experts in different communication fields. For example, when an expert is dealing with a technical problem, he can share the problem on the platform and ask experts in other fields for support or suggestions. The BERT model is used to encode the text content of the expert communication to obtain high-dimensional vectors, which can be used for subsequent decision analysis.

[0137] When it is put into use in the early stage, simple work in the early stage can automatically call up existing work report templates through the existing scenario template library, such as the work report template for project order signing, for work support. Users only need to provide the necessary elements of the work items, and the system automatically generates a work report based on scenario recognition, perception, content analysis adaptation, and case analysis. For complex work, this method generates a basic template, and then pulls in experts to supplement and improve the content to form a final analysis report. The final work analysis report, during the early operation period, needs to undergo an internal final report review and revision to form a formal work analysis report, and then summarize and feedback this work to revise the applicability of the method, continuously optimize and iterate, continuously promote the expansion and upgrading of scenarios, and promote the completeness of the work analysis report method, thereby realizing the transition of the method from modeling to standardization, and ultimately achieving efficient work support.

[0138] refer to Figure 3 , Figure 3 is a business application matching diagram proposed in an embodiment of the present application, such as Figure 3 As shown, firstly, the work scenario characteristics, work scenario tasks and work scenario requirements are input into the deep learning model, then the work scenario is adapted through the decision tree classifier, matters are arranged through weights, and cross-domain expert collaboration is achieved through natural language processing technology and cloud computing platform, so as to obtain the optimal analysis and optimization algorithm and generate the final work report.

[0139] refer to Figure 4 , Figure 4 : is a schematic diagram of the overall flow of the working method proposed in one embodiment of the present application. Figure 4 As shown, when a work report needs to be generated, data collection is first performed to collect work scenario features, work scenario tasks, and work scenario requirements, which are then input into the deep learning model. The report template is then filled in through a machine learning classification algorithm (decision tree classifier). Natural language processing is used to identify the task importance assessment and task time for the report, and then the work report is scheduled. Cross-domain expert opinions are received through the cloud computing platform, and then overall data analysis is performed, the algorithm is optimized, and overall task scheduling and resource scheduling are achieved to form the final work report.

[0140] In the above embodiments of the present application, a scene recognition model is obtained through recurrent neural network training, and the scene is perceived and recognized, which increases the flexibility and intelligence of scene recognition, effectively reduces manpower input, and improves work efficiency. A decision tree machine learning classification algorithm is used to select the most suitable solution support for the work scene according to predefined rules or rules obtained through learning, further improve the work report, and improve the accuracy of the work report. A collaborative working environment is provided for experts in different fields through a cloud computing platform, forming an orderly work scheduling template, expert scheduling and solution support, which facilitates the joint decision-making of staff, is conducive to the improvement of work reports, promotes the sharing of technology and information in various fields, improves overall work efficiency, and is conducive to improving the generation efficiency and accuracy of work reports in various fields.

[0141] Based on the same inventive concept, an embodiment of the present application provides a report generating device. Figure 5 , Figure 5 FIG. 5 is a schematic diagram of a report generating device 500 proposed in an embodiment of the present application. Figure 5 As shown, the device comprises: The scene recognition module 501 is used to receive the scene data of the target work scene through the pre-trained scene recognition model, and obtain the work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; A decision classification module 502 is used to fill the work report template with content according to preset rules through a decision tree classifier to obtain a first work report; A task scheduling module 503 is used to schedule the work tasks corresponding to the first work report to obtain a second work report; The technical supplement module 504 is used to perform technical supplements on the second working report according to the cross-domain expert opinions to obtain a third working report.

[0142] Optionally, the device further comprises: An opinion receiving module, used for receiving analysis opinions corresponding to the third work report; The parameter optimization module is used to optimize the parameters of the model and algorithm used in the work report generation process according to the analysis opinions.

[0143] Optionally, the training step of the scene recognition model includes: Collect work scenario data; Adding corresponding scene labels to the work scene data to obtain a training data set; The training data set is input into the model to be trained, and the model is trained to obtain the scene recognition model.

[0144] Optionally, the scene recognition module includes: A type determination submodule, used to analyze the scene data through the scene recognition model to determine the scene type and target task corresponding to the target work scene; The template generation module is used to generate the corresponding work report template according to the scenario type and the target task.

[0145] Optionally, the decision classification module includes: A decision chain acquisition submodule, used to classify the scene features corresponding to the target work scene through the decision tree classifier to obtain a corresponding decision chain; A support solution acquisition submodule, used to generate a support solution corresponding to the target work scenario according to the decision chain; The first work report acquisition submodule is used to fill the work report template with the support solution to obtain the first work report.

[0146] Optionally, the task scheduling module includes: A task evaluation submodule, configured to evaluate the importance of each task in the first work report according to the expert opinions corresponding to the first work report, and obtain the importance weight and task time corresponding to each task; The second work report acquisition submodule is used to schedule each of the work tasks according to the importance weight and the task time to obtain the second work report.

[0147] Optionally, the technical supplement module includes: A cross-domain expert opinion acquisition submodule, used to obtain the cross-domain expert opinion for the second work report; A vector acquisition submodule, used for inputting the cross-domain expert opinion into a vector model to obtain a vector corresponding to the cross-domain expert opinion; A technical opinion determination submodule, used to determine the technical opinion for each of the tasks according to the vectors corresponding to the cross-domain expert opinions; The third working report obtaining submodule is used to add the technical opinions into the second working report to obtain the third working report.

[0148] Based on the same inventive concept, another embodiment of the present application provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the report generating method as described in any of the above embodiments of the present application.

[0149] Based on the same inventive concept, another embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps in the report generation method described in any of the above embodiments of the present application.

[0150] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0151] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0152] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0153] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0156] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0157] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0158] The report generation method, device, equipment and storage medium provided by the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A report generation method, characterized in that: The method comprises: Receive scene data of a target work scene through a pre-trained scene recognition model, and obtain a work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; Filling the work report template with content according to preset rules through a decision tree classifier to obtain a first work report; Arrange tasks corresponding to the first work report to obtain a second work report; Based on the opinions of cross-domain experts, the second working report was technically supplemented to obtain the third working report.

2. The report generation method according to claim 1, characterized in that: The method further comprises: Receiving analysis opinions corresponding to the third work report; Based on the analysis opinions, parameters of the models and algorithms used in the work report generation process are optimized.

3. The report generation method according to claim 1, characterized in that: The training steps of the scene recognition model include: Collect work scenario data; Adding corresponding scene labels to the work scene data to obtain a training data set; The training data set is input into the model to be trained, and the model is trained to obtain the scene recognition model.

4. The report generation method according to claim 1, characterized in that: The receiving of scene data of a target work scene through a pre-trained scene recognition model to obtain a work report template corresponding to the target work scene includes: Analyzing the scene data through the scene recognition model to determine the scene type and target task corresponding to the target work scene; Generate the corresponding work report template according to the scenario type and the target task.

5. The report generation method according to claim 1, characterized in that: The decision tree classifier is used to fill the work report template with content according to preset rules to obtain a first work report, including: Classifying the scene features corresponding to the target work scene by the decision tree classifier to obtain a corresponding decision chain; According to the decision chain, a support plan corresponding to the target work scenario is generated; The work report template is filled in using the support solution to obtain the first work report.

6. The report generation method according to claim 1, characterized in that: The step of arranging the work tasks corresponding to the first work report to obtain a second work report includes: According to the expert opinions corresponding to the first work report, an importance assessment is performed on each task in the first work report to obtain an importance weight and a task time corresponding to each task; According to the importance weight and the task time, each of the work tasks is scheduled to obtain the second work report.

7. The report generation method according to claim 1, characterized in that: The second working report is technically supplemented according to the cross-domain expert opinions to obtain a third working report, including: Obtaining the cross-domain expert opinion on the second work report; Inputting the cross-domain expert opinions into a vector model to obtain a vector corresponding to the cross-domain expert opinions; Determining a technical opinion for each of the tasks according to the vectors corresponding to the cross-domain expert opinions; The technical opinions are added to the second working report to obtain the third working report.

8. A report generating device, characterized in that: The device comprises: A scene recognition module, used to receive scene data of a target work scene through a pre-trained scene recognition model, and obtain a work report template corresponding to the target work scene, wherein the scene data includes scene feature data, scene task data, and scene requirement data; A decision classification module, used to fill the work report template with content according to preset rules through a decision tree classifier to obtain a first work report; A task scheduling module, used for scheduling the work tasks corresponding to the first work report to obtain a second work report; The technical supplement module is used to provide technical supplements to the second working report based on cross-domain expert opinions to obtain a third working report.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images