Intelligent electricity demand identification system and method for multi-element scene

Abstract

The invention discloses an intelligent electricity demand identification system and method for a multi-element scene, and belongs to the field of power supply services, and the system comprises a scene presetting module which is used for presetting a plurality of electricity service scenes, and each electricity service scene is provided with at least one intention classification label; the model building module is used for building a natural language processing model; the intention recognition module is used for performing intention recognition on the customer expression text to obtain an intention classification result; the demand generation module is used for generating electricity demand information; the topology mapping module is used for establishing a topology mapping relationship between the WeChat group and the customer manager; and the information pushing module is used for automatically pushing the electricity demand information to a corresponding customer manager. The customer demand intention in the WeChat group is accurately recognized through the natural language processing model, and the technical effect of improving the power supply service efficiency and quality is achieved in combination with the topological mapping relation between the customer manager and the WeChat group.

Description

Technical Field

[0001] The present invention relates to the field of power supply services, and in particular to a system and method for intelligently identifying power demand for multiple scenarios. Background Art

[0002] In recent years, many power supply service companies have used WeChat groups, a widely used social platform, to interact and communicate with customers, and to promptly understand and respond to various service demands of customers. However, there are some problems with the existing technology in WeChat group power supply services. First, it is difficult to accurately understand the real needs of customers for various service demands raised by customers in the WeChat group; second, after identifying customer needs, it is difficult to quickly and accurately match customer demands to the account manager responsible for the WeChat group, resulting in untimely service responses and improper handling. These problems affect the efficiency and quality of power supply services. Summary of the invention

[0003] This application provides an intelligent identification system and method for electricity demand in multiple scenarios, aiming to solve the technical problems in the prior art of low accuracy in identifying various customer service requests in WeChat groups and difficulty in matching customer managers, resulting in low efficiency and quality of power supply services.

[0004] In view of the above problems, the present application provides a system and method for intelligently identifying electricity demand for multiple scenarios.

[0005] The first aspect disclosed in the present application provides an intelligent identification system for electricity demand for multiple scenarios, the system comprising: a scenario preset module, used to preset multiple electricity service scenarios, wherein each electricity service scenario has at least one intent classification label; a model building module, used to build a natural language processing model, the natural language processing model uses a labeled data set in the power supply service field to perform fine-tuning on a pre-trained NLP large model, and the labeled data set is constructed based on multiple electricity service scenarios and corresponding intent classification labels; an intent recognition module, used to use the natural language processing model to perform intent recognition on customer expression text to obtain intent classification results, wherein the customer expression text is sent by customers in a WeChat group; a demand generation module, used to extract key information elements from the customer expression text according to the intent classification results, and generate electricity demand information in combination with the intent classification results; a topology mapping module, used to use the unique identifier of the account manager as a topological boundary condition to establish a topological mapping relationship between the WeChat group and the account manager; an information push module, used to automatically push the electricity demand information to the corresponding account manager based on the topological mapping relationship.

[0006] Another aspect disclosed in the present application provides an intelligent identification method for electricity demand in multiple scenarios, the method comprising: presetting multiple electricity service scenarios, wherein each electricity service scenario has at least one intent classification label; building a natural language processing model, the natural language processing model using a labeled data set in the power supply service field to perform fine-tuning on a pre-trained NLP large model, the labeled data set being constructed based on multiple electricity service scenarios and corresponding intent classification labels; using the natural language processing model to perform intent identification on customer expression text to obtain intent classification results, wherein the customer expression text is sent by customers in a WeChat group; based on the intent classification results, extracting key information elements from the customer expression text, and generating electricity demand information in combination with the intent classification results; using the unique identifier of the account manager as a topological boundary condition to establish a topological mapping relationship between the WeChat group and the account manager; and automatically pushing the electricity demand information to the corresponding account manager based on the topological mapping relationship.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The scenario preset module is used to preset multiple electricity service scenarios, each of which corresponds to at least one intent classification label, which lays the foundation for the subsequent identification of customer demand intentions. The natural language processing model is built through the model building module. The model uses the labeled data set in the power supply service field to perform fine-tuning on the pre-trained NLP large model. The labeled data set is constructed based on multiple electricity service scenarios and corresponding intent classification labels, so that the model can better understand and process natural language expressions in power supply service scenarios. Through the intent recognition module, the trained natural language processing model is used to identify the intent of the text sent by customers in the WeChat group, and the intent classification results are obtained, realizing the intelligent identification of customer demand intentions. Through the demand generation module, according to the intent classification results, key information elements are extracted from the customer expression text, and structured The electricity demand information converts unstructured customer expressions into structured demand information for subsequent processing. Through the topology mapping module, the unique identifier of the account manager is used as the topological boundary condition to establish a topological mapping relationship between the WeChat group and the account manager, thereby realizing the intelligent association between customer demand and the account manager. Through the information push module, the generated electricity demand information is automatically pushed to the corresponding account manager based on the topological mapping relationship between the WeChat group and the account manager, thereby realizing a technical solution for accurate matching and push of customer demand. This solves the technical problems in the prior art of low accuracy in identifying various customer service demands in the WeChat group and difficulty in matching account managers, resulting in low efficiency and quality of power supply services. Through the natural language processing model, the customer demand intentions in the WeChat group are accurately identified, and combined with the topological mapping relationship between the account manager and the WeChat group, the technical effect of improving the efficiency and quality of power supply services is achieved.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A structural schematic diagram of an intelligent identification system for electricity demand in multiple scenarios is provided for an embodiment of the present application.

[0010] Figure 2 A flowchart of a method for intelligently identifying electricity demand for multiple scenarios is provided for an embodiment of the present application; Explanation of the accompanying drawings: scenario preset module 11, model building module 12, intention recognition module 13, demand generation module 14, topology mapping module 15, information push module 16. DETAILED DESCRIPTION

[0011] The overall idea of ​​the technical solution provided by this application is as follows: The embodiment of the present application provides an intelligent identification system and method for electricity demand for multiple scenarios. First, a plurality of electricity service scenarios and corresponding intent classification labels are preset through the scenario preset module, laying the foundation for subsequent intent recognition. Then, the model building module is used to fine-tune the pre-trained NLP large model using the labeled data set in the power supply service field to obtain a professional model that can efficiently process natural language expressions in this field. Next, the intent recognition module is used to identify the intent of the text sent by the customer in the WeChat group, and then the key information elements are extracted through the demand generation module to generate structured electricity demand information. At the same time, the topology mapping module establishes a topology mapping relationship between the WeChat group and the account manager to realize the intelligent association between customer needs and the account manager. After that, the information push module accurately pushes the electricity demand information to the corresponding account manager based on the topology mapping relationship.

[0012] To sum up, this application realizes the automatic identification, association and push of customer needs in the WeChat group scenario, improves the pertinence, timeliness and accuracy of power supply services, and improves the efficiency and quality of power supply services.

[0013] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification. Example

[0014] like Figure 1 As shown, the embodiment of the present application provides an intelligent identification system for power demand in multiple scenarios, and the system includes: The scenario preset module 11 is used to preset multiple electricity service scenarios, wherein each electricity service scenario has at least one intent classification label.

[0015] Specifically, the scene preset module 11 summarizes a number of typical electricity service scenarios by analyzing historical electricity service data. These scenarios cover various situations that users may encounter during electricity use, such as fault repair, payment inquiry, business handling, etc. Among them, each electricity service scenario represents a type of electricity demand or problem of the user. In order to facilitate the subsequent classification and identification of user needs, at least one intent classification label is set for each electricity service scenario during the scene preset process. The intent classification label is a high-level summary of the user's needs or problem types in the scenario. Through the preset intent classification labels, the system can quickly determine which type of electricity service scenario the specific needs expressed by the user belong to, thereby laying the foundation for providing targeted services. For example, for the electricity service scenario of "fault repair", intent classification labels such as "power outage repair" and "line fault" are set; for the "payment inquiry" scenario, intent classification labels such as "arrears inquiry" and "payment method" are set.

[0016] A plurality of electricity service scenarios are preset through the scenario preset module 11, and an intent classification label is set for each scenario, laying a solid foundation for realizing intelligent electricity demand identification.

[0017] The model building module 12 is used to build a natural language processing model, which uses a labeled data set in the power supply service field to fine-tune the pre-trained NLP large model. The labeled data set is constructed based on multiple power service scenarios and corresponding intent classification labels.

[0018] Specifically, the model building module 12 builds a customized natural language processing model based on a pre-trained NLP (natural language processing) model. The NLP model refers to a language model pre-trained on a large amount of general corpus, which has strong language understanding and expression capabilities. The pre-trained NLP model is used as a basis, and targeted fine-tuning is performed on this basis to adapt to the specific language expression habits and professional terms in the power supply service field.

[0019] In order to enable the natural language processing model to better understand the user needs in the electricity service scenario, a labeled data set in the power supply service field is constructed. The construction of the labeled data set is based on the aforementioned multiple electricity service scenarios and corresponding intent classification labels. Specifically, typical user expression samples in each scenario are extracted from the historical electricity service records, and these samples are labeled with intent by professionals to form a user expression corpus with intent classification labels. When building a natural language processing model, the model building module 12 fine-tunes the pre-trained NLP large model according to the labeled data set. By adding the intent classification task for the electricity service scenario to the pre-trained model, and using the labeled data set to train and optimize the model parameters, the model can learn the professional knowledge and language characteristics in the power supply service field, thereby improving the recognition accuracy of the user's expression intention. Among them, the fine-tuning process adopts a multi-round iteration method to continuously evaluate the performance of the model on the labeled data set, and optimize the model structure and parameters according to the evaluation results until a natural language processing model with excellent performance suitable for the electricity service scenario is obtained.

[0020] Through the model building module 12, based on the pre-trained NLP large model, targeted fine-tuning is performed using the labeled data set in the power supply service field to build a natural language processing model that can accurately understand user needs and perform intent recognition, providing strong technical support for subsequent smart electricity consumption services.

[0021] The intention recognition module 13 is used to use the natural language processing model to perform intention recognition on the customer expression text to obtain an intention classification result, wherein the customer expression text is sent by a customer in the WeChat group.

[0022] Specifically, the intention recognition module 13 receives customer expression text, which is various messages sent by customers in the WeChat group, such as consultation, complaint, repair, etc. For the customer expression text, the intention recognition module 13 first inputs it into the natural language processing model for feature extraction and semantic understanding. The natural language processing model performs a series of processing such as word segmentation, syntactic analysis, and semantic encoding on the customer expression text, and converts the text into a structured representation that can be understood and processed by the computer. On the basis of the semantic understanding of the text, the intention recognition module 13 uses the natural language processing model to perform intent recognition on the customer needs expressed in the customer expression text, and obtains the customer need type as the intention classification result.

[0023] Through the intent recognition module, the intent classification results are obtained, indicating the specific demand type expressed by the customer, such as "power outage repair" and "arrears inquiry", etc., providing a basis for subsequent service decisions. The system automatically triggers the corresponding service process according to the identified intent type, thereby realizing intelligent and precise customer service.

[0024] The demand generation module 14 is used to extract key information elements from the customer expression text according to the intention classification result, and generate electricity demand information in combination with the intention classification result.

[0025] Specifically, the demand generation module 14 receives the intent classification result output by the intent recognition module 13, which indicates the demand type to which the customer's expression text belongs. On this basis, the demand generation module 14 performs in-depth information extraction on the customer's expression text, and identifies key information elements related to the intent type from the customer's expression text. For example, for customer demands of the type "power outage repair", the demand generation module 14 focuses on extracting key information elements such as the power outage address, power outage time, and contact information of the person reporting the repair involved in the customer's expression text; for demands of the type "arrears inquiry", key information elements such as the customer's electricity user number and meter number are extracted. The extracted key information elements provide the necessary contextual data for subsequent service execution.

[0026] After extracting the key information elements, the demand generation module 14 comprehensively considers the intention classification results, matches and combines the key information elements with the intention types according to the preset demand generation rules and templates, and generates structured electricity demand information, including the customer's basic information, demand type, key appeal points and other dimensions to express the customer's specific needs.

[0027] The demand generation module 14 generates electricity demand information to drive subsequent service processes, such as automatic dispatching, task allocation, manual processing, etc. Due to the structuring and standardization of electricity demand information, subsequent service links can be connected and processed based on a unified data format, which improves the automation and efficiency of the service process and improves the efficiency and quality of the service.

[0028] The topology mapping module 15 is used to establish a topology mapping relationship between the WeChat group and the account manager by taking the unique identifier of the account manager as a topology boundary condition.

[0029] Specifically, the topological mapping module 15 constructs a topological mapping relationship from the WeChat group to the account manager by using the unique identifier of the account manager as a topological boundary condition. In this topological mapping relationship, each WeChat group corresponds to one or more account managers responsible for the group, and each account manager may also be responsible for managing multiple WeChat groups at the same time. Among them, when the topological mapping module 15 actually constructs the mapping relationship, it can use technologies such as graph databases, use WeChat groups and account managers as nodes in the graph, and establish directed edges based on the association characteristics between the nodes. Through the graphical organizational form, the relationship between the WeChat group and the account manager is clearly expressed.

[0030] Based on the established topological mapping relationship, the system can accurately push the electricity demand information to the account manager responsible for the corresponding WeChat group. When a service demand is generated in a WeChat group, the topological mapping module 15 quickly finds the corresponding account manager according to the intention recognition result, and pushes the electricity demand information to the account manager for processing. At the same time, the account manager can also timely understand and manage the service needs and customer dynamics of the WeChat group he is responsible for through the topological mapping relationship. Specifically, when constructing the topological mapping relationship between the WeChat group and the account manager, the account manager corresponding to each WeChat group is identified, and the electricity service scenario that each account manager is responsible for in the group is further identified. After the intention recognition module 13 identifies the intention classification result of a certain customer, the topological mapping module 15 quickly finds the account manager responsible for the electricity service scenario corresponding to the intention type according to the intention type of the intention classification result, thereby accurately pushing the electricity demand information to the corresponding account manager, ensuring that the service demand is handled in a timely and professional manner. For example, suppose there are customer managers A and B in a WeChat group, where manager A is responsible for the two electricity service scenarios of "fault repair" and "complaint and suggestion", and manager B is responsible for the two business scenarios of "business processing" and "consultation and inquiry"; when a service demand for "power outage repair" appears in the group, the intent recognition module 13 recognizes it as an intent of the "fault repair" type; then, the topology mapping module 15 quickly locates the customer manager A who is responsible for the "fault repair" scenario based on the intent type and the topology mapping relationship, and pushes the service demand to manager A for processing.

[0031] Through the topology mapping module 15, a mapping relationship between the WeChat group and the account manager is established, which provides strong support for the accurate push of electricity demand information, ensures that service needs can quickly reach the account manager responsible for handling, and improves the response speed and quality of the service.

[0032] The information push module 16 is used to automatically push the electricity demand information to the corresponding customer manager based on the topology mapping relationship.

[0033] Specifically, the information push module 16 pushes the electricity demand information to the corresponding account manager based on the topological mapping relationship established by the topological mapping module 15, without manual intervention, thereby improving the circulation efficiency and accuracy of the electricity demand information. First, the information push module 16 receives the electricity demand information output by the demand generation module 14, and extracts key fields therein, such as the intent type, WeChat group identifier, etc. Then, the information push module 16 obtains the information of the account manager responsible for the WeChat group according to the WeChat group identifier in the electricity demand information, including the account manager's work number, name, contact information, etc. Next, the information push module 16 further screens the target account manager who is directly responsible for handling the demand according to the business division of the account manager, and determines the final push object. After determining the push object, the information push module 16 sends the electricity demand information to the target account manager. After receiving the electricity demand information, the target account manager quickly launches subsequent service processes, such as customer return visits, on-site surveys, and plan formulation.

[0034] By using the intelligent push mechanism through the information push module 16, the electricity demand information is accurately sent to the account manager responsible for processing, realizing the closed-loop management of the power supply service process and improving the service response speed and service quality.

[0035] Furthermore, the execution steps of the scene preset module include: Extracting historical electricity service data, analyzing the historical electricity service data, and obtaining multiple electricity service scenarios according to the analysis results; In combination with the historical electricity service data, a corresponding intent classification label is set for each electricity service scenario; Map and store multiple electricity service scenarios with corresponding intent classification labels.

[0036] In a feasible implementation, first, the scene preset module 11 extracts historical electricity service data. The historical electricity service data comes from multiple channels such as the business system, customer service records, and work order distribution of the power supply company. By collecting and integrating massive historical electricity service data, the scene preset module 11 obtains a comprehensive and real set of electricity service data, providing original material for subsequent scene analysis. Then, the scene preset module 11 analyzes the extracted historical electricity service data to discover the common characteristics and behavior patterns of user needs from the massive data. For example, through clustering, classification and other algorithms, the scene preset module 11 classifies similar user needs into the same category and obtains multiple electricity service scenarios, such as "fault repair", "arrears collection", "business processing", etc.

[0037] Next, based on the identification of the electricity service scenario, the scenario preset module 11 further combines the historical electricity service data to set corresponding intent classification labels for each electricity service scenario. The intent classification label is a summary and refinement of the user's specific demand intentions in the electricity service scenario, expressing the service content or problem that the user hopes to obtain in the scenario. For example, in the "fault repair" scenario, multiple intent classification labels such as "power outage repair", "line fault", and "equipment maintenance" are set to correspond to different types of fault repair needs. When setting the intent classification label, the scenario preset module 11 fully considers factors such as user expression methods and business keywords in the historical electricity service data to ensure that the intent classification label can accurately and comprehensively cover various demand intentions in the electricity service scenario. Afterwards, the scenario preset module 11 maps and stores the identified multiple electricity service scenarios with the corresponding set intent classification labels, and builds a complete "electricity service scenario-intent classification label" mapping relationship library, which provides a solid foundation for subsequent user intent identification and demand matching.

[0038] Furthermore, the execution steps of the model building module include: Based on the pre-trained NLP big model, construct an initial version of the natural language processing model, and add a scene intent classification task layer at the end of the NLP big model; Acquire the annotated data set, and fine-tune the parameters of the natural language processing model using a supervised learning method, wherein the annotated data set is constructed based on historical electricity service data; The performance of the fine-tuned natural language processing model is evaluated, and the natural language processing model is iteratively optimized to obtain a natural language processing model for intent recognition.

[0039] In a feasible implementation, first, the model building module 12 builds an initial version of the natural language processing model based on the pre-trained NLP big model. Among them, the NLP big model refers to a large-scale language model pre-trained on a large amount of general corpus, such as the Pangu NLP big model. The model building module 12 uses the language feature extraction and semantic modeling capabilities provided by the Pangu NLP big model to build a preliminary natural language processing model framework. On this basis, the model building module 12 adds a scene intent classification task layer at the end of the Pangu NLP big model so that the model can adapt to the intent recognition task in a specific field. The scene intent classification task layer adopts a neural network structure of a fully connected layer, a convolutional layer, etc., and is designed according to specific task requirements. By adding a dedicated task layer to the general Pangu NLP big model, the model building module 12 realizes the initial transformation of the natural language processing model from general to dedicated. Then, the model building module 12 obtains a labeled data set and fine-tunes the parameters of the natural language processing model in a supervised learning manner. The labeled data set refers to a training data set with intent labels constructed based on historical electricity service data. The model building module 12 extracts representative samples from the historical electricity service data, and business experts annotate the samples with intent to form a high-quality annotated data set. In the fine-tuning stage, the model building module 12 uses the input text and corresponding intent labels in the annotated data set to adjust and optimize the parameters of the natural language processing model through supervised learning. Through the back propagation algorithm and gradient descent strategy, the model's prediction results on the annotated data set are as close to the real labels as possible, thereby improving the model's performance in the intent recognition task in the electricity field.

[0040] After completing the initial fine-tuning, the model building module 12 performs a performance evaluation on the fine-tuned natural language processing model, comprehensively examines the performance of the model in the actual intent recognition task, and identifies possible problems and optimization space. Among them, the evaluation indicators include machine learning evaluation indicators such as precision, recall, and F1 value. Business indicators specific to the power consumption field can also be designed, such as intent recognition coverage, response time, etc. According to the results of the performance evaluation, the model building module 12 iteratively optimizes the natural language processing model, such as adjusting the model's hyperparameters, expanding training data, introducing new feature representations, and improving the task layer structure. Through multiple rounds of iterative optimization, the model building module 12 continuously improves the performance of the natural language processing model in the intent recognition task in the power consumption field, and obtains a natural language processing model suitable for intelligent recognition of electricity demand in multiple scenarios, providing high-quality intent recognition capabilities.

[0041] Furthermore, the execution steps of the model building module also include: Extracting historical electricity service data, and for each electricity service scenario, extracting a preset number of customer expression text samples from the historical electricity service data according to the corresponding intent classification label to form candidate data; Manually annotate the candidate data to obtain an initial annotated data set, wherein the initial annotated data set includes a plurality of annotated samples, each of which includes a customer expression text sample, an electricity service scenario sample, an intent classification label sample, and a key information element sample; Performing data enhancement processing on the initial annotated data set to obtain the annotated data set.

[0042] In a preferred embodiment, first, the model building module 12 extracts customer expression texts related to each electricity service scenario from the massive historical electricity service data to form candidate data. Specifically, for each electricity service scenario, the model building module 12 uses keyword matching, semantic similarity and other technologies according to its corresponding intent classification label to extract a preset number of customer expression text samples from the historical electricity service data, covering various typical requirements and problem descriptions under the electricity service scenario as candidate data. Then, the model building module 12 manually annotates the candidate data to obtain an initial annotated data set. The purpose of manual annotation is to add accurate and standardized intent labels and key information labels to the customer expression text samples in the candidate data, so as to provide high-quality supervision information for subsequent model training. During the annotation process, professional annotators will carefully read and analyze each customer expression text sample, and annotate the corresponding electricity service scenario sample, intent classification label sample and key information element sample according to the annotation specifications and business knowledge. Among them, the electricity service scenario sample represents the electricity service scenario to which the text belongs; the intent classification label sample represents the specific customer demand intent expressed by the text; and the key information element sample annotates the key information in the text, such as the fault address, etc. Through manual annotation, the model building module 12 obtains a high-quality initial annotation data set, which provides reliable training data for subsequent model fine-tuning.

[0043] Afterwards, in order to further expand the scale and diversity of the annotated data set, the model building module 12 performs data enhancement processing on the initial annotated data set. New samples are generated by transforming, combining, perturbing and other operations on the original data, thereby improving the generalization ability and robustness of the model. For example, by adopting a variety of data enhancement strategies, including synonym replacement, sentence rewriting, random combination of key information, etc., the model building module 12 generates a large number of new samples while keeping the original annotation information unchanged, so that the scale and coverage of the annotated data set are expanded, providing richer and more diverse training materials for the fine-tuning of the natural language processing model, and obtaining an annotated data set, thereby improving the generalization performance and practical application effect of the model.

[0044] Furthermore, the execution steps of the demand generation module include: Obtaining a preset key information element extraction rule, wherein the key information element extraction rule includes a key information element type corresponding to each intent classification label; Based on the key information element extraction rule, determine the target key information element type corresponding to the customer expression text through the intention classification result; Extract key information elements corresponding to the target key information element type from the customer expression text.

[0045] In a feasible implementation, first, the demand generation module 14 obtains preset key information element extraction rules. Key information element extraction rules refer to a rule system for extracting key information from customer expression texts, summarized based on business experience and data analysis. The rule system establishes a correspondence between intent classification labels and key information element types. Specifically, for each intent classification label, the key information element extraction rules preset one or more corresponding key information element types. For example, for the intent classification label "power outage repair", the corresponding key information element types include power outage address, power outage time, customer contact information, etc. By presetting key information element extraction rules, the demand generation module 14 can quickly determine which types of key information elements need to be extracted from the text based on the intent classification results of the customer expression text, thereby achieving accurate and efficient information extraction.

[0046] Then, the demand generation module 14 determines the target key information element type corresponding to the customer expression text through the intention classification result based on the key information element extraction rule. Specifically, the demand generation module 14 first receives the intention classification result output by the intention recognition module 13, which indicates the intention classification label to which the customer expression text belongs. Subsequently, the demand generation module 14 searches for the intention classification label in the key information element extraction rule and obtains the key information element type corresponding to it, which is the target key information element type that the demand generation module 14 needs to extract from the customer expression text, laying the foundation for subsequent accurate information extraction. After that, the demand generation module 14 extracts the key information elements corresponding to the target key information element type from the customer expression text. Specifically, the demand generation module 14 uses natural language processing technology, such as named entity recognition, keyword matching, dependency analysis, etc., to deeply analyze and extract information from the customer expression text. In the extraction process, the demand generation module 14 focuses on the text fragments and information points that match the target key information element type, and uses the predefined extraction mode and rules to accurately identify and extract the corresponding key information elements, providing the necessary information support for the subsequent user service demand generation.

[0047] Furthermore, the execution steps of the demand generation module also include: Extracting multiple target intent classification labels according to the intent classification results; According to multiple target intent classification labels, multiple corresponding electricity demand information generation templates are obtained; Fill the extracted key information elements into corresponding positions of multiple electricity demand information generation templates to obtain initial electricity demand information; The initial electricity demand information is semantically optimized and completed to generate the electricity demand information.

[0048] In a preferred embodiment, first, the demand generation module 14 extracts multiple target intent classification labels based on the intent classification results. In actual business scenarios, the needs expressed by customers may involve multiple intents, such as "power outage repair + complaint suggestions". In order to comprehensively and accurately generate electricity demand information, the demand generation module 14 comprehensively considers the various intentions contained in the customer's expression text. Specifically, the demand generation module 14 analyzes the intent classification results output by the intent recognition module 13, and identifies multiple intent classification labels with higher confidence as the target intent for generating electricity demand information. By extracting multiple target intent classification labels, the demand generation module 14 can more comprehensively understand and meet the diverse needs of customers.

[0049] Then, the demand generation module 14 obtains the corresponding multiple electricity demand information generation templates according to the multiple target intent classification labels. The electricity demand information generation template refers to a structured representation template of electricity demand information pre-defined for different intent types. Each template contains the key elements and basic format of the electricity demand information under the intent type. For example, the electricity demand information generation template corresponding to the "power outage repair report" intent includes elements such as the power outage address, power outage time, fault description, and customer information. The demand generation module 14 searches and obtains the corresponding multiple electricity demand information generation templates in the predefined template library based on the extracted multiple target intent classification labels. Next, the demand generation module 14 fills the extracted key information elements into the corresponding positions of the multiple electricity demand information generation templates to obtain the initial electricity demand information. Specifically, the demand generation module 14 fills the extracted key information elements, such as the power outage address, customer contact information, etc., into the corresponding fields of the corresponding power demand information generation template according to their types and meanings, so as to map and merge the unstructured key information elements with the structured power demand information generation template to generate a preliminary structured power demand information representation, i.e., the initial power demand information. Afterwards, the demand generation module 14 performs semantic optimization and completion on the initial power demand information to generate the final power demand information. The purpose of semantic optimization and completion is to further improve the readability, completeness and professionalism of the power demand information. Specifically, the demand generation module 14 uses natural language generation technology to perform grammatical analysis, semantic understanding and text generation on the initial power demand information. In the optimization process, the demand generation module 14 focuses on the logical relationship between the information elements, and makes the expressions between the various information elements more coherent and smooth by adding conjunctions, modifiers, etc. In the completion process, the demand generation module 14 infers and supplements the missing key information according to business rules and common sense, so that the power demand information is more complete and professional. After semantic optimization and completion, electricity demand information is generated, which accurately and completely expresses the customer's service needs and provides a high-quality information foundation for subsequent service responses.

[0050] Furthermore, the execution steps of the information push module include: Obtaining a WeChat group ID corresponding to the electricity demand information; According to the topological mapping relationship, determine the unique identifier of the account manager corresponding to the WeChat group identifier; Associating the electricity demand information with the unique identifier of the customer manager to generate an electricity demand information push task; The power demand information push task is triggered to push the power demand information to the customer manager account corresponding to the customer manager's unique identifier.

[0051] In a feasible implementation, first, the information push module 16 obtains the WeChat group identifier corresponding to the electricity demand information. The WeChat group identifier refers to the unique identification information of the WeChat group, such as the group number, group name, etc. The information push module 16 extracts the WeChat group identifier information contained therein by parsing the electricity demand information generated by the demand generation module 14, determines which specific WeChat group the electricity demand information comes from, and provides necessary clues for the subsequent customer manager matching. Then, the information push module 16 determines the unique identifier of the customer manager corresponding to the WeChat group identifier based on the topological mapping relationship. The information push module 16 uses the extracted WeChat group identifier as a query condition, searches in the topological mapping relationship, and obtains one or more unique identifiers of the customer manager corresponding to the WeChat group identifier. The unique identifier of the customer manager can be the unique identification information of the customer manager, such as the work number and telephone number. When there are multiple unique identifiers of customer managers, according to the intention recognition result, the unique identifier of the customer manager responsible for this type of electricity service business is determined as the object of subsequent information push.

[0052] Next, the information push module 16 associates the power demand information with the unique identifier of the account manager and generates a power demand information push task. Specifically, the information push module 16 binds the content, source, time and other information of the power demand information with the unique identifier of the account manager to form a structured push task object. The push task object contains complete information such as the target account manager to be pushed, the content to be pushed and the related parameters to be pushed, providing support for automated and batch information push. Afterwards, the information push module 16 triggers the power demand information push task and pushes the power demand information to the account manager account corresponding to the unique identifier of the account manager. Specifically, the information push module 16 searches for the corresponding account manager account information, such as the WeChat account, email address, etc. of the account manager according to the unique identifier of the account manager specified in the push task. Then, the information push module 16 pushes the content of the power demand information to the account manager's account by calling the corresponding message push interface, such as the WeChat message interface, the email sending interface, etc., to realize the automated and precise push of the power demand information to the target account manager, thereby improving the efficiency and reliability of information transmission and the timeliness and pertinence of the power supply service response.

[0053] In summary, the intelligent identification system for power demand in multiple scenarios provided by the embodiments of the present application has the following technical effects: The scenario preset module is used to preset multiple electricity service scenarios, where each electricity service scenario has at least one intent classification label, which provides a basis for the subsequent customer demand intention identification and ensures that various common electricity service scenarios can be covered; the model building module is used to build a natural language processing model. The natural language processing model uses the annotated data set in the power supply service field to fine-tune the pre-trained NLP large model. The annotated data set is built based on multiple electricity service scenarios and corresponding intent classification labels, so that the model can better understand and process the natural language expressions in this field and improve the accuracy of intent identification. The intent identification module is used to use the natural language processing model to identify the intent of the customer expression text and obtain the intent classification result. The customer expression text is sent by the customer in the WeChat group, which realizes the intelligent identification of the customer demand intention. The demand generation module is used to extract key information elements from the customer expression text according to the intent classification result, and generate electricity demand information in combination with the intent classification result, converting the unstructured customer expression into structured information that is easy to process, providing a data basis for subsequent intelligent matching and push. The topology mapping module is used to use the unique identifier of the account manager as the topology boundary condition to establish the topology mapping relationship between the WeChat group and the account manager, realizing the intelligent association between the WeChat group and the account manager responsible for the WeChat group, ensuring the pertinence and professionalism of the service. The information push module is used to automatically push the electricity demand information to the corresponding account manager based on the topology mapping relationship, realizing the accurate matching and push of customer needs, ensuring that customer demands can be handled in a timely and effective manner, and improving service quality and efficiency. Example

[0054] Based on the same inventive concept as the intelligent identification system for power demand in multiple scenarios in the aforementioned embodiment, Figure 2 As shown, the embodiment of the present application provides a method for intelligently identifying power demand for multiple scenarios, the method comprising: Preset multiple electricity service scenarios, where each electricity service scenario has at least one intent classification label; Build a natural language processing model that is fine-tuned on a pre-trained NLP model using a labeled dataset in the power supply service field, where the labeled dataset is constructed based on multiple power service scenarios and corresponding intent classification labels; Using the natural language processing model to perform intent recognition on the customer expression text to obtain an intent classification result, wherein the customer expression text is sent by a customer in the WeChat group; Extract key information elements from the customer's expression text according to the intention classification result, and generate electricity demand information in combination with the intention classification result; The unique identifier of the account manager is used as the topological boundary condition to establish the topological mapping relationship between the WeChat group and the account manager; The electricity demand information is automatically pushed to the corresponding customer manager based on the topology mapping relationship.

[0055] Furthermore, multiple electricity service scenarios are preset, including: Extracting historical electricity service data, analyzing the historical electricity service data, and obtaining multiple electricity service scenarios according to the analysis results; In combination with the historical electricity service data, a corresponding intent classification label is set for each electricity service scenario; Map and store multiple electricity service scenarios with corresponding intent classification labels.

[0056] Furthermore, a natural language processing model is built, including: Based on the pre-trained NLP big model, construct an initial version of the natural language processing model, and add a scene intent classification task layer at the end of the NLP big model; Acquire the annotated data set, and fine-tune the parameters of the natural language processing model using a supervised learning method, wherein the annotated data set is constructed based on historical electricity service data; The performance of the fine-tuned natural language processing model is evaluated, and the natural language processing model is iteratively optimized to obtain a natural language processing model for intent recognition.

[0057] Further, obtaining the annotated data set includes: Extracting historical electricity service data, and for each electricity service scenario, extracting a preset number of customer expression text samples from the historical electricity service data according to the corresponding intent classification label to form candidate data; Manually annotate the candidate data to obtain an initial annotated data set, wherein the initial annotated data set includes a plurality of annotated samples, each of which includes a customer expression text sample, an electricity service scenario sample, an intent classification label sample, and a key information element sample; Performing data enhancement processing on the initial annotated data set to obtain the annotated data set.

[0058] Furthermore, according to the intent classification result, key information elements are extracted from the customer expression text, including: Obtaining a preset key information element extraction rule, wherein the key information element extraction rule includes a key information element type corresponding to each intent classification label; Based on the key information element extraction rule, determine the target key information element type corresponding to the customer expression text through the intention classification result; Extract key information elements corresponding to the target key information element type from the customer expression text.

[0059] Furthermore, combining the intention classification result, electricity demand information is generated, including: Extracting multiple target intent classification labels according to the intent classification results; According to multiple target intent classification labels, multiple corresponding electricity demand information generation templates are obtained; Fill the extracted key information elements into corresponding positions of multiple electricity demand information generation templates to obtain initial electricity demand information; The initial electricity demand information is semantically optimized and completed to generate the electricity demand information.

[0060] Furthermore, the power demand information is automatically pushed to the corresponding customer manager based on the topology mapping relationship, including: Obtaining a WeChat group ID corresponding to the electricity demand information; According to the topological mapping relationship, determine the unique identifier of the account manager corresponding to the WeChat group identifier; Associating the electricity demand information with the unique identifier of the customer manager to generate an electricity demand information push task; The power demand information push task is triggered to push the power demand information to the customer manager account corresponding to the customer manager's unique identifier.

[0061] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0062] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these changes and variations.

Claims

1. Intelligent identification system for electricity demand in multiple scenarios, It is characterized in that The system comprises: A scene preset module, wherein the scene preset module is used to preset multiple electricity service scenes, wherein each electricity service scene has at least one intent classification label; A model building module, which is used to build a natural language processing model. The natural language processing model is fine-tuned on a pre-trained NLP large model using a labeled data set in the power supply service field. The labeled data set is constructed based on multiple power service scenarios and corresponding intent classification labels; An intention recognition module, the intention recognition module is used to use the natural language processing model to perform intention recognition on the customer expression text to obtain an intention classification result, wherein the customer expression text is sent by a customer in the WeChat group; A demand generation module, the demand generation module is used to extract key information elements from the customer expression text according to the intention classification result, and generate electricity demand information in combination with the intention classification result; A topology mapping module, wherein the topology mapping module is used to establish a topology mapping relationship between the WeChat group and the account manager by using the unique identifier of the account manager as a topology boundary condition; An information push module is used to automatically push the electricity demand information to the corresponding customer manager based on the topology mapping relationship.

2. According to claim 1, the intelligent identification system for power demand in multiple scenarios, It is characterized in that The execution steps of the scene preset module include: Extracting historical electricity service data, analyzing the historical electricity service data, and obtaining multiple electricity service scenarios according to the analysis results; In combination with the historical electricity service data, a corresponding intent classification label is set for each electricity service scenario; Map and store multiple electricity service scenarios with corresponding intent classification labels.

3. According to claim 2, the intelligent identification system for power demand in multiple scenarios, It is characterized in that The execution steps of the model building module include: Based on the pre-trained NLP big model, construct an initial version of the natural language processing model, and add a scene intent classification task layer at the end of the NLP big model; Acquire the annotated data set, and fine-tune the parameters of the natural language processing model using a supervised learning method, wherein the annotated data set is constructed based on historical electricity service data; The performance of the fine-tuned natural language processing model is evaluated, and the natural language processing model is iteratively optimized to obtain a natural language processing model for intent recognition.

4. According to claim 3, the intelligent identification system for power demand in multiple scenarios, It is characterized in that The execution steps of the model building module also include: Extracting historical electricity service data, and for each electricity service scenario, extracting a preset number of customer expression text samples from the historical electricity service data according to the corresponding intent classification label to form candidate data; Manually annotate the candidate data to obtain an initial annotated data set, wherein the initial annotated data set includes a plurality of annotated samples, each of which includes a customer expression text sample, an electricity service scenario sample, an intent classification label sample, and a key information element sample; Performing data enhancement processing on the initial annotated data set to obtain the annotated data set.

5. According to claim 1, the intelligent identification system for power demand in multiple scenarios, It is characterized in that The execution steps of the demand generation module include: Obtaining a preset key information element extraction rule, wherein the key information element extraction rule includes a key information element type corresponding to each intent classification label; Based on the key information element extraction rule, determine the target key information element type corresponding to the customer expression text through the intention classification result; Extract key information elements corresponding to the target key information element type from the customer expression text.

6. According to claim 5, the intelligent identification system for power demand in multiple scenarios, It is characterized in that The execution steps of the demand generation module also include: Extracting multiple target intent classification labels according to the intent classification results; According to multiple target intent classification labels, multiple corresponding electricity demand information generation templates are obtained; Fill the extracted key information elements into corresponding positions of multiple electricity demand information generation templates to obtain initial electricity demand information; The initial electricity demand information is semantically optimized and completed to generate the electricity demand information.

7. The intelligent identification system for power demand in multiple scenarios according to claim 1, It is characterized in that The execution steps of the information push module include: Obtaining a WeChat group ID corresponding to the electricity demand information; According to the topological mapping relationship, determine the unique identifier of the account manager corresponding to the WeChat group identifier; Associating the electricity demand information with the unique identifier of the customer manager to generate an electricity demand information push task; The power demand information push task is triggered to push the power demand information to the customer manager account corresponding to the customer manager's unique identifier.

8. Intelligent identification method of electricity demand for multiple scenarios, It is characterized in that The method comprises: Preset multiple electricity service scenarios, where each electricity service scenario has at least one intent classification label; Build a natural language processing model that is fine-tuned on a pre-trained NLP model using a labeled dataset in the power supply service field, where the labeled dataset is constructed based on multiple power service scenarios and corresponding intent classification labels; Using the natural language processing model to perform intent recognition on the customer expression text to obtain an intent classification result, wherein the customer expression text is sent by a customer in the WeChat group; Extract key information elements from the customer's expression text according to the intention classification result, and generate electricity demand information in combination with the intention classification result; The unique identifier of the account manager is used as the topological boundary condition to establish the topological mapping relationship between the WeChat group and the account manager; The electricity demand information is automatically pushed to the corresponding customer manager based on the topology mapping relationship.

Images