CN-122022938-A - Electric power full-channel accurate service recommendation method based on dynamic multidimensional user image

CN122022938ACN 122022938 ACN122022938 ACN 122022938ACN-122022938-A

Abstract

The invention provides a power full-channel accurate service recommendation method based on dynamic multidimensional user images, and belongs to the technical field of power service recommendation. The method comprises the steps of extracting a user electricity consumption mode and channel preference through a historical interaction log, grouping similar behaviors by adopting a clustering algorithm, further fusing weather and electricity consumption peak data to predict a change trend, generating a preliminary pushing strategy when the trend exceeds a threshold value, optimizing content and opportunity by combining dynamic user portraits and channel preference, continuously updating the pushing mode through adjusting a parameter set and feedback deviation analysis, and finally forming an accurate service sequence.

Inventors

HU JING
ZHUANG LEI
WANG PIN
WU LINGLING
NI YANYAN
CAO YOUXIA
ZHANG QIAN
WANG YONG
DUAN YUQING
LI XINRAN
WU QIAN
CHANG LI
GONG SHU
Cheng Huanglun
SUN WEIHONG

Assignees

国网安徽省电力有限公司营销服务中心

Dates

Publication Date: 20260512
Application Date: 20251209

Claims (10)

1. The electric power full channel accurate service recommendation method based on the dynamic multidimensional user image is characterized by comprising the following steps of: s1, collecting user behavior data and preference records through multiple contact points, and extracting a power consumption mode and channel use frequency from a historical interaction log to obtain a user behavior data set; s2, grouping similar behavior patterns by adopting a clustering algorithm according to a user behavior data set, analyzing channel preference and interaction frequency, and determining a user cognitive model; s3, acquiring real-time environment data, fusing the environment data aiming at a user cognitive model, and judging a behavior change trend; S4, if the behavior change trend exceeds a preset threshold, generating personalized service content through a recommendation algorithm, and matching energy-saving suggestions and pushing opportunities from user preferences to obtain a preliminary pushing strategy; S5, fusing channel preference analysis of the dynamic multidimensional user portraits and personalized content matching of the electric power full-channel accurate service recommendation according to a preliminary pushing strategy, and determining fusion association parameters; S6, adopting a user cognitive model, taking the fusion associated parameters as a quantization scale, verifying the matching degree of the preliminary pushing strategy and scene perception, extracting inconsistent parts from the preliminary pushing strategy and scene perception, and determining an adjustment parameter set; S7, updating a service pushing mode through adjusting a parameter set to obtain an optimized pushing strategy; S8, fusing behavior trend judgment of the dynamic multidimensional user portraits and feedback deviation analysis of the power full-channel accurate service recommendation according to an optimized pushing strategy, and obtaining associated optimization indexes by combining analysis results; And S9, monitoring user feedback data according to the optimized pushing strategy, analyzing deviation from satisfaction indexes, judging further dynamic adjustment requirements and generating a final service sequence.
2. The method for recommending full-channel accurate services for power based on dynamic multidimensional user images according to claim 1, wherein in step S2, the grouping of similar behavior patterns according to the user behavior data set by using a clustering algorithm comprises: Inputting the electricity utilization mode and channel use frequency in the user behavior data set as feature vectors into a clustering algorithm; Determining a behavior mode group to which a user belongs by calculating the distance between the user behavior data vector and the clustering center feature vector and combining the clustering standard deviation parameter; Calculating a user cognition model score according to the weight coefficient of each behavior mode group; and analyzing channel preference tendency and interaction frequency characteristics of the user based on the user cognitive model score to form quantitative description of the user behavior mode.
3. The power full channel accurate service recommendation method based on dynamic multidimensional user images according to claim 2, wherein the calculation formula of the user cognitive model construction process is as follows: ; Wherein, the Represent the first The cognitive model score of the individual user is calculated, Representing the total number of clustering algorithm groupings, Represent the first The weight coefficients of the individual behavior pattern groups, Represent the first The behavior data vector of the individual user contains channel preferences and interaction frequencies, Represent the first The feature vectors of the individual cluster centers, Represent the first Standard deviation parameters for each cluster.
4. The method for recommending full-channel accurate services for electric power based on dynamic multi-dimensional user images according to claim 1, wherein in step S3, the environmental data includes weather conditions and peak electricity consumption indicators; the judging of the behavior change trend includes: Weighting and calculating a weather condition quantization index, an electricity consumption peak index value and a user behavior characteristic value in a historical time window by fusing weight coefficients; Calculating a threshold judgment value of a behavior change trend according to the change amplitude of the characteristic value of the user behavior at each moment in the length of the historical time window and combining the weather condition and the influence degree of the electricity consumption peak; When the threshold judgment value exceeds a preset threshold range, recognizing that the user behavior mode is changed obviously, and triggering a dynamic adjustment mechanism of the re-matching and pushing strategy of the personalized service content.
5. The method for recommending electric power full channel accurate service based on dynamic multidimensional user images according to claim 1, wherein in step S4, a calculation formula for determining whether the behavior change trend exceeds a preset threshold is as follows: ; Wherein, the A threshold judgment value representing a behavior change trend, 、 And The fusion weight coefficients representing the different factors, A quantified indicator representative of the weather condition, The peak index value of electricity consumption is represented, The length of the historical time window is indicated, And the characteristic value of the user behavior at the t-th moment is represented.
6. The method for recommending electric power full-channel accurate services based on dynamic multidimensional user images according to claim 1, wherein in step S5, the fusion association parameters are calculated based on channel preference weights, content matching degree coefficients, pushing opportunity sensitivity thresholds and scene adaptation degree references.
7. The method for power full channel accurate service recommendation based on dynamic multidimensional user portraits according to claim 1, wherein in step S7, said determining adjustment parameter sets includes channel selection and content optimization; The optimized pushing strategy is obtained through strategy recommendation matching degree score, user feedback satisfaction degree predicted value, adjustment punishment coefficient and difference calculation of the preliminary pushing parameter and the adjusted parameter value.
8. The power full channel accurate service recommendation method based on dynamic multidimensional user images according to claim 7, wherein the calculation formula of the optimized push strategy generation process is as follows: ; Wherein, the Indicating an optimal service push policy is provided, Representing policies Is a recommendation matching degree score of (1), Representing policies Is used for the user feedback satisfaction prediction value, Representing the adjustment of the penalty coefficient, Indicating the total number of adjustment parameters, Represent the first The number of preliminary push parameters is set to be, Represent the first And the adjusted parameter values.
9. The method for recommending full-channel accurate services for electric power based on dynamic multidimensional user images according to claim 1, wherein in step S8, the quantization indexes of the feedback deviation analysis comprise response rate deviation, satisfaction score deviation, behavior transformation deviation, channel adaptation deviation and time response deviation, and the association optimization indexes are calculated based on the quantization indexes of the feedback deviation analysis.
10. The method for power full channel accurate service recommendation based on dynamic multidimensional user images according to claim 1, wherein in step S9, the generating a final service sequence includes: According to the analysis result of the deviation influence factors, a dynamic weight adjustment algorithm is adopted to recalculate the weight coefficients of the portrait matching score, the channel adaptation score and the situation perception score in the recommendation algorithm, when the matching degree deviation of the recommended content is large, the portrait matching score weight is improved, and when the pushing channel effect is poor, the channel adaptation score weight is adjusted, so that optimized recommendation algorithm parameters are obtained; continuously tracking interactive behavior changes of users in various channels through a real-time monitoring module, analyzing recent channel preference change trend of the users by adopting a sliding window mechanism, and triggering a user portrait dynamic updating mechanism to obtain the latest user channel preference label when the use habit of the user channel is changed obviously; Generating an adjusted service recommendation strategy by adopting a recalculated recommendation algorithm according to the optimized recommendation algorithm parameters and the latest user channel preference label, and preferentially distributing high-quality service resources and a proprietary client manager to follow up when the user belongs to a high-value client and the satisfaction degree continuously decreases, so as to obtain a personalized service recommendation scheme; Performing effect verification on the adjusted service recommendation scheme by adopting an A/B test mechanism, and judging that the adjustment strategy is effective when the satisfaction degree improving amplitude of the experimental group exceeds a preset threshold value through the satisfaction degree index comparison analysis of the control group and the experimental group so as to obtain a recommendation strategy passing verification; According to the recommendation strategy passing verification, adopting a full-channel collaborative engine to accurately push the optimized service content according to channels and opportunities preferred by users, and simultaneously recording push tracks and expected effects in a client relationship management system to obtain a complete service push execution record; And continuously collecting response data of a user to the optimized push strategy through a closed loop feedback mechanism, adopting an incremental learning algorithm to integrate new feedback data into a training set of a user portrait model and a recommendation algorithm, and generating a stable final service sequence and solidifying recommendation strategy parameters when the user satisfaction degree of three continuous periods reaches a preset target.

Description

Electric power full-channel accurate service recommendation method based on dynamic multidimensional user image Technical Field The invention relates to the technical field of electric power service recommendation, in particular to an electric power full channel accurate service recommendation method based on dynamic multi-dimensional user images. Background In the field of electric power service, along with the promotion of market innovation and the diversification of user demands, an electric power enterprise gradually changes from traditional production guiding to a service mode taking customers as centers, the transformation has a crucial meaning for improving user satisfaction and enterprise competitiveness, particularly, in a full-channel service environment, how to provide consistent and careful service for users through various contact points becomes a core issue of industry development, however, when the current service mode is in response to complex demands and dynamic changes of the users, a great deal of challenges still remain, and technical innovation is needed to break through bottlenecks. The conventional method has obvious defects in practical application, particularly in the aspects of integrating user information and realizing service accuracy, a plurality of service recommendations often cannot fully sense the real demands of users in different scenes, and the strategies are difficult to adjust in time according to the changes of user behaviors, the limitation causes larger deviation between service contents and user expectations, the user experience and the service efficiency of enterprises are influenced, and the problem is that service pushing often lacks overall consideration of user contact channels, so that resource waste and user trouble are caused. Further analysis can find that two core technical difficulties faced in the field are closely related, namely, how to comprehensively capture and integrate user behaviors and preferences in various service contact points to form a coherent and complete user cognition, user information of different contact points is often scattered and different in format, if the user information cannot be effectively fused, the user needs and habits are difficult to accurately judge, secondly, how to flexibly adjust the timing and modes of service pushing according to the changes of real-time environments and the differences of user preferences on the basis of the user cognition, and if the changes cannot be dynamically adapted, the situation that the service pushing is not matched with the current state of the user can occur, for example, a certain user may need energy-saving advice in the peak period of summer power consumption, but if the service pushing cannot be combined with the current weather condition and the historical behavior characteristics of the user, recommended content may be ignored, and even user dislikes can be caused. How to integrate user behavior information in various service contact points and dynamically adjust service pushing strategies according to real-time environment and user preferences becomes a key problem of improving power service accuracy and user satisfaction, and the problem is solved, so that the optimization of user experience is not only concerned, but also service efficiency and market performance of power enterprises in competition are directly affected, and the power full-channel accurate service recommendation method based on dynamic multidimensional user images is provided to solve the problem. Disclosure of Invention The invention aims to provide a power full-channel accurate service recommendation method based on dynamic multidimensional user images, which aims to solve the problems of how to integrate user behavior information in various service contact points and dynamically adjust service pushing strategies according to real-time environments and user preferences. In order to achieve the purpose, the invention provides the technical scheme that the electric power full channel accurate service recommendation method based on dynamic multidimensional user images comprises the following steps: s1, collecting user behavior data and preference records through multiple contact points, and extracting a power consumption mode and channel use frequency from a historical interaction log to obtain a user behavior data set; s2, grouping similar behavior patterns by adopting a clustering algorithm according to a user behavior data set, analyzing channel preference and interaction frequency, and determining a user cognitive model; s3, acquiring real-time environment data, fusing the environment data aiming at a user cognitive model, and judging a behavior change trend; S4, if the behavior change trend exceeds a preset threshold, generating personalized service content through a recommendation algorithm, and matching energy-saving suggestions and pushing opportunities from user preferences to obtain a preliminary pus