Power business self-service intelligent recommendation method for enterprise users

Abstract

The invention relates to the technical field of data processing, in particular to an enterprise user-oriented power business self-service intelligent recommendation method, which comprises the following steps of: acquiring a static feature vector of each enterprise user and each historical dynamic feature vector when each power business is accessed; determining a historical dynamic feature concentration ratio of each enterprise user; determining the user concentration degree of each power business; determining the recommendation degree of each power business to each enterprise user; and based on the recommendation degree, realizing intelligent recommendation of power business self-service for enterprise users. According to the invention, through double feature construction and clustering analysis, the user demand stability and diversity are accurately measured; business differentiation modeling is realized, and the defect of a unified strategy is avoided; dynamic adjustment of the recommendation degree is achieved through comprehensive evaluation, preference changes are adapted, cold start compensation is provided for new users, and recommendation accuracy and personalization are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to an intelligent recommendation method for self-service electric power services for enterprise users. Background Art

[0002] With the continuous development of information technology, more and more corporate users are choosing to handle their electricity business online. Corporate user groups are remarkably diverse, and their needs are also diverse. A single transaction by a corporate user typically involves only one or a few services. For corporate users, selecting the electricity services they need from a vast array of options can easily lead to inefficient operations and a poor user experience. Therefore, providing accurate service recommendations to corporate users is key to optimizing the user experience.

[0003] A Chinese patent document with announcement number CN113469731B discloses an intelligent recommendation method, platform, and server based on power Internet of Things big data services. The application document obtains user data carried in a service request; determines service data that matches the service request based on the user data, and recommends the determined service data to the target user. This method achieves the purpose of pushing corresponding service information to the user based on consumer preference information.

[0004] However, the above patent document only relies on consumption preference information to recommend user services, fails to deeply explore and analyze the dynamic behavioral characteristics generated by multiple visits of users, and makes it difficult to measure the stability and diversity of user needs; at the same time, there will be obvious differences in the handling of different power services. Some services may be frequently handled by many companies (such as electricity bill payment, with high concentration), and some services may only be occasionally handled by a few companies (such as power equipment upgrade application, with low concentration). It fails to perform differentiated modeling for different power services, resulting in all businesses using the same set of recommendation strategies; and for new users visiting for the first time, there is also a lack of an effective cold start compensation mechanism; in addition, the calculation of the recommendation degree only relies on simple linear calculations, and lacks nonlinear adjustment means for dynamic amplification or smoothing based on the stability of user preferences, making it difficult to adapt to dynamic changes in user preferences. Summary of the Invention

[0005] To address the problems that existing technologies rely solely on consumer preference information to recommend user services, fail to deeply explore and analyze the dynamic behavioral characteristics generated by multiple user visits, and have difficulty measuring the stability and diversity of user needs; fail to perform differentiated modeling for different power services, resulting in the use of the same recommendation strategy for all services; and lack an effective cold start compensation mechanism for new users visiting for the first time; and furthermore, the calculation of recommendation degree relies solely on simple linear calculations, which is difficult to adapt to the dynamic changes in user preferences. This invention proposes an intelligent recommendation method for self-service power services for enterprise users. The method comprises the following steps: Any enterprise user is recorded as a target user. Based on the static features of the target user at the time of registration, the static feature vector of the target user is determined. Based on each historical dynamic feature of the target user when accessing each power service, each historical dynamic feature vector of the target user when accessing each power service is determined. All historical dynamic feature vectors of the target user when accessing all power services are clustered to obtain multiple dynamic feature clusters of the target user. The historical dynamic feature concentration of the target user is determined based on the maximum and minimum values ​​of the distances between the cluster centers of all dynamic feature clusters of the target user. Any power service is recorded as a target service, and the enterprise users who have handled the target service are recorded as service users. The user concentration of the target service is determined based on the mean of the historical dynamic feature concentrations of all service users, the mean of the cosine similarities between the static feature vectors of all service users, and the proportion of the number of times each service user handles the target service to the number of times they handle all power services. The recommendation degree of the target service for the target user is determined based on the number of times the target user handles the target service, the historical dynamic feature vector concentration, and the user concentration. Based on the recommendation degree, intelligent recommendation of power service self-service for enterprise users is achieved.

[0006] The beneficial effects are as follows: by constructing dual features of static feature vectors and historical dynamic feature vectors and using cluster analysis to quantify the concentration of user behavior, an accurate measurement of the stability and diversity of user needs is achieved, overcoming the limitation of traditional methods that only rely on consumption preferences; by calculating the mean historical dynamic feature concentration of the business user group, the mean cosine similarity of the static feature vectors, and the proportion of business handling frequency, a user concentration index for the target business is constructed, which realizes differentiated modeling of different power businesses and avoids the drawbacks of a unified recommendation strategy; by introducing a comprehensive evaluation of the historical dynamic feature concentration and user concentration, a nonlinear dynamic adjustment of the recommendation degree is achieved, which can dynamically amplify or smooth according to the stability of user preferences and effectively adapt to the dynamic changes of user preferences; through the group similarity analysis of static feature vectors and the statistical characteristics of business handling, an effective cold start compensation mechanism is provided for new users, and reasonable recommendations are made based on the behavioral patterns of similar users, significantly improving the initial service experience of new users; the complete recommendation framework constructed based on dynamic feature clustering, user concentration analysis and nonlinear recommendation degree calculation realizes intelligent recommendation of enterprise users' power business self-service, improving the accuracy and personalization of recommendations.

[0007] Furthermore, the static features include enterprise size, electricity consumption type, industry classification, region, and remaining electricity.

[0008] Furthermore, the status features include the length of stay, keyword input and visit period.

[0009] Furthermore, determining the static feature vector of the target user includes: combining the normalization result of the continuous data in the static features of the target user at the time of registration with the encoding result of the category data in the static features of the target user at the time of registration to obtain the static feature vector of the target user.

[0010] Furthermore, the determining of each historical dynamic feature vector of the target user when accessing each power service includes: combining the normalization result of the continuous data in each historical dynamic feature of the target user when accessing each power service with the encoding result of the category data in each historical dynamic feature of the target user when accessing each power service, to obtain each historical dynamic feature vector of the target user when accessing each power service.

[0011] Furthermore, the clustering adopts a mean shift clustering algorithm.

[0012] Furthermore, the historical dynamic feature concentration satisfies: Where, is the historical dynamic feature concentration of the target user, is the maximum value of the distances between the cluster centers of all dynamic feature clusters of the target user, is the minimum value of the distance between the cluster centers of all dynamic feature clusters of the target user, for Type curve function.

[0013] The beneficial effect is that by calculating the difference between the maximum and minimum distances between the centers of all dynamic feature clusters of the target user, and using Function is used for mapping, which realizes the standardized quantification of historical dynamic feature concentration; Function The characteristic of the curve limits the concentration range to range, which is convenient for subsequent calculation and comparison.

[0014] Furthermore, the user concentration satisfies: Where, For the The user concentration of each power business, For all those who have handled The mean of the historical dynamic characteristic concentration of corporate users of electricity business, For all those who have handled The mean value of the cosine similarity between the static feature vectors of the enterprise users of the power business, For all those who have handled The number of corporate users of electricity services, For all those who have handled Among the corporate users of electricity business Enterprise users handle the The number of electricity business For all those who have handled Among the corporate users of electricity business The number of times a corporate user handles all electricity business.

[0015] The beneficial effects are: by comprehensively considering the three dimensions of the historical dynamic feature concentration mean, static feature similarity mean and business handling frequency ratio of business users, a comprehensive evaluation model for the concentration of power business users is constructed, which can accurately reflect the user group characteristics and business handling rules of different power businesses, and realize the differentiated characterization of high-frequency and low-frequency businesses; the homogeneity of the user group is measured by the cosine similarity mean, and the popularity of the business is reflected in combination with the handling frequency ratio, making the user concentration index more comprehensive and objective.

[0016] Furthermore, the recommendation degree satisfies: Where, For the The recommendation degree of each power business for the target user, For target users The number of electricity business is the historical dynamic feature concentration of the target user, For the The user concentration of each power business, is a linear normalization function.

[0017] The beneficial effects are: by performing a power operation on the normalized value of the number of user historical behaviors and the dynamic feature concentration, and adding it to the business user concentration, a nonlinear recommendation degree calculation model is constructed, which can be dynamically adjusted according to the different stability of user behavior. When the user behavior concentration is high, the power function amplifies the influence of the user's historical behavior and strengthens the recommendation effect. When the user behavior concentration is low, the power function is smoothed to avoid over-reliance on unstable historical data; by introducing the business user concentration, the group characteristics of the business itself are comprehensively considered, so that the recommendation results reflect both individual preferences and business characteristics; linear normalization processing ensures the comparability of data of different dimensions and improves the accuracy and stability of the recommendation degree calculation.

[0018] Furthermore, the intelligent recommendation of electricity business self-service for corporate users includes: sorting the recommendation degree of all power properties for target users from large to small, selecting a preset number of multiple electricity businesses in order as top recommendations for target users, and completing the intelligent recommendation of electricity business self-service for corporate users.

[0019] The beneficial effects are: by sorting all power businesses from high to low according to the degree of recommendation, and selecting a preset number of businesses for top recommendation, the orderly display and accurate push of recommendation results are achieved, which can intuitively provide users with the most relevant business recommendations and improve the efficiency of users in finding and handling businesses; by controlling the preset number, it not only ensures the comprehensiveness of the recommendations, but also avoids information overload and improves the user experience; the sorting mechanism ensures the priority display of high-priority businesses, so that the recommendation results are more in line with the actual needs of users and improves the intelligence level of business self-service.

[0020] The present invention has the following beneficial effects: (1) It breaks through the limitation of existing technology that only relies on consumption preference information. By clustering the historical dynamic feature vectors generated by multiple visits of users, it generates dynamic feature clusters and calculates the concentration of historical dynamic features. It quantifies the stability (high concentration indicates focused demand) and diversity (low concentration indicates dispersed demand) of user needs. It can capture the dynamic changes in user behavior, avoid recommendation bias caused by the one-sidedness of static preference information, and make the recommendation results more in line with the real needs of users.

[0021] (2) By calculating the user concentration of the target business, accurately distinguish high concentration business (such as electricity payment, handle user, high frequency) and low concentration business (such as power equipment upgrading application, handle user, low frequency), based on the difference of business characteristics, formulate targeted recommendation strategy, avoid the same recommendation mode, for example, for high concentration business, focus on the stable demand of high frequency user, for low concentration business, focus on the mining of potential demand, significantly improve the adaptability of different types of power business recommendation.

[0022] (3) For the first time, the new user accesses, although it lacks historical dynamic behavior, but can participate in the calculation of business user concentration through the static feature vector at the time of registration (such as using the mean of static feature similarity), providing a basis for cold start recommendation, compared with the blank of existing technology for new user recommendation, the present application effectively fills the lack of cold start compensation mechanism, expands the coverage of intelligent recommendation.

[0023] (4) The recommendation degree calculation fuses the historical frequency of user handling business, the dynamic feature concentration of itself (reflecting the demand stability) and the user concentration of business (reflecting the business universality), forming a nonlinear adjustment mechanism; for example, for the user with stable demand (high dynamic concentration), the recommendation weight of high frequency handling business can be amplified; for the user with diverse demand (low dynamic concentration), the recommendation weight can be smoothed to cover more potential business, so that the recommendation result can dynamically adapt to the change of user preference, improving the flexibility and accuracy of recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a step flow chart of a power business self-service intelligent recommendation method for enterprise users according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below. The described embodiments are part of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0026] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0027] Please refer to Figure 1 , which shows a step flow chart of a power business self-service intelligent recommendation method for enterprise users according to an embodiment of the present application. The method comprises the following steps: S1: Obtain the static feature vector of each enterprise user and each historical dynamic feature vector when accessing each power business.

[0028] Any corporate user is recorded as a target user. Based on the target user's static characteristics at the time of registration (including corporate size, electricity consumption type, industry classification, region, and remaining power), the target user's static feature vector (including stay duration, keyword input, and visit time period, etc.) is determined. Based on each historical dynamic feature of the target user when visiting each power business, each historical dynamic feature vector of the target user when visiting each power business is determined.

[0029] Specifically, determining the static feature vector of the target user includes: Normalize the continuous data of the static features of the target user at the time of registration (such as maximum and minimum normalization) and encode the category data of the static features of the target user at the time of registration (such as one-hot encoding. Assuming that a static feature has 4 categories, for example, the type of electricity consumption is divided into 4 categories: residential, commercial, industrial, and agricultural. The first category is encoded as , where the second category is coded as , and so on) to combine the results (for example, after normalization of enterprise size, , the electricity usage type is coded as , then the static eigenvector is ), and obtain the static feature vector of the target user.

[0030] Specifically, determining each historical dynamic feature vector of the target user when accessing each power service includes: The normalized results of the continuous data in each historical dynamic feature when the target user accesses each electricity service (including online payment, application, bill inquiry, fault repair, and energy efficiency analysis report application, etc.) are combined with the encoded results of the category data in each historical dynamic feature when the target user accesses each electricity service to obtain each historical dynamic feature vector of the target user when accessing each electricity service.

[0031] S2: Determine the historical dynamic feature concentration of each enterprise user.

[0032] It should be noted that different corporate users have different business needs for the power business platform, and the realization of different business needs corresponds to different dynamic characteristics. If the historical dynamic characteristics of the corporate users are all the same, then the next time the corporate users visit the power business platform, the power business corresponding to the historical dynamic characteristics can be placed in a priority position for the corporate users to choose. However, in fact, for an corporate user, the dynamic characteristics generated each time they visit the power business platform are not exactly the same, and the power business they conduct is not exactly the same. Therefore, it is impossible to simply recommend power business based on the dynamic characteristics of the corporate users. In order to avoid the physical examination results being out of touch with the real needs of the users, this step obtains the concentration of the historical dynamic characteristics of the corporate users based on the historical dynamic characteristics of the corporate users in each power business.

[0033] All historical dynamic feature vectors of the target user when accessing all power services are clustered to obtain multiple dynamic feature clusters of the target user. The historical dynamic feature concentration of the target user is determined based on the maximum and minimum values ​​of the distances between the cluster centers of all dynamic feature clusters of the target user.

[0034] Specifically, the clustering adopts a mean shift clustering algorithm.

[0035] Specifically, the historical dynamic feature concentration satisfies: ; Where, is the historical dynamic feature concentration of the target user, is the maximum value of the distances between the cluster centers of all dynamic feature clusters of the target user, is the minimum value of the distance between the cluster centers of all dynamic feature clusters of the target user, for Type curve function.

[0036] in, Represents the gap between the dynamic feature vectors of the target user. The larger the value, the more dispersed the dynamic feature vectors of the target user are, and the smaller the historical dynamic feature concentration of the target user is; the smaller the value, the closer the dynamic feature vectors of the target user are, and the larger the historical dynamic feature concentration of the target user is. In order to facilitate subsequent calculations, Curve function pair Normalize, where , For users who access the power business platform for the first time, since they have no historical dynamic feature vectors, their historical dynamic feature vector concentration is set to 0 to facilitate subsequent calculations.

[0037] S3: Determine the user concentration of each power business.

[0038] It should be noted that, since different electricity services have different functions, different corporate users have different needs for electricity services. Some services may be handled by most corporate users (such as payment). For these services, priority should be given to most corporate users. Some services may only be handled by a few types of corporate users. For these electricity services, it is only necessary to give priority to these types of corporate users. Therefore, in order to measure which electricity services need to be widely recommended and which electricity services need to be recommended to special types of corporate users, this step obtains the user concentration of each electricity service based on the historical dynamic feature concentration of corporate users and the historical electricity business handling records of corporate users.

[0039] Any electricity business is recorded as the target business, and the corporate users who have handled the target business are recorded as business users. The user concentration of the target business is determined based on the mean of the historical dynamic feature concentration of all business users, the mean of the cosine similarity between the static feature vectors of all business users, and the proportion of the number of times each business user handles the target business to the number of times they handle all electricity businesses.

[0040] Specifically, the user concentration satisfies: ; Where, For the The user concentration of each power business, For all those who have handled The mean of the historical dynamic characteristic concentration of corporate users of electricity business, For all those who have handled The mean value of the cosine similarity between the static feature vectors of the enterprise users of the power business, For all those who have handled The number of corporate users of electricity services, For all those who have handled Among the corporate users of electricity business Enterprise users handle the The number of electricity business For all those who have handled Among the corporate users of electricity business The number of times a corporate user handles all electricity business.

[0041] in, Representatives have handled The degree of similarity of static characteristics between enterprise users of electricity business. The larger the value, the more The more similar the enterprise users of the electricity business are, the more likely they are to The more likely the corporate users of electricity business are to be concentrated on a certain type of corporate users, the more likely The greater the user concentration of the electricity business, the smaller the value, the more The greater the difference between the corporate users of the electricity business, the more The more likely a business user of a power business is to be a business user of various types, the more likely The smaller the user concentration of an electricity business. Representatives have handled The similarity of access habits between enterprise users of power business. The larger the value, the more similar the access habits between enterprise users of power business. The more similar the access habits of the enterprise users of the power business platform are, the more likely they are to handle the first The more likely the corporate users of electricity business are to be concentrated on a certain type of corporate users, the more likely The greater the user concentration of the electricity business, the smaller the value, the more The greater the difference in access habits of corporate users of power business on the power business platform, the more The more likely the enterprise users of power business are to be enterprise users of various types, the more likely they are to be enterprise users of various types. The smaller the user concentration of the power business, the smaller the The weights of , implementers can adjust the weights according to actual conditions Representatives have handled electricity business of corporate users have electricity business The larger the value, the more The more the corporate user of a power business tends to handle the power business, the more likely the The greater the user concentration of the electricity business, the smaller the value, the more The less the corporate user of a power business has the tendency to handle the power business, the The smaller the user concentration of an electricity business.

[0042] S4: Determine the recommendation level of each power service for each corporate user.

[0043] It should be noted that the historical dynamic feature concentration of corporate users can reflect the users' access habits to the power business platform, and the user concentration of power business reflects the access tendency of various types of corporate users to different power businesses. Since the business handling of corporate users is affected by both individual factors (such as electricity bill payment) and overall factors (such as various factors leading to the industry as a whole handling a certain business), this step obtains the recommendation degree of each power business for each corporate user based on the user concentration of each power business and the historical dynamic feature concentration of each corporate user.

[0044] The recommendation degree of the target service to the target user is determined based on the number of times the target user handles the target service, the historical dynamic feature vector concentration, and the user concentration.

[0045] Specifically, the recommendation degree satisfies: ; Where, For the The recommendation degree of each power business for the target user, For target users The number of electricity business is the historical dynamic feature concentration of the target user, For the The user concentration of each power business, is a linear normalization function.

[0046] in, Represents the target user The historical handling frequency of the electricity business, the larger the value, the more the target user The more inclined the handling of a power business is, the The greater the recommendation of the first power business to the target user, the smaller the value is, the higher the target user's recommendation of the first power business is. The less biased the handling of a power business is, the The lower the recommendation of a power business to the target user, the lower the historical dynamic feature concentration of the target user. The larger the value is, the more stable the target user's business handling preference is. Therefore, the more frequently handled businesses should be recommended to the target customer, so that the recommendation degree of the target customer's frequently handled electricity business is higher than that of other electricity businesses. Therefore, through In the form of exponential function Make adjustments, When the value is larger, the power business that the target user frequently handles is more recommended to the target user relative to other power businesses; The smaller the value, the closer the recommendation of each power business to the target user. is 0, which makes it impossible to effectively recommend power services. To supplement the recommendation degree, the electricity services that corporate users frequently handle as a whole are recommended to new target users.

[0047] S5: Based on the recommendation degree, intelligent recommendation of electricity business self-service for corporate users is realized.

[0048] Specifically, the intelligent recommendation for self-service electricity services for corporate users includes: All power properties are sorted from the highest to the lowest in terms of their recommendation for target users, and a preset number of power businesses are selected in order as top recommendations for target users, completing the intelligent recommendation of power business self-service for corporate users.

[0049] Implementers can set the preset number according to the specific implementation situation, for example, 3.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent recommendation method for self-service electricity business for enterprise users, characterized in that: include: Record any enterprise user as a target user, determine the target user's static feature vector based on the target user's static features when registering, and determine each historical dynamic feature vector of the target user when accessing each power service based on each historical dynamic feature of the target user when accessing each power service; Clustering all historical dynamic feature vectors of the target user when accessing all power services to obtain multiple dynamic feature clusters of the target user, and determining the historical dynamic feature concentration of the target user based on the maximum and minimum values ​​of the distances between the cluster centers of all dynamic feature clusters of the target user; Record any electricity business as the target business, and record the corporate users who have handled the target business as business users. Determine the user concentration of the target business based on the mean of the historical dynamic feature concentration of all business users, the mean of the cosine similarity between the static feature vectors of all business users, and the proportion of the number of times each business user has handled the target business to the number of times they have handled all electricity businesses. Determining the recommendation degree of the target service for the target user based on the number of times the target user has handled the target service and the historical dynamic feature vector concentration, as well as the user concentration; Based on the recommendation degree, intelligent recommendation of electricity business self-service for corporate users is achieved.

2. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1 is characterized in that: The static features include enterprise size, electricity consumption type, industry classification, region, and remaining power.

3. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1 is characterized in that: The dynamic features include stay time, keyword input and visit time period.

4. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: Determining the static feature vector of the target user includes: The normalization result of the continuous data in the static features of the target user at the time of registration is combined with the encoding result of the category data in the static features of the target user at the time of registration to obtain the static feature vector of the target user.

5. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1 is characterized in that: The determining of each historical dynamic feature vector of the target user when accessing each power service includes: The normalized result of the continuous data in each historical dynamic feature when the target user accesses each power service is combined with the encoding result of the category data in each historical dynamic feature when the target user accesses each power service to obtain each historical dynamic feature vector of the target user when accessing each power service.

6. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: The clustering adopts the mean shift clustering algorithm.

7. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: The historical dynamic characteristic concentration satisfies: ； Where, is the historical dynamic feature concentration of the target user, is the maximum value of the distances between the cluster centers of all dynamic feature clusters of the target user, is the minimum value of the distance between the cluster centers of all dynamic feature clusters of the target user, for Type curve function.

8. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: The user concentration satisfies: ； Where, For the The user concentration of each power business, For all those who have handled The mean of the historical dynamic characteristic concentration of corporate users of electricity business, For all those who have handled The mean value of the cosine similarity between the static feature vectors of the enterprise users of the power business, For all those who have handled The number of corporate users of electricity services, For all those who have handled Among the corporate users of electricity business Enterprise users handle the The number of electricity business For all those who have handled Among the corporate users of electricity business The number of times a corporate user handles all electricity business.

9. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: The recommendation degree meets: ； Where, For the The recommendation degree of each power business for the target user, For target users The number of electricity business is the historical dynamic feature concentration of the target user, For the The user concentration of each power business, is a linear normalization function.

10. The method for intelligent recommendation of electric power business self-service for enterprise users according to claim 1, characterized in that: The intelligent recommendation of electricity business self-service for enterprise users includes: All power properties are sorted from the highest to the lowest in terms of their recommendation for target users, and a preset number of power businesses are selected in order as top recommendations for target users, completing the intelligent recommendation of power business self-service for corporate users.

Images