Electric power customer satisfaction evaluation method and system based on customer behavior data mining

Abstract

The invention discloses an electric power customer satisfaction evaluation method and system based on customer behavior data mining, and belongs to the field of data mining, and the method comprises the steps: collecting multi-source heterogeneous data of an electric power customer, carrying out the fusion processing of the multi-source heterogeneous data, extracting customer behavior characteristics and emotion keywords, and generating a recessive emotion fluctuation index; based on the historical behavior data and the current behavior characteristics, a dynamic threshold rule is constructed, client behavior abnormity is monitored in real time, and graded early warning is triggered; a causal reasoning network is constructed according to the customer behavior event sequence, key event paths causing satisfaction change are identified, parameters are dynamically corrected, and a satisfaction correction coefficient is generated for reverse verification; and dividing customer life cycle stages to generate and execute differentiated service strategies. The emotional state of the customer is deeply understood, the dissatisfaction reason of the customer is recognized, an enterprise can quickly respond to the customer satisfaction change, measures are taken in time, customer loss is reduced, and the customer satisfaction is improved.

Description

Technical Field

[0001] The present invention relates to the field of data mining, and in particular to a method and system for evaluating electric power customer satisfaction based on customer behavior data mining. Background Art

[0002] With the rapid development of information technology, especially the continuous advancement of big data and data mining technologies, the application of customer satisfaction assessment in the power industry has been significantly improved. In the past, power companies mainly relied on customer feedback and regular surveys to assess customer satisfaction. However, traditional methods are often limited by insufficient sample representativeness and data acquisition delays, making it difficult to reflect customers' true needs and emotional state in real time. In recent years, the continuous accumulation of customer behavior data (such as payment records, service inquiries, and complaint information) has provided new opportunities for the power industry, making data mining-based satisfaction assessment methods possible. In particular, the collection and analysis of multi-source heterogeneous data has greatly broadened the information dimension, enabling power companies to comprehensively evaluate customer satisfaction from multiple perspectives and gain insight into customers' potential needs.

[0003] Although existing technologies have achieved breakthroughs in customer behavior analysis, several shortcomings remain. First, most existing satisfaction assessment methods fail to effectively integrate data from disparate sources, such as structured data generated by internal business systems and unstructured data from external sources like social media. This lack of integration limits a comprehensive understanding of customer sentiment, as true customer emotions are often expressed not only in formal complaints but also through informal channels like social media and forums. Second, existing methods for extracting customer behavior features and analyzing sentiment often rely on simple sentiment lexicons, failing to deeply explore complex emotional tendencies and their associations with customer behavior, resulting in insufficient ability to identify emotional fluctuations. Furthermore, existing methods often fail to flexibly address dynamic changes in customer behavior, particularly in the event of a sudden drop in satisfaction, and lack effective real-time monitoring and early warning mechanisms. Therefore, our proposed "electricity customer satisfaction assessment method based on customer behavior data mining" offers the potential to provide a comprehensive solution to these shortcomings through multi-dimensional data fusion, deep sentiment analysis, and dynamic monitoring mechanisms. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: how to effectively collect and process multi-source heterogeneous data of power customers to obtain accurate customer behavior characteristics and sentiment analysis, thereby providing a quantitative assessment of customer satisfaction; how to use data mining technology to establish dynamic threshold rules to monitor customer behavior in real time, and promptly discover and warn of abnormal changes in customer satisfaction; and how to formulate differentiated service strategies at different stages of the customer life cycle to improve customer satisfaction and extend the customer life cycle.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for evaluating electric power customer satisfaction based on customer behavior data mining, which includes the following steps:

[0007] Collect multi-source heterogeneous data from electricity customers, including structured data from internal business systems and external unstructured data, and divide the collected multi-source heterogeneous data into text data and voice data; fuse the multi-source heterogeneous data, extract customer behavior characteristics and emotional keywords, and generate an implicit emotion fluctuation index; build dynamic threshold rules based on historical behavior data and current behavior characteristics, monitor customer behavior anomalies in real time and trigger graded warnings; build a causal reasoning network based on customer behavior event sequences, identify the key event paths that lead to changes in satisfaction, dynamically correct parameters, and generate a satisfaction correction coefficient for reverse verification; divide the customer life cycle stages to generate differentiated service strategies and implement them.

[0008] As a preferred solution of the method for evaluating electric power customer satisfaction based on customer behavior data mining according to the present invention, wherein: the fusion processing includes fusion processing of text data and voice data;

[0009] The text data includes identifying sentiment-oriented words in comments through a pre-trained natural language processing model, matching negative sentiment phrases with a standard dictionary of the power industry for semantic enhancement, extracting phrases with negative sentiment orientation, and determining whether to increase the relevance strength of negative sentiment words in the context;

[0010] The voice data includes segmenting the customer service recording into single-round conversation segments, detecting the intervals of sudden increases in speech rate and sudden decreases in pitch in each segment, marking calls with high emotional fluctuations, and calculating the emotional fluctuation intensity value of calls with high emotional fluctuations;

[0011] When both text and voice data come from the same customer, the emotion fluctuation intensity value calculated from the voice is used first; when only a single data type exists, the missing items are filled with the average value of the same type of customer group;

[0012] The frequency of emotional words in the text and the intensity of emotional fluctuations are normalized and superimposed according to a preset ratio to generate an implicit emotional fluctuation index. When the implicit emotional fluctuation index exceeds the preset baseline, the data labeling process is triggered, and manual reviewers review the abnormal data samples.

[0013] As a preferred embodiment of the method for evaluating electricity customer satisfaction based on customer behavior data mining described in the present invention, the dynamic threshold rule construction includes establishing a customer behavior baseline database, conducting stratified sampling based on customer type and electricity consumption scale, defining normal behavior intervals, and calculating behavioral indicator baseline values ​​for each group, including payment punctuality and service sensitivity;

[0014] K-means clustering is performed on each subgroup to generate an initial group. The tertile range of the behavioral indicators of the historical data of the initial group is calculated, and the initial threshold range is set. When the current customer behavior is monitored to deviate from the normal range of the group to which it belongs, the time series prediction model is activated to predict the potential abnormal trend in the next week: if the prediction result shows that the probability of abnormality continues to increase, the threshold range is automatically relaxed; if the prediction result shows that the probability of abnormality continues to decrease, the threshold range is automatically tightened.

[0015] As a preferred embodiment of the method for evaluating electric power customer satisfaction based on customer behavior data mining described in the present invention, the construction of a causal inference network includes abstracting customer behavior events as nodes in the causal inference network, connecting different customer behavior events as edges in the causal inference network, and quantifying the strength of the causal relationship between nodes as edge weights in the causal inference network;

[0016] Paths with weight values ​​higher than a preset threshold are screened from the causal graph and marked as high-risk paths for satisfaction degradation. For isolated nodes that are not covered by any high-risk paths, the current isolated nodes are removed from the causal reasoning network.

[0017] The marked high-risk paths are encoded as feature vectors, and the target area causal reasoning network of the marked high-risk paths is saved. When cross-regional sharing is required and cross-regional model correction is performed, the causal reasoning network containing the high-risk path feature vector is injected first.

[0018] As a preferred embodiment of the method for evaluating electric power customer satisfaction based on customer behavior data mining according to the present invention, the dynamic correction parameters include: when performing cross-regional model correction, extracting the high-risk path feature vector of customers in the target area, calculating the similarity between the high-risk path feature vector and the source area path through cosine similarity, and if the similarity is less than 0.6, performing reinforcement learning on the path edge weight parameters in the model to update the path edge weight parameters;

[0019] The updated path edge weights are input into reverse verification and the parameter change timestamp is locked. If the parameter update causes the correction coefficient to fluctuate by more than 20%, it is rolled back to the previous stable version and a manual review is triggered.

[0020] As a preferred solution of the power customer satisfaction evaluation method based on customer behavior data mining described in the present invention, the generating of the satisfaction correction coefficient for reverse verification includes receiving the updated path edge weight output by the dynamic correction parameter and normalizing and fusing it with the current area infection rate β:

[0021] γ′=γ×(1+W new / W base ×tanh(θ·C active ))

[0022] Among them, W base is the initial edge weight, C active is the number of active complaints, θ is the current regional complaint propagation rate, γ′ is the updated path edge weight coefficient, γ is the updated path edge weight coefficient, and γ max is the maximum coefficient allowed by the system;

[0023] When γ′>1.5γ max When γ′=1.2γ max +0.3×(γ′ / 1+γ′) for nonlinear scaling;

[0024] The corrected coefficients are injected into the causal inference network and the weight values ​​of high-risk paths are recalculated. If the direction of path weight change is inconsistent with the coefficient adjustment direction, a conflict warning is triggered and the coefficient output is frozen.

[0025] As a preferred solution of the power customer satisfaction evaluation method based on customer behavior data mining described in the present invention, wherein: the customer life cycle stages include an introduction period, a stable period, and a decline period;

[0026] The introduction period is when the first electricity payment is completed and there are no complaints. The frequency of collecting customer electricity APP operation heat map data every day is used to calculate daily activity, monitor the click rate of the electricity function module, and judge customer satisfaction based on daily activity and click rate;

[0027] During the stable period, customer satisfaction is judged based on the fluctuation rate of customer electricity consumption and complaint records;

[0028] The decline period defaults to low customer satisfaction. When a customer delays electricity bill payments for two consecutive times or the number of consultations in a single month increases suddenly, satisfaction is considered to have declined, and a high-risk path scan is initiated. If the customer's behavior sequence matches a high-risk path, the customer is marked as an emergency decline customer and a path blocking strategy is pushed in real time.

[0029] If the customer accepts the path blocking strategy, the weight of the associated path will be reduced and the duration of the lifecycle phase will be extended. If the customer does not accept the path blocking strategy or ignores it, the monitoring frequency will be increased to once every 15 minutes, and staff will be prompted to intervene.

[0030] If the success rate of any strategy continues to be lower than 40%, the dynamic correction parameters will be triggered to be forced to be corrected, and customer satisfaction will be continuously judged to see if it recovers.

[0031] Another object of the present invention is to provide an electricity customer satisfaction evaluation system based on customer behavior data mining.

[0032] In order to solve the above technical problems, the present invention provides the following technical solutions: an electric power customer satisfaction evaluation system based on customer behavior data mining, comprising: a data acquisition module, a data fusion module, a dynamic monitoring module, and a satisfaction analysis module;

[0033] The data acquisition module is responsible for collecting multi-source heterogeneous data and performing preliminary data cleaning and preprocessing;

[0034] The data fusion module fuses the collected text data and voice data to generate an implicit emotion fluctuation index for monitoring customer emotion changes;

[0035] The dynamic monitoring module builds dynamic threshold rules to monitor customer behavior in real time, triggers warnings based on behavioral deviations, analyzes potential abnormal trends, and adjusts threshold ranges based on the degree of abnormality;

[0036] The satisfaction analysis module constructs a causal reasoning network, identifies customer critical paths, dynamically modifies critical paths, and generates satisfaction correction coefficients; divides customers according to their life cycle stages, and designs differentiated service strategies to improve customer retention rates.

[0037] The present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program and is characterized in that when the processor executes the computer program, the steps of the method for evaluating electricity customer satisfaction based on customer behavior data mining are implemented.

[0038] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method for evaluating electricity customer satisfaction based on customer behavior data mining are implemented.

[0039] The beneficial effects of the present invention are as follows: By collecting multi-source heterogeneous data from electricity customers, including internal structured data and external unstructured data, the present invention achieves a comprehensive understanding of customer behavior, laying the foundation for subsequent analysis. On this basis, customer behavior characteristics and emotional keywords are extracted through data fusion processing to generate a hidden emotion fluctuation index, which provides a deep understanding of the customer's emotional state, helps identify the reasons for customer dissatisfaction, and improves customer satisfaction. By constructing dynamic threshold rules and monitoring customer behavior anomalies in real time, the company can quickly respond to changes in customer satisfaction and take timely measures to reduce customer churn.

[0040] Furthermore, by building a causal inference network to identify key event paths, the model achieves adaptive capabilities, enabling corrective actions when satisfaction declines and accurately identifying customer pain points. Finally, by categorizing the customer lifecycle stages, differentiated service strategies are generated, enabling personalized service and strengthening customer relationship management, effectively improving customer satisfaction and retention rates and maximizing customer value.

[0041] The advanced data analysis technology of the present invention has built a sensitive customer relationship management system, providing a strong guarantee for improving the service quality and cultivating customer loyalty of the power company. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 This is an overall flow chart of a method for evaluating electric power customer satisfaction based on customer behavior data mining, provided by one embodiment of the present invention.

[0044] Figure 2 This is a system solution module diagram of an electric power customer satisfaction evaluation system based on customer behavior data mining provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0045] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0046] Example 1, reference Figure 1, which is the first embodiment of the present invention, provides a method for evaluating electric power customer satisfaction based on customer behavior data mining, comprising:

[0047] S1: Collect multi-source heterogeneous data from electricity customers, including structured data from internal business systems and external unstructured data, and divide the collected multi-source heterogeneous data into text data and voice data.

[0048] Among them, it collects multi-source heterogeneous data of power customers, including structured data from internal business systems and external unstructured data;

[0049] Structured data from internal business systems includes but is not limited to payment records, fault repair work orders, and electricity consumption time series data;

[0050] External unstructured data includes but is not limited to social media comments, customer service voice recordings, and community forum posts.

[0051] In summary, by collecting multi-source, heterogeneous data from electricity customers (internal structured data and external unstructured data), we achieve a comprehensive understanding of customer behavior. Integrating data from different sources lays the foundation for subsequent analysis, ensuring we capture true customer sentiment and satisfaction. This enables us to more accurately identify customer needs and pain points, providing guidance for improving service.

[0052] S2: Fusing the multi-source heterogeneous data, extracting customer behavior characteristics and emotional keywords, and generating an implicit emotion fluctuation index.

[0053] Furthermore, the fusion processing includes fusion processing of text data and voice data;

[0054] The text data includes using the BERT model to embed context into social media comments, identifying sentiment-oriented words in the comments through a pre-trained natural language processing model, matching negative sentiment phrases (including keywords in the fields of "voltage instability" and "complex payment") with the power industry standard dictionary for semantic enhancement, extracting phrases with negative sentiment orientation, and performing dependency syntactic analysis on the matched phrases. If the negative word is a subject-predicate structure (such as "voltage instability causes equipment damage"), the association strength F of the negative sentiment word in the context is increased. neg .

[0055] The voice data includes segmenting customer service recordings into single-turn conversation segments. Within each segment, the user detects intervals where the speech rate increases (exceeding 20% ​​of the average speech rate) and the pitch decreases (below 30Hz of the baseline). If the proportion of these abnormal intervals exceeds 40% within a call, the call is marked as a call with high emotional fluctuations. The emotional fluctuation intensity value is calculated by multiplying the standard deviation of speech rate by the pitch change rate.

[0056] When both text and voice data come from the same customer, the emotion fluctuation intensity value V calculated by voice is preferred. int ; When there is only a single data type, the missing items are filled with the average value of the customer group of the same type;

[0057] Normalize the frequency of emotional words and the intensity of emotional fluctuations in the text, and superimpose them according to the preset ratio to generate the implicit emotional fluctuation index:

[0058]

[0059] Among them, T neg is the frequency of negative sentiment words in the text data (such as "power outage" and "fault"), F neg is the association strength of negative sentiment words in the context (e.g., “frequent power outages” are stronger than “occasional power outages”), V int is the intensity of emotional fluctuation in speech data, D pause is the ratio of the total duration of silence intervals in the conversation;

[0060] When the implicit emotion fluctuation index exceeds the preset baseline, the data labeling process is triggered, and a manual reviewer will review the abnormal data samples.

[0061] By fusing data, extracting customer behavioral characteristics and emotional keywords, we generate a latent sentiment fluctuation index. This approach aims to leverage advanced natural language processing techniques (such as the BERT model) and sentiment fluctuation detection to gain a deeper understanding of customers' emotional states. By enhancing the correlation between voice and text emotions, we achieve more accurate sentiment analysis, helping identify the causes of customer dissatisfaction and ultimately improving customer satisfaction.

[0062] S3: Based on historical behavior data and current behavior characteristics, dynamic threshold rules are constructed to monitor abnormal customer behavior in real time and trigger graded warnings.

[0063] The dynamic threshold rule construction includes establishing a customer behavior baseline library, wherein customer behavior includes payment operations (success / failure), fault reporting, package changes, customer service consultations, and complaint tickets;

[0064] Conduct stratified sampling by customer type (residential / corporate) and electricity consumption scale (average monthly electricity consumption), define normal behavior intervals (e.g., customers whose bill payments are concentrated between the 1st and 5th of each month), and calculate baseline behavioral indicators for each group, including on-time payment rate and service sensitivity.

[0065] Among them, the payment punctuality rate is the number of on-time payments / total number of payments, and the service sensitivity is the proportion of work orders for complaints that have a response time lower than the average.

[0066] Perform K-means clustering on each subgroup to generate an initial group. Calculate the interquartile range (IQR) of the behavioral indicator for the historical data of the initial group, setting the initial threshold interval to [Q1-1.5IQR, Q3+1.5IQR]. Q1 (first quartile) and Q3 (third quartile) are important indicators for describing data distribution and are used to calculate the interquartile range (IQR), where IQR = Q3-Q1.

[0067] When the current customer behavior is monitored to deviate from the normal range of the group (for example, if the customer still has not paid the bill on the 10th of this month and has not applied for an extension), the time series prediction model is activated to predict potential abnormal trends in the next week based on the behavioral trajectory of the past 6 months:

[0068] B t+1 =α·B t +β·∈ t +γ·X t+1

[0069] Among them, B t+1 is the current behavior indicator value, such as the number of payment delay days or complaints this month; α is the autoregressive coefficient, which represents the historical behavior B t The influence weight on the current value is fitted by maximum likelihood estimation; β is the moving average coefficient, which represents the historical forecast error ∈ t The correction weight of the current value; γ is the external variable coefficient, which quantifies the external factor X t+1 The impact of electricity price adjustment (percentage) or number of extreme weather days; X t+1 is a vector of external variables, including electricity price policies, weather data and other external factors related to customer behavior.

[0070] If the prediction results show that the probability of abnormality continues to increase, the threshold range will be automatically relaxed to [Q1-2.5IQR, Q3+2.5IQR] to avoid false alarms (such as allowing a delay of 3 days). If the prediction results show that the probability of abnormality continues to decrease, the threshold will be tightened to [Q1-1.0IQR, Q3+1.0IQR] to improve detection sensitivity.

[0071] By building dynamic threshold rules and monitoring customer behavior anomalies in real time, we can instantly respond to changes in customer satisfaction and establish a behavioral baseline, facilitating the timely identification of any deviations from normal behavior. By triggering graded alerts, we can quickly implement measures to reduce customer churn, ultimately building a more responsive customer relationship management system and enhancing customer loyalty.

[0072] S4: Construct a causal reasoning network based on the customer behavior event sequence, identify the key event path that leads to changes in satisfaction and dynamically modify the parameters, and generate a satisfaction correction coefficient for reverse verification.

[0073] It should be noted that the construction of the causal inference network includes abstracting customer behavior events as nodes in the causal inference network, connecting different customer behavior events as edges in the causal inference network, and quantifying the causal relationship strength between nodes as the edge weight W in the causal inference network. i→j :

[0074]

[0075] Among them, N i→j N represents the number of times event j occurs after event i occurs. i is the total number of occurrences of event i, T j is the duration of the impact of event j on satisfaction, T avg is the average impact duration of all events;

[0076] Paths with weights above a preset threshold (e.g., "payment failure → payment reminder → customer service consultation → complaint ticket") are filtered from the causal graph and marked as high-risk paths for satisfaction degradation. For isolated nodes not covered by any high-risk paths (e.g., a single electricity bill inquiry operation), the isolated nodes are removed from the causal reasoning network to reduce computational complexity.

[0077] The marked high-risk paths (e.g., “payment failure → payment reminder → complaint”) are encoded into feature vectors, including path length, sum of node weights, and time span.

[0078] The causal reasoning network of the target area marked with high-risk paths is saved. When cross-regional model correction is required, the causal reasoning network containing high-risk paths is injected first to ensure that the model's sensitivity to paths is improved.

[0079] It should also be noted that when performing cross-region model correction, the high-risk path feature vectors of customers in the target region are extracted, and the similarity between the high-risk path feature vectors and the source region paths is calculated using cosine similarity;

[0080] If the similarity is lower than 0.6, the path edge weight parameter in the model is reinforced learning to update the path edge weight parameter W new :

[0081] W new =W old +η·(Reward+λ·maxW future -W old )

[0082] Among them, W old is the original path weight parameter, η is the learning rate, and the control parameter update step size (usually set to 0.01~0.1); Reward is the reward function, calculated as 1-|γ pred -γ real ∣ / γreal , γ pred is the prediction correction coefficient, γ real is the actual value; λ is the discount factor (usually set to 0.9), which measures the current value of future weights; maxW future is the estimated maximum possible weight in the future.

[0083] The updated path edge weights are input into reverse verification and the parameter change timestamp is locked. If the parameter update causes the correction coefficient to fluctuate by more than 20%, it is rolled back to the previous stable version and a manual review is triggered.

[0084] It should also be noted that the generation of the satisfaction correction coefficient for reverse verification includes receiving the updated path edge weight output by the dynamic correction parameter, and normalizing and fusing it with the current area infection rate β to perform coefficient correction:

[0085] γ′=γ×(1+W new / W base ×tanh(θ·C active ))

[0086] Among them, W base is the initial edge weight, C active is the number of active complaints, θ is the current regional complaint propagation rate, γ′ is the updated path edge weight coefficient, γ is the updated path edge weight coefficient, and γ max is the maximum coefficient allowed by the system;

[0087] When γ′>1.5γ max When γ′=1.2γ max +0.3×(γ′ / 1+γ′) for nonlinear scaling to avoid model overfitting;

[0088] The corrected coefficients are injected into the causal inference network and the weight values ​​of high-risk paths are recalculated. If the direction of path weight change is inconsistent with the coefficient adjustment direction, a conflict warning is triggered and the coefficient output is frozen.

[0089] By building a causal inference network and identifying the key event paths that lead to changes in satisfaction, we achieve in-depth analysis of the factors influencing customer satisfaction. By dynamically adjusting the parameters in the path, we achieve the model's adaptive capabilities, enabling timely corrective measures when customer satisfaction declines. This allows companies to accurately identify customer pain points and effectively improve service quality and customer satisfaction.

[0090] S5: Divide the customer life cycle stages to generate differentiated service strategies and implement them.

[0091] The customer life cycle stages include introduction, stabilization and decline;

[0092] The introduction period is when the first electricity payment is completed and there are no complaints. The frequency of collecting customer electricity APP operation heat map data every day is used to calculate daily activity and monitor the click rate of the electricity function module.

[0093] If daily activity continues to increase, or the click-through rate slope is ≥0.1 and there are no negative sentiment words triggered, customer satisfaction is considered to be increasing; if daily activity continues to decrease, or the click-through rate slope is <0.1 and the implicit emotion fluctuation index exceeds the baseline, customer satisfaction is considered to be increasing;

[0094] The stable period is when the customer's electricity consumption fluctuation rate is lower than the threshold for three consecutive months and there is no complaint record, in which case the customer satisfaction is considered to be increasing; otherwise, the customer satisfaction is considered to be decreasing;

[0095] The decline period is when a customer has two consecutive electricity bill payment delays or a sudden increase in the number of consultations in a single month. This indicates a decline in customer satisfaction and initiates a high-risk path scan. If a customer's behavior sequence matches a high-risk path (e.g., payment delay → bill reminder → complaint path), the customer is marked as an emergency decline customer and a path blocking strategy is pushed in real time (e.g., personalized payment reminders sent 48 hours in advance).

[0096] If the customer accepts the path blocking strategy (e.g., clicks on the reminder link and pays), the weight of the associated path is reduced, and the stay time in the life cycle stage is extended;

[0097] If the customer does not accept the path blocking policy or ignores it, the monitoring frequency will be increased to once every 15 minutes, and staff will be prompted to intervene (such as an offline account manager to conduct an on-site inspection);

[0098] If the success rate of any strategy continues to be lower than 40%, the dynamic correction parameters will be triggered forcibly. When the blocking success rate is greater than the set threshold and there are no new complaints in the subsequent 10 days, or the high-risk path matching degree continues to decline, it is considered that the satisfaction has rebounded.

[0099] By categorizing the customer lifecycle stages and generating differentiated service strategies to achieve personalized service, we can flexibly adjust service strategies based on evolving customer behaviors and needs, strengthening customer relationship management. Implementing tailored strategies for customers at different lifecycle stages can effectively improve customer satisfaction and retention, ultimately maximizing customer value.

[0100] Example 2 is the second embodiment of the present invention, which is different from the previous embodiment in that:

[0101] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0102] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0103] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0104] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0105] Example 3, reference Figure 2 , which is the third embodiment of the present invention, provides an electric power customer satisfaction evaluation system based on customer behavior data mining, including: a data acquisition module, a data fusion module, a dynamic monitoring module, and a satisfaction analysis module;

[0106] The data acquisition module is responsible for collecting multi-source heterogeneous data and performing preliminary data cleaning and preprocessing;

[0107] The data fusion module integrates the collected text data and voice data to generate an implicit emotion fluctuation index for monitoring customer emotion changes;

[0108] Dynamic monitoring module: build dynamic threshold rules, monitor customer behavior in real time, trigger warnings based on behavioral deviations, analyze potential abnormal trends, and adjust threshold ranges based on the degree of abnormality;

[0109] The satisfaction analysis module builds a causal reasoning network, identifies customer critical paths, dynamically modifies critical paths, and generates satisfaction correction coefficients; it divides customers according to their life cycle stages and designs differentiated service strategies to improve customer retention rates.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for evaluating electric power customer satisfaction based on customer behavior data mining, characterized by: include, Collect multi-source heterogeneous data from power customers, including structured data from internal business systems and external unstructured data, and divide the collected multi-source heterogeneous data into text data and voice data; Performing fusion processing on the multi-source heterogeneous data, extracting customer behavior characteristics and emotional keywords, and generating an implicit emotion fluctuation index; Based on historical behavior data and current behavior characteristics, dynamic threshold rules are built to monitor abnormal customer behavior in real time and trigger graded warnings; Build a causal reasoning network based on the customer behavior event sequence, identify the key event path that leads to changes in satisfaction, dynamically modify parameters, and generate satisfaction modification coefficients for reverse verification; Divide the customer life cycle stages to generate differentiated service strategies and implement them.

2. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 1, characterized in that: The fusion processing includes fusion processing of text data and voice data; The text data includes identifying sentiment-oriented words in comments through a pre-trained natural language processing model, matching negative sentiment phrases with a standard dictionary of the power industry for semantic enhancement, extracting phrases with negative sentiment orientation, and determining whether to increase the relevance strength of negative sentiment words in the context; The voice data includes segmenting customer service recordings into single-round conversation segments, detecting intervals with sudden increases in speech rate and sudden decreases in pitch in each segment, marking calls with high emotional fluctuations, and calculating the emotional fluctuation intensity value of calls with high emotional fluctuations; When both text and voice data come from the same customer, the emotion fluctuation intensity value calculated from the voice is used first; when only a single data type exists, the missing items are filled with the average value of the same type of customer group; The frequency of emotional words in the text and the intensity of emotional fluctuations are normalized and superimposed according to a preset ratio to generate an implicit emotional fluctuation index. When the implicit emotional fluctuation index exceeds the preset baseline, the data labeling process is triggered, and manual reviewers review the abnormal data samples.

3. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 2, characterized in that: The construction of dynamic threshold rules includes establishing a customer behavior baseline database, conducting stratified sampling by customer type and electricity consumption scale, defining normal behavior intervals, and calculating behavioral indicator baseline values ​​for each group, including payment punctuality and service sensitivity; K-means clustering is performed on each subgroup to generate an initial group. The tertile range of the behavioral indicators of the historical data of the initial group is calculated, and the initial threshold range is set. When the current customer behavior is monitored to deviate from the normal range of the group to which it belongs, the time series prediction model is activated to predict the potential abnormal trend in the next week: if the prediction result shows that the probability of abnormality continues to increase, the threshold range is automatically relaxed; if the prediction result shows that the probability of abnormality continues to decrease, the threshold range is automatically tightened.

4. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 3, characterized in that: The construction of the causal inference network includes abstracting customer behavior events as nodes in the causal inference network, connecting different customer behavior events as edges in the causal inference network, and quantifying the strength of the causal relationship between nodes as edge weights in the causal inference network; Paths with weight values ​​higher than a preset threshold are screened from the causal graph and marked as high-risk paths for satisfaction degradation. For isolated nodes that are not covered by any high-risk paths, the current isolated nodes are removed from the causal reasoning network. The marked high-risk paths are encoded as feature vectors, and the target area causal reasoning network of the marked high-risk paths is saved. When cross-regional sharing is required and cross-regional model correction is performed, the causal reasoning network containing the high-risk path feature vector is injected first.

5. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 4, characterized in that: The dynamic correction parameters include, when performing cross-region model correction, extracting the high-risk path feature vector of customers in the target area, calculating the similarity between the high-risk path feature vector and the source area path through cosine similarity, and if the similarity is lower than 0.6, performing reinforcement learning on the path edge weight parameters in the model to update the path edge weight parameters; The updated path edge weights are input into reverse verification and the parameter change timestamp is locked. If the parameter update causes the correction coefficient to fluctuate by more than 20%, it is rolled back to the previous stable version and a manual review is triggered.

6. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 5, characterized in that: The generation of the satisfaction correction coefficient for reverse verification includes receiving the updated path edge weight output by the dynamic correction parameter and normalizing and fusing it with the current area infection rate β: γ′=γ×(1+W new / W base ×tanh(θ·C active )) Among them, W base is the initial edge weight, C active is the number of active complaints, θ is the current regional complaint propagation rate, γ′ is the updated path edge weight coefficient, γ is the updated path edge weight coefficient, and γ max is the maximum coefficient allowed by the system; When γ′>1.5γ max When γ′=1.2γ max +0.3×(γ′ / 1+γ′) for nonlinear scaling; The corrected coefficients are injected into the causal inference network and the weight values ​​of high-risk paths are recalculated. If the direction of path weight change is inconsistent with the coefficient adjustment direction, a conflict warning is triggered and the coefficient output is frozen.

7. The method for evaluating electric power customer satisfaction based on customer behavior data mining according to claim 6, characterized in that: The customer life cycle stages include introduction, stabilization and decline; The introduction period is when the first electricity payment is completed and there are no complaints. The frequency of collecting customer electricity APP operation heat map data every day is used to calculate daily activity, monitor the click rate of the electricity function module, and judge customer satisfaction based on daily activity and click rate; During the stable period, customer satisfaction is judged based on the fluctuation rate of customer electricity consumption and complaint records; The decline period defaults to low customer satisfaction. When a customer delays electricity bill payments for two consecutive times or the number of consultations in a single month increases suddenly, satisfaction is considered to have declined, and a high-risk path scan is initiated. If the customer's behavior sequence matches a high-risk path, the customer is marked as an emergency decline customer and a path blocking strategy is pushed in real time. If the customer accepts the path blocking strategy, the weight of the associated path is reduced and the stay time in the life cycle stage is extended; If the customer does not accept the path blocking strategy or ignores it, the monitoring frequency will be increased to once every 15 minutes, and staff will be prompted to intervene; If the success rate of any strategy continues to be lower than 40%, the dynamic correction parameters will be triggered to be forced to be corrected, and customer satisfaction will be continuously judged to see if it recovers.

8. An electric power customer satisfaction evaluation system based on customer behavior data mining, applying the electric power customer satisfaction evaluation method based on customer behavior data mining according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, data fusion module, dynamic monitoring module, satisfaction analysis module; The data acquisition module is responsible for collecting multi-source heterogeneous data and performing preliminary data cleaning and preprocessing; The data fusion module fuses the collected text data and voice data to generate an implicit emotion fluctuation index for monitoring customer emotion changes; The dynamic monitoring module builds dynamic threshold rules to monitor customer behavior in real time, triggers warnings based on behavioral deviations, analyzes potential abnormal trends, and adjusts threshold ranges based on the degree of abnormality; The satisfaction analysis module constructs a causal reasoning network, identifies customer critical paths, dynamically modifies critical paths, and generates satisfaction correction coefficients; divides customers according to their life cycle stages, and designs differentiated service strategies to improve customer retention rates.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for evaluating electric power customer satisfaction based on customer behavior data mining according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for evaluating electric power customer satisfaction based on customer behavior data mining according to any one of claims 1 to 7 are implemented.

Images