Client complaint risk prediction method, device and equipment based on community multi-dimensional features

Abstract

The invention discloses a customer complaint risk prediction method, device and equipment based on community multi-dimensional features. The method comprises the following steps: acquiring interaction data of power customers through a power community, and acquiring global data of the power customers through a power business system; based on the interaction data and the global data, respectively extracting appeal intensity features, emotion dynamic features, customer background features and behavior pattern features of the power customers; fusing at least one of the appeal intensity feature, the emotion dynamic feature, the customer background feature and the behavior pattern feature to form a feature vector for quantifying the power customer complaint risk; and inputting the feature vector into a pre-trained prediction model to obtain a complaint risk probability value of the power customer output by the prediction model. Through the multi-dimensional data of the power customer, the feature vector for quantifying the complaint risk of the power customer is formed to predict the complaint risk of the power customer, so that the possible complaint behavior is perceived in advance, and the complaint behavior is convenient to deal with in advance.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, apparatus, and equipment for predicting customer complaint risks based on multidimensional community characteristics. Background Technology

[0002] With the continuous development of the electricity market, the increasing diversity of customer needs, and the growing awareness of their rights, electricity customers have higher and higher expectations and requirements for the quality of power supply services.

[0003] The integration of grid services with power communities represents an innovative approach in the power industry, incorporating grid-based management into community services. By dividing power supply areas into dedicated power grids and then establishing or integrating various power communities accordingly, grid service teams can leverage these communities to provide refined electricity services, better meeting the electricity needs of customers and reducing customer complaints.

[0004] However, when faced with massive amounts of interactive data from the power industry community, traditional methods cannot effectively uncover the emotional fluctuations and behavioral patterns behind customer demands, resulting in a lag in the perception of potential customer complaint risks. Summary of the Invention

[0005] The purpose of this invention is to provide a method, device, and equipment for predicting customer complaint risks based on multi-dimensional community features. By obtaining multi-dimensional data of power customers from power communities and power business systems, information such as the intensity of customer demands and emotional dynamics is extracted to form a feature vector that can quantify the complaint risks of power customers, thereby predicting the complaint risks of power customers and solving the problem of delayed perception of potential complaint risks of power customers.

[0006] This invention is achieved through the following technical solution:

[0007] The first aspect of this application provides a method for predicting customer complaint risk based on multi-dimensional community characteristics, including:

[0008] The system acquires interactive data from electricity customers through electricity communities and comprehensive data from electricity customers through electricity business systems; the comprehensive data includes at least basic information of electricity customers, historical service orders, and electricity consumption behavior data.

[0009] Based on the interactive data and global data, the demand intensity characteristics, emotional dynamic characteristics, customer background characteristics and behavioral pattern characteristics of electricity customers are extracted respectively.

[0010] By integrating at least one of the aforementioned characteristics of demand intensity, emotional dynamics, customer background, and behavioral patterns, a feature vector is formed to quantify the risk of electricity customer complaints.

[0011] The feature vector is input into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

[0012] In one feasible implementation, the intensity characteristics of electricity customers' demands are extracted, specifically including:

[0013] Extract text information describing the request from the interaction data;

[0014] Semantic analysis is performed on the text information to determine the level of detail in the electricity customer's description of their request;

[0015] The intensity score of electricity customers' demands is quantified based on the level of detail mentioned above, and used as the feature value of the demand intensity characteristic.

[0016] In one feasible implementation, the emotional dynamics of electricity customers are extracted, specifically including:

[0017] Extract various emotional information that characterizes customer emotions from interaction data within a preset time period;

[0018] Based on the time of issuance of each emotional information, the emotional values ​​of each emotional information are weighted and quantified.

[0019] The weighted result of the combined emotional values ​​is used as the feature value of the emotional dynamic feature.

[0020] In one feasible implementation, behavioral pattern characteristics of electricity customers are extracted, specifically including:

[0021] Information characterizing the behavior of electricity customers is obtained from the interaction data;

[0022] The number of valid behaviors is extracted from the information characterizing electricity customer behavior and used as a feature value of the behavior pattern.

[0023] In one feasible implementation, the method further includes:

[0024] Based on the interaction data of the electricity customer, identify the electricity customer's intent;

[0025] When the intention of the electricity customer is identified as a complaint, various features are extracted.

[0026] In one feasible implementation, identifying the electricity customer's intent based on the customer's interaction data specifically includes:

[0027] Based on the semantic information and emotional information of the electricity customer in the interaction data, the electricity customer's intention is determined.

[0028] In one feasible implementation, after identifying that the electricity customer's intent is to file a complaint, the method further includes:

[0029] The interaction data of electricity customers is input into the topic discovery model to obtain the demand topics output by the topic discovery model; the demand topics represent the core demands of electricity customers.

[0030] Establish a mapping relationship between the core demands and intentions, and store the mapping relationship in the database.

[0031] A second aspect of this application provides a customer complaint risk prediction device based on multi-dimensional community characteristics, comprising:

[0032] The data acquisition unit is used to acquire interactive data of electricity customers through the electricity community and to acquire full-domain data of electricity customers through the electricity business system; the full-domain data includes at least the basic information of electricity customers, historical service work orders and electricity consumption behavior data;

[0033] The feature extraction unit extracts the demand intensity features, emotional dynamic features, customer background features, and behavioral pattern features of electricity customers based on the interaction data and the global data, respectively.

[0034] The feature fusion unit is used to fuse at least one of the following features: demand intensity feature, emotional dynamic feature, customer background feature and behavioral pattern feature, to form a feature vector that quantifies the risk of electricity customer complaints.

[0035] The risk prediction unit is used to input the feature vector into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

[0036] A third aspect of this application provides an electronic device, comprising: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the above-described method.

[0037] A fourth aspect of this application provides a storage medium, comprising: storing a program or instructions on the storage medium, wherein the program or instructions, when executed by a processor, implement the steps of the above-described method.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] This application embodiment obtains multi-dimensional data on electricity customers by acquiring interaction data from electricity communities and full-domain data of electricity customers from the electricity business system. Based on this multi-dimensional data, it extracts features that can influence user complaints, such as the intensity of customer demands, emotional dynamics, customer background, and behavioral patterns, to form a feature vector that quantifies the risk of electricity customer complaints. This allows for the prediction of electricity customer complaint risks, thereby enabling early perception of potential complaint behaviors and facilitating proactive responses. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0041] Figure 1 A flowchart illustrating a customer complaint risk prediction method based on multidimensional community characteristics, provided for an embodiment of this application;

[0042] Figure 2 A flowchart illustrating a specific implementation of a customer complaint risk prediction method based on multi-dimensional community characteristics provided in this application embodiment;

[0043] Figure 3 This application provides a schematic diagram of the structure of a customer complaint risk prediction device based on multi-dimensional community features.

[0044] Figure 4 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for explanation only and are not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0046] As will be known to those skilled in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0047] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, product, or apparatus.

[0048] Example 1

[0049] Embodiment 1 of this application provides a customer complaint risk prediction method based on multi-dimensional community characteristics to solve the current problem of lagging perception of potential complaint risks of electricity customers.

[0050] The subject executing this method can be any computing device capable of implementing the method, such as a server, mobile phone, personal computer, smart wearable device, smart robot, etc.

[0051] Furthermore, the embodiments of this application do not limit the execution order of different steps. When using the method provided in the embodiments of this application, the execution order of different steps can be adjusted according to actual needs.

[0052] For ease of description, the following uses a customer complaint risk prediction device based on multi-dimensional community features as the execution subject of this method to provide a detailed description of the method provided in this application embodiment.

[0053] like Figure 1 The diagram shown is a flowchart illustrating the specific implementation of a customer complaint risk prediction method based on multi-dimensional community features provided in this application, including the following steps 11 to 14:

[0054] Step 11: Obtain interactive data of electricity customers through the electricity community and obtain full-domain data of electricity customers through the electricity business system; the full-domain data includes at least the basic information of electricity customers, historical service work orders and electricity consumption behavior data.

[0055] The power community is centered around "power services," encompassing needs such as electricity consultation, fault reporting, energy-saving advice, and business processing. It includes power companies, power customers, and potential stakeholders, such as residential customers, business customers, property management personnel, and electricians, all of whom can join the power community. Through the power community, power companies can push information to power customers, and communication and collaboration can occur between power customers and between power companies and customers.

[0056] Common application scenarios for electricity communities include: residential electricity communities: embedding electricity services into community owner groups to promptly respond to power outage notices, payment reminders, and share energy-saving tips for home appliances; enterprise electricity communities: providing dedicated power line guarantees, interpretation of electricity pricing policies, and exchange of energy consumption optimization solutions for industrial and commercial users; and special scenario communities: such as new energy vehicle owner communities (focusing on charging pile installation, charging discounts, and electricity safety) and electricity renovation communities for old residential areas (coordinating renovation needs and keeping progress synchronized).

[0057] The electricity community can contain interactive data initiated by electricity customers regarding their electricity needs. This interactive data can include multimodal information such as text messages, images, and emoticons.

[0058] The power business system refers to a series of integrated and professional software applications and information platforms that support the core production and operation activities of power enterprises. Covering all aspects of the power industry, the power business system achieves integrated coordination of power flow, information flow, and business flow through deep integration of information technology and power business, aiming to ensure the safe, stable, and economical operation of the power grid and improve operational efficiency and customer service quality.

[0059] Based on the core business processes of the power grid, the power business system includes multiple business systems such as production monitoring and operation management. This embodiment focuses on systems related to power customers, such as marketing systems and operation platforms. It can collect customer data from marketing systems and operation platforms, covering comprehensive data such as basic customer information, historical service orders, and electricity consumption behavior characteristics.

[0060] Specifically, this could involve: obtaining basic customer information, such as name, home address, and account number, from the customer profile module in the marketing system; obtaining historical service orders, such as order number, order type, order status, and creation time, from the customer service work order module; obtaining customer electricity consumption data, which can be considered as a type of electricity consumption behavior, from the metering over-meter module; and obtaining precise load data and event information (such as power outages and voltage over-limits) from the operation management platform, which can also be considered as a type of electricity consumption behavior.

[0061] In addition, electricity consumption behavior can also be statistical data determined based on historical data from the power business system, such as data characterizing the electricity consumption trend of electricity customers (year-on-year / month-on-month electricity consumption change rate, seasonal electricity consumption patterns, etc.).

[0062] In one feasible implementation, a domain-specific text cleaning rule base is constructed to address the characteristics of the power business scenario, focusing on the identification and processing of power-related professional terms; a domain-adaptive word segmentation method combined with a power dictionary is adopted to ensure accurate segmentation of professional terms; and a unified data standardization process is established to ensure standardized data formats.

[0063] Among them, the text cleaning rule base can be a set of rules that are pre-defined to identify, correct and filter non-standard, redundant and ambiguous information in power text data, based on the characteristics of power business scenarios (such as multi-source data containing power professional terms, colloquial power expressions in social media, etc.), with a core focus on the processing of power professional terms.

[0064] The text cleaning rule base can: accurately identify power-related technical terms (including colloquial and abbreviation forms) in multi-source data (social media texts, work orders, electricity usage records, etc.); standardize terminology (e.g., "power outage" → "power supply interruption," "PMS" → "production management system") to eliminate ambiguity; and filter redundant information unrelated to power business (e.g., casual conversation and garbled text in social media texts). The text cleaning rule base is used to improve data purity, laying the foundation for subsequent word segmentation and analysis.

[0065] The power dictionary integrates standard terminology, equipment names, business vocabulary, common abbreviations, and aliases (such as "distribution transformer," "business expansion application," "SVG," and "integrated automation station") in the power industry. It can be categorized by "equipment," "business," and "parameter" to form a structured power-specific dictionary. Based on this dictionary, the word segmentation process for auxiliary interactive data and global data ensures that specialized terms are not separated.

[0066] The process of word segmentation based on the power dictionary can be as follows: Importing the dictionary: Import the power dictionary into the word segmentation tool (such as jieba or HanLP) and use it as a custom dictionary for priority matching; Long term priority matching: Set rules to allow the word segmentation tool to match long terms in the dictionary first (such as "intelligent substation integrated monitoring system"), avoiding splitting into single characters or short words; Context-assisted correction: For new terms not included in the dictionary (such as new equipment names), combine them with the power business context (such as community repair reports and work order description scenarios) to judge the semantics, avoid missegmentation, and ensure that power professional terms are not split, omitted, or missegmented.

[0067] Data standardization processes can be:

[0068] 1) Establish standards and specifications to clarify the standard data format for multi-source data (social media text, images, emoticons, marketing systems, and operations and maintenance systems). For example, for text: terminology should conform to the cleaning rule base specifications, and word segmentation results should be unified; for numerical values: electricity consumption, voltage, etc. should be retained to two decimal places, and the encoding should be unified (e.g., electricity type "resident = 01"); for dates: unified as "YYYY-MM-DD HH:MM:SS"; for addresses: unified as "province-city-district-street-community-building" format; for images: format is JPG, naming rule is "social media ID_time_type_serial number.jpg"; for emoticons: mapped to "tag + quantitative score", etc.

[0069] 2) Batch data conversion: Perform conversion on multi-source data using scripting tools (such as Python). For example, for unstructured data (text / images): text cleaning → word segmentation → extraction of key fields (problem type, equipment, fault code, etc.); image OCR recognition → extraction of structured information (meter reading, error code); semi-structured data (emojis): convert to tags and scores according to mapping rules; structured data (customer information, work orders): standardize field formats according to format specifications, eliminate redundancy (duplicate work orders) and ambiguity ("repaired" → "completed").

[0070] 3) Validation and correction: This is used to perform standard validations such as formatting on the processed data and to correct data that fails the validation. For example, the program checks whether the data conforms to standards (such as whether the terminology is standardized, whether the date format is correct, and whether key fields are missing), and marks abnormal data; abnormal data (such as unrecognized terms or incorrect formatting) is manually reviewed and corrected.

[0071] 4) Archive storage is used to store standardized data. For example, standardized data can be classified into "power grid-community-customer" levels and stored in a unified data warehouse to achieve full uniformity in data format, meaning, and terminology.

[0072] In one feasible implementation, to quickly identify the electricity customer's intent and reduce memory usage (referring to the content usage of the subsequent complaint risk prediction process), this implementation includes: identifying the electricity customer's intent based on the customer's interaction data. When the customer's intent is identified as a complaint, step 12 is executed to extract various features.

[0073] The process of identifying the intent of an electricity customer can be as follows: determining the intent of an electricity customer based on the semantic information and emotional information of the electricity customer in the interaction data.

[0074] After identifying the intent of the electricity customer, the process may further include: inputting the electricity customer's interaction data into a topic discovery model to obtain the demand topic output by the topic discovery model; the demand topic represents the core demand of the electricity customer; establishing a mapping relationship between the core demand and the determined intent of the electricity customer, and storing the mapping relationship in a database.

[0075] This embodiment uses the LDA (Latent Dirichlet Allocation) topic discovery model. Based on the characteristics of electricity customer demands, a power business keyword guidance mechanism is added to the LDA topic model. By using preset power service keywords to guide the direction of topic discovery, the clustering purity of business subjects such as "electricity billing" and "voltage quality" is improved. The model automatically identifies and summarizes the core demands of customers, such as typical issues like "frequent power outages," "electricity billing questions," and "unstable voltage," and establishes a dynamically updated topic thesaurus.

[0076] Key terms related to the power industry may include:

[0077] Electricity billing: electricity billing, electricity price standards, payment failure, bill questions, tiered pricing; Power supply quality: unstable voltage, low / high voltage, frequent power outages, power outage duration, power restoration; Equipment services: charging pile malfunctions, meter abnormalities, aging lines, circuit breaker tripping, equipment repair; Business processing: transfer of ownership, account cancellation, capacity increase, application for installation, name change.

[0078] Traditional LDA models have low purity in clustering professional domain topics; for example, they easily confuse "electricity bill questions" with "payment channels." By adding a keyword guiding factor to the "topic-word" probability calculation of the LDA model, the bias of guiding keywords towards the topic is ensured.

[0079] During the training phase of the LDA model, a database of electricity business keywords is used as prior knowledge input to force an increase in the probability of keywords appearing in the corresponding topics.

[0080] Based on the trained LDA model, after inputting a text vector formed from interactive data, the output is a demand theme that represents the core demands of electricity customers.

[0081] The confirmation of electricity customer intent can be achieved by: building a classifier based on a pre-trained natural language understanding model, combined with knowledge of the power industry, to distinguish intents such as inquiries, complaints, suggestions, and repair requests, so as to accurately identify different types of customer intents such as inquiries, complaints, suggestions, and repair requests.

[0082] Furthermore, a "topic-intent" association mapping is established, and this association mapping from historical data is stored in the database to facilitate learning in the subsequent intent recognition process and to facilitate accurate understanding of customer intent.

[0083] An example of constructing an association mapping:

[0084] 1) Capture the customer's original statement: "Our community has already experienced three power outages this month. When will this be completely resolved? If this continues, I'm going to make a complaint call."

[0085] 2) Parallel analysis and automatic annotation:

[0086] The LAD topic discovery model analyzes the vocabulary of this statement ("power outage", "three times", "neighborhood") and matches it with known topics. The model calculates that this statement has the highest similarity to the topic "frequent power outages", for example, 95%. Therefore, the model labels this statement with the topic "[topic: frequent power outages]".

[0087] Intent understanding model: This model deeply understands the semantics and tone of the statement ("This has happened three times already," "When will it happen again," "I'm going to complain if this continues"), determining that it contains strong dissatisfaction, urging, and threats. The model determines the intent to be "complain." Therefore, the same statement is labeled with the intent: "complain."

[0088] 3) Establish association mapping: Associate topics and intentions and store them in the database along with the original speech data.

[0089] The final structured data records are shown in the table below:

[0090] Electricity Customer ID Speech content theme intention Customer A "Our neighborhood has already experienced three power outages this month..." Frequent power outages complaint

[0091] Step 12: Based on the interactive data and the global data, extract the demand intensity characteristics, emotional dynamic characteristics, customer background characteristics and behavioral pattern characteristics of electricity customers respectively.

[0092] In this step, the vague feeling of "dissatisfaction" from customers is transformed into a series of quantifiable "risk indicators." Risk characteristics are extracted and quantified from four key dimensions: intensity of demand, emotional dynamics, customer background, and behavioral patterns.

[0093] Among them, the intensity of the demand dimension assesses the specificity of the problem description, the clarity of the solution requirements, and the urgency of the time expression through semantic analysis.

[0094] The extraction of the intensity feature of the demand specifically includes: extracting text information describing the demand from the interaction data; performing semantic analysis on the text information to determine the level of detail in the electricity customer's description of the demand; and quantifying the demand intensity score of the electricity customer based on the level of detail, which is used as the feature value of the demand intensity feature.

[0095] The intensity of a customer's complaint is quantified based on the level of detail. Specifically, it could be: a general description or a vague evaluation, such as "the service is terrible," scoring 0.3 points; stating a specific problem but not providing detailed information such as a home address for feedback, such as "My power went out at 8 pm last night and it hasn't come back on yet," scoring 0.6 points; or describing a specific problem and providing a path for the power company to provide feedback, such as "I live in apartment * in building * of * community on * road in * city. My power went out at 8 pm on January 1st and it hasn't come back on yet," scoring 1 point.

[0096] The emotional dynamic dimension analyzes customers' real-time emotional state, historical emotional accumulation trends, and adversarial behavior during interactions based on an emotional computing model.

[0097] The extraction of emotional dynamic features specifically includes: extracting various emotional information representing customer emotions from interaction data within a preset time period; weighting the emotional values ​​of each emotional information based on the time of its issuance; and fusing the weighted results of the emotional values ​​as the feature values ​​of the emotional dynamic features.

[0098] The preset time period can be the current month.

[0099] The weighting of sentiment values ​​changes linearly based on the time elapsed, with recent sentiment values ​​being given higher weight. In other words, sentiment values ​​over historical periods are subject to linear decay.

[0100] The weighted result of combining various sentiment values, used as the feature value of the sentiment dynamic feature, can be:

[0101] The feature value is calculated as Σ[emotional value * (1 - i / N)], where N is the maximum time window considered (e.g., 30 days, associated with a preset time period). For each day further away from today, the weight decreases linearly by 1 / N.

[0102] The customer background dimension is combined with business data to assess the customer's value level, satisfaction with historical problem resolution, and the business relevance of current needs.

[0103] Business data can include electricity customer electricity consumption behavior data (electricity consumption fluctuations, payment timeliness, load curves), service work order data (historical repair / consultation / complaint records, processing time, solutions), satisfaction data (work order evaluation scores, follow-up feedback), and business processing data (electricity application type, processing frequency, business compliance), etc.

[0104] Example: "3 repair requests in the past year, average processing time of 2 hours, satisfaction rating of 4.8 points, no outstanding payments"—this type of data directly supports the assessment and calculation of customer value level and historical problem resolution satisfaction.

[0105] The assessment of the business relevance of current demands can be quantified by evaluating the degree of alignment between the demands and the core power business.

[0106] The behavioral pattern dimension is quantified by monitoring the discussion intensity of the issue within the group, whether the customer has tagged the account manager, posted pictures as evidence, or spread the issue in different groups.

[0107] The extraction of behavioral pattern characteristics of electricity customers specifically includes: obtaining information representing electricity customer behavior from the interaction data; and extracting the number of effective behaviors from the information representing electricity customer behavior as feature values ​​of behavioral pattern characteristics.

[0108] For each additional valid action, the feature value of the behavioral pattern characteristic increases by 1. For example, @AccountManager + on-site photo, the corresponding behavioral pattern feature value is 2.

[0109] Step 13: Integrate at least one of the following characteristics: intensity of the complaint, emotional dynamics, customer background, and behavioral pattern, to form a feature vector that quantifies the risk of electricity customer complaints.

[0110] After standardizing or normalizing the feature vectors of each dimension, the corresponding dimensions can be selected based on the specific theme (determined by the LDA model in step 13), and the features of each dimension can be weighted and fused to form a feature vector that quantifies the risk of electricity customer complaints.

[0111] Step 14: Input the feature vector into the pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

[0112] The prediction model can be an ensemble learning framework based on gradient boosting decision trees. This learning framework uses "decision trees" as weak learners and "gradient descent" as error correction logic. It constructs multiple decision trees through sequential training (each subsequent tree depends on the result of the previous tree) and finally integrates the prediction results of all trees into a framework.

[0113] During the training of the prediction model, grid search and cross-validation are used to optimize model parameters and ensure the model's generalization ability. Grid search involves listing candidate values ​​for multiple parameters of the gradient boosting decision tree model (e.g., number of trees, learning rate, tree depth), iterating through all parameter combinations, and finding the combination with the best prediction performance. Cross-validation involves splitting the data multiple times to allow the selected parameter combinations to be evaluated more comprehensively and to ensure that the selected parameters are adaptable to different data scenarios.

[0114] The feature vectors determined in step 13 are input into the trained gradient boosting decision tree model. After comprehensive judgment by multiple decision trees, a quantified probability value of the complaint risk is output. A three-level risk classification mechanism is established based on the probability value: a probability value below 0.3 is considered low risk, between 0.3 and 0.7 is medium risk, and above 0.7 is high risk. A confidence assessment mechanism is also set up to specially mark predictions with low confidence to ensure the reliability of the early warning information.

[0115] like Figure 2 As shown, in a specific implementation, the prediction method of this embodiment includes: multi-source data collection: including interaction data, basic customer information, historical service orders, and electricity consumption behavior data; data sample processing: standardizing the collected multi-source data; deep semantic analysis of customer demands: based on multi-source data, determining the customer's demand theme through an LDA model, determining the customer's intent through a constructed classifier, and establishing a theme-intent mapping relationship; construction of multi-dimensional risk feature vectors: extracting the demand intensity features, emotional dynamic features, customer background features, and behavioral pattern features of electricity customers to form a feature vector that quantifies the risk of electricity customer complaints; machine learning risk prediction: using a gradient boosting decision tree as the prediction model, inputting the feature vector into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

[0116] This application embodiment obtains multi-dimensional data on electricity customers by acquiring interaction data from electricity communities and full-domain data of electricity customers from the electricity business system. Based on this multi-dimensional data, it extracts features that can influence user complaints, such as the intensity of customer demands, emotional dynamics, customer background, and behavioral patterns, to form a feature vector that quantifies the risk of electricity customer complaints. This allows for the prediction of electricity customer complaint risks, thereby enabling early perception of potential complaint behaviors and facilitating proactive responses.

[0117] Example 2

[0118] To address the problem of delayed perception of potential complaint risks from electricity customers in existing technologies, and based on the same inventive concept as Embodiment 1, this application also provides a customer complaint risk prediction device based on multi-dimensional community characteristics.

[0119] A schematic diagram of the specific structure of the device is shown below. Figure 3 As shown, it includes the following functional units 301~304:

[0120] The data acquisition unit 301 is used to acquire interactive data of electricity customers through the electricity community and to acquire full-domain data of electricity customers through the electricity business system; the full-domain data includes at least the basic information of electricity customers, historical service work orders and electricity consumption behavior data.

[0121] The feature extraction unit 302 extracts the demand intensity features, emotional dynamic features, customer background features, and behavioral pattern features of the electricity customer based on the interaction data and the global data.

[0122] The feature extraction unit is specifically used to extract the intensity features of the demands of electricity customers, including: extracting text information describing the demands from the interaction data; performing semantic analysis on the text information to determine the level of detail in the electricity customer's description of the demands; and quantifying the intensity score of the electricity customer's demands based on the level of detail, which is used as the feature value of the demand intensity feature.

[0123] The feature extraction unit is specifically used to extract the emotional dynamic features of electricity customers, including: extracting various emotional information that characterizes customer emotions from interaction data within a preset time period; weighting the emotional values ​​of each emotional information based on the time of its issuance; and fusing the weighted results of the emotional values ​​as the feature values ​​of the emotional dynamic features.

[0124] The feature extraction unit is specifically used to extract behavioral pattern features of electricity customers, including: obtaining information representing electricity customer behavior from the interaction data; and extracting the number of valid behaviors from the information representing electricity customer behavior as feature values ​​of behavioral pattern features.

[0125] The feature fusion unit 303 is used to fuse at least one of the following features: demand intensity feature, emotional dynamic feature, customer background feature and behavioral pattern feature, to form a feature vector that quantifies the risk of electricity customer complaints.

[0126] The risk prediction unit 304 is used to input the feature vector into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

[0127] In one implementation, the prediction device of this embodiment preferably includes an intent understanding unit, specifically used to identify the intent of the electricity customer based on the interaction data of the electricity customer. When the intent of the electricity customer is identified as a complaint, a feature extraction unit is invoked to extract various features.

[0128] The intent understanding unit is also used to determine the intent of the electricity customer based on the semantic information and emotion representation information of the electricity customer in the interaction data.

[0129] After identifying the electricity customer's intent, the intent understanding unit is further configured to: input the electricity customer's interaction data into the topic discovery model to obtain the demand topic output by the topic discovery model; the demand topic represents the electricity customer's core demand; establish a mapping relationship between the core demand and the determined electricity customer intent, and store the mapping relationship in the database.

[0130] This application embodiment obtains interactive data of electricity customers from the electricity community and full-domain data of electricity customers from the electricity business system, thereby obtaining multi-dimensional data of electricity customers. Based on this multi-dimensional data, it extracts features that can affect user complaints from multiple dimensions such as the intensity of customer demands, emotional dynamics, customer background and behavioral patterns, forming a feature vector that quantifies the risk of electricity customer complaints, so as to predict the risk of electricity customer complaints, thereby perceiving possible complaint behaviors in advance and facilitating early response.

[0131] Based on the same inventive concept as the foregoing embodiments of this application, this application also provides a computing device.

[0132] like Figure 4 As shown, the computing device includes a memory 41 and a processor 42. The memory 41 can be configured to store various other data to support operation on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device. The memory 41 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0133] The processor 42, coupled to the memory 41, is used to execute the computer program stored in the memory 41 to perform a customer complaint risk prediction method based on multidimensional community features as described in the foregoing embodiments.

[0134] When the processor 42 executes the computer program to perform a customer complaint risk prediction method based on multi-dimensional community features, it obtains multi-dimensional data of power customers by acquiring interaction data of power customers from the power community and full-domain data of power customers from the power business system. Based on this multi-dimensional data, it extracts features that can affect user complaints from multiple dimensions, such as the intensity of customer demands, emotional dynamics, customer background, and behavioral patterns, to form a feature vector that quantifies the risk of power customer complaints, so as to predict the risk of power customer complaints, thereby perceiving possible complaint behaviors in advance and facilitating early response.

[0135] When the processor 42 executes the computer program in the memory 41, in addition to the functions described above, it can also perform other functions, as detailed in the descriptions of the preceding embodiments.

[0136] Furthermore, such as Figure 4 As shown, the computing device also includes other components such as a display 44, a communication component 43, a power supply component 45, and an audio component 46. Figure 4The diagram only shows some components and does not mean that the computing device includes only these components. Figure 4 The components shown.

[0137] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the methods provided in the above embodiments.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0140] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting customer complaint risk based on multidimensional community characteristics, characterized in that, include: We acquire interactive data from electricity customers through electricity communities and comprehensive data on electricity customers through electricity business systems. The comprehensive data includes at least the basic information of electricity customers, historical service orders, and electricity consumption behavior data. Based on the interactive data and global data, the demand intensity characteristics, emotional dynamic characteristics, customer background characteristics and behavioral pattern characteristics of electricity customers are extracted respectively. By integrating at least one of the aforementioned characteristics of demand intensity, emotional dynamics, customer background, and behavioral patterns, a feature vector is formed to quantify the risk of electricity customer complaints. The feature vector is input into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

2. The method according to claim 1, characterized in that, Extracting the intensity characteristics of electricity customers' demands, specifically including: Extract text information describing the request from the interaction data; Semantic analysis is performed on the text information to determine the level of detail in the electricity customer's description of their request; The intensity score of electricity customers' demands is quantified based on the level of detail mentioned above, and used as the feature value of the demand intensity characteristic.

3. The method according to claim 1, characterized in that, Extracting the emotional dynamics of electricity customers, specifically including: Extract various emotional information that characterizes customer emotions from interaction data within a preset time period; Based on the time of issuance of each emotional information, the emotional values ​​of each emotional information are weighted and quantified. The weighted result of the combined emotional values ​​is used as the feature value of the emotional dynamic feature.

4. The method according to claim 1, characterized in that, Extracting behavioral pattern characteristics of electricity customers, specifically including: Information characterizing the behavior of electricity customers is obtained from the interaction data; The number of valid behaviors is extracted from the information characterizing electricity customer behavior and used as a feature value of the behavior pattern.

5. The method according to claim 1, characterized in that, The method further includes: Based on the interaction data of the electricity customer, identify the electricity customer's intent; When the intention of the electricity customer is identified as a complaint, various features are extracted.

6. The method according to claim 5, characterized in that, Based on the interaction data of the electricity customer, the intent of the electricity customer is identified; Specifically, it includes: Based on the semantic information and emotional information of the electricity customer in the interaction data, the electricity customer's intention is determined.

7. The method according to claim 5, characterized in that, After identifying the electricity customer's intent, the method further includes: The interaction data of electricity customers is input into the topic discovery model to obtain the demand topics output by the topic discovery model; the demand topics represent the core demands of electricity customers. Establish a mapping relationship between the core demands and the determined intentions of electricity customers, and store the mapping relationship in a database.

8. A customer complaint risk prediction device based on multi-dimensional community characteristics, characterized in that, include: The data acquisition unit is used to acquire interactive data of electricity customers through the electricity community and to acquire full-domain data of electricity customers through the electricity business system. The comprehensive data includes at least the basic information of electricity customers, historical service orders, and electricity consumption behavior data. The feature extraction unit extracts the demand intensity features, emotional dynamic features, customer background features, and behavioral pattern features of electricity customers based on the interaction data and the global data, respectively. The feature fusion unit is used to fuse at least one of the following features: demand intensity feature, emotional dynamic feature, customer background feature and behavioral pattern feature, to form a feature vector that quantifies the risk of electricity customer complaints. The risk prediction unit is used to input the feature vector into a pre-trained prediction model to obtain the probability value of the electricity customer's complaint risk output by the prediction model.

9. An electronic device, characterized in that, include: It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1-7.

10. A storage medium, characterized in that, include: The storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1-7.

Images