Recommendation scheme generation method, device and equipment for customer service system

Abstract

The invention provides a recommendation scheme generation method, device and equipment for a customer service system, and relates to the technical field of data processing, and the method comprises the steps: carrying out the clustering analysis of users and recommendation contents in user data, and obtaining a plurality of recommendation schemes and user groups corresponding to the recommendation schemes; based on each recommendation scheme and the corresponding user group, generating a training set of recommendation scheme satisfaction estimation; based on the training set of each recommendation scheme, determining an evaluation value of the recommendation scheme by adopting tendency score matching and average causal effect evaluation; and through Pearson similarity calculation, determining user characteristics adapted to each first recommendation scheme. Therefore, a user group with the same recommendation scheme characteristic is screened out, a training set is formed, effective recommendation schemes in the recommendation schemes are evaluated through tendency score matching and an average causal effect, finally, user characteristics matched with the recommendation schemes are determined through Pearson similarity calculation, and analysis support is provided for recommendation content decision making of a customer service system.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device and equipment for generating a recommendation solution for a customer service system. Background Art

[0002] When users enter the customer service interface of the customer service system, they will be shown many recommended contents of functions and configurations, such as user operation instructions, business knowledge, page shortcut function display, etc. With the development of the customer service system, the recommended contents are becoming more and more abundant and the dimensions are getting higher and higher. Different recommended contents can be combined to form a variety of solutions.

[0003] At present, business personnel often determine the recommended combination scheme based on experience, but lack user data support. Therefore, obtaining potential effective recommended combination schemes from a large amount of real system data in the customer service system is of great significance for improving the service quality of the customer service system. Summary of the invention

[0004] In view of this, the present application provides a method, device and equipment for generating a recommendation plan for a customer service system, aiming to set up an effective recommendation combination plan and improve the service quality of the customer service system.

[0005] In a first aspect, a method for generating a recommendation solution of a customer service system includes:

[0006] Perform cluster analysis on users and recommended contents in user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme;

[0007] Based on each recommendation scheme and its corresponding user group, generating a training set of satisfaction estimation of the recommendation scheme;

[0008] Based on the training set of each recommended solution, the evaluation value of the recommended solution is determined by using propensity score matching and average causal effect evaluation;

[0009] The user characteristics that are adapted to the recommendation scheme when the evaluation value exceeds the preset value through Pearson similarity calculation.

[0010] Optionally, clustering analysis is performed on users and recommended contents in the user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme, including:

[0011] Based on the characteristics of the user and the characteristics of the recommended content, a bipartite graph of user recommended content characteristics is established;

[0012] Establish the adjacency matrix of the bipartite graph of user recommendation content features;

[0013] Solving the generalized eigenvector of the adjacency matrix to obtain a characteristic matrix;

[0014] The feature matrix is ​​clustered using a K-means algorithm to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

[0015] Optionally, generating a training set of satisfaction estimation of the recommendation schemes based on each recommendation scheme and its corresponding user group includes:

[0016] Get a recommendation solution without generating a training set;

[0017] The intervention variable indicating the satisfaction estimate configured in the first user data is set to 1, and the intervention variable configured in the second user data is set to 0, the first user data and the second user data are used as training sets of the recommendation scheme, the first user data is user data included in the user group corresponding to the recommendation scheme, and the second user data is user data included in the user group not corresponding to the recommendation scheme; the step of obtaining a recommendation scheme for which a training set has not been generated is continued until there is no recommendation scheme for which a training set has not been generated.

[0018] Optionally, the method of determining the evaluation value of the recommended solution by using propensity score matching and average causal effect evaluation based on the training set of each recommended solution includes:

[0019] Based on each user data, a logistic regression calculation is performed with the user characteristics as the independent variable and whether the recommended solution is accepted as the dependent variable to obtain the propensity score of the user corresponding to the user data to accept the recommended solution;

[0020] Based on the propensity score, a one-to-one nearest neighbor matching method is used to match users in the training set who accept the recommended solution and users who do not accept the recommended solution, and the users who accept the recommended solution are formed into a recommended solution group, and the users who do not accept the recommended solution are formed into a non-recommended solution group;

[0021] The average causal effect evaluation value between user acceptance of the recommended solution and user satisfaction is calculated as the evaluation value of the recommended solution.

[0022] Optionally, the user characteristics adapted by the recommendation scheme whose evaluation value calculated by Pearson similarity exceeds a preset value include:

[0023] Calculating a similarity value of each user feature between the recommendation solution group and the unaccepted recommendation solution group;

[0024] User features with similarity values ​​less than a threshold are selected as user features adapted by the recommendation scheme.

[0025] Optionally, before clustering the users and recommended contents in the user data, the method further includes preprocessing the user data;

[0026] The preprocessing of the user data includes: data cleaning, missing value processing and feature extraction of the user data.

[0027] In a second aspect, the present application further provides a recommendation solution generating device for a customer service system, the device comprising:

[0028] A division unit, used to perform cluster analysis on users and recommended contents in the user data, and obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme;

[0029] A training set generating unit, used for generating a training set of satisfaction estimation of the recommendation schemes based on each recommendation scheme and its corresponding user group;

[0030] An evaluation unit, used to determine the evaluation value of the recommended scheme based on the training set of each recommended scheme by using propensity score matching and average causal effect evaluation;

[0031] The analyzing unit is used to calculate the user characteristics of the recommendation scheme adapted by the evaluation value exceeding the preset value through Pearson similarity calculation.

[0032] Optionally, the division unit is specifically used to establish a bipartite graph of user recommendation content features based on user features and recommended content features; establish an adjacency matrix of the bipartite graph of user recommendation content features; solve the generalized eigenvector of the adjacency matrix to obtain a feature matrix; and use a K-means algorithm to cluster the feature matrix to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

[0033] Optionally, the evaluation unit is specifically used to perform logistic regression calculation based on each user data, with user characteristics as independent variables and whether to accept the recommended plan as the dependent variable, to obtain a propensity score for the user corresponding to the user data to accept the recommended plan; based on the propensity score, a one-to-one nearest neighbor matching method is used to match users who accept the recommended plan and users who do not accept the recommended plan in the training set, and the users who receive the recommended plan are formed into a recommended plan group, and the users who do not accept the recommended plan are formed into a non-recommended plan group; the average causal effect evaluation value between user acceptance of the recommended plan and user satisfaction is calculated as the evaluation value of the recommended plan.

[0034] In a third aspect, the present application also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes any one of the methods for generating a recommendation solution for a customer service system described above.

[0035] The present application provides a method, device and equipment for generating a recommendation scheme for a customer service system. The present application includes clustering analysis of users and recommended contents in user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme; generating a training set for estimating the satisfaction of the recommendation scheme based on each recommendation scheme and its corresponding user group; determining the evaluation value of the recommendation scheme based on the training set of each recommendation scheme by using propensity score matching and average causal effect evaluation; and determining the user characteristics adapted to each first recommendation scheme by Pearson similarity calculation. In this way, user groups with the same recommendation scheme characteristics are screened out, which facilitates the subsequent provision of intervention variables for user satisfaction evaluation to form a training set. Furthermore, the bias and influence of confounding variables in the study are reduced by propensity score matching, so that the effective recommendation schemes in each recommendation scheme are evaluated by average causal effect. Finally, the user characteristics adapted to each recommendation scheme are determined by Pearson similarity calculation, providing analytical support for the recommendation content decision of the customer service system, so as to improve the user satisfaction with the customer service system service. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 A flowchart of a method for generating a recommendation solution for a customer service system is provided for an embodiment of the present application;

[0038] Figure 2 A schematic diagram of a training set generation process provided in an embodiment of the present application;

[0039] Figure 3 A schematic diagram of a satisfaction estimation process provided in an embodiment of the present application;

[0040] Figure 4 A schematic diagram of the structure of a recommendation solution generating device for a customer service system provided in an embodiment of the present application;

[0041] Figure 5 A schematic diagram of the structures of the training set generation unit, evaluation unit and analysis unit provided in the embodiments of the present application. DETAILED DESCRIPTION

[0042] In order to be able to understand the features and technical contents of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure is described in detail below in conjunction with the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present disclosure. In the following technical description, for the convenience of explanation, a full understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, to simplify the drawings, well-known structures and devices can be simplified for display.

[0043] The terms "first", "second", etc. in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the embodiments of the present disclosure described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0044] Unless otherwise stated, the term "plurality" means two or more.

[0045] In the embodiment of the present disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B indicates: A or B.

[0046] The term "and / or" is a description of the association relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B.

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] See also Figure 1 , Figure 1 A flowchart of a method for generating a recommendation solution for a customer service system provided in an embodiment of the present application includes:

[0049] S101 , clustering analysis is performed on users and recommended contents in user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme.

[0050] The user data is the user data collected by the customer service system, which may include user behavior records, interaction logs, user portraits and other information. User data can be represented by a data set, for example, D = {(E i , Fi , G i , Y i )}, where E i It is the content recommended by the user customer service system; F i It is the user's historical behavior data, such as the user's unique identification (such as ID number, phone number, etc.), age, shopping data or loan repayment record, etc.; G i is the user's current context information, such as the interaction log information with the client system; Y i is the user experience score. i and G i As user characteristics.

[0051] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0052] The recommendation scheme is a recommendation content combination scheme. Each time, the customer service system will recommend multiple recommended contents to the user, and the combination of multiple recommended contents forms a recommendation scheme.

[0053] Bicluster analysis is used to determine the user group characteristics corresponding to various recommendation scheme characteristics.

[0054] S102: Generate a training set of satisfaction estimates of the recommendation schemes based on the recommendation schemes and their corresponding user groups.

[0055] According to step S101, each recommendation scheme has a corresponding user group. It can be understood that through clustering, the user group corresponding to one type of recommendation scheme has high satisfaction, while the user groups corresponding to other types of recommendation schemes have low satisfaction. In a specific implementation step, step S102 sets an intervention variable for satisfaction estimation for each user group for each recommendation scheme (the intervention variable set for satisfaction is 1, and the intervention variable set for dissatisfaction is 0), forming a training set for the satisfaction estimation of the recommendation scheme. The setting of the intervention variable indicates whether the user accepts the recommendation scheme.

[0056] S103. Based on the training set of each recommended solution, the evaluation value of the recommended solution is determined by using propensity score matching and average causal effect evaluation.

[0057] Calculate the propensity score of each user (one user in the user group with intervention variable 1 and one user in the user group with intervention variable 0 in each training set) for the recommendation scheme corresponding to the training set, and then determine the user group that receives the recommendation scheme (experimental group) and the user group that does not receive the recommendation scheme (control group) based on the user's propensity score through a one-to-one nearest neighbor matching method. Finally, calculate the average causal effect based on the user data of the experimental group and the control group to evaluate the recommendation scheme. In this way, the distribution of user characteristics between user groups is balanced through propensity scoring and matching, thereby reducing the influence of bias and confounding variables.

[0058] Optionally, the first recommended solution is a recommended solution whose average causal effect value exceeds a preset value.

[0059] S104: Determine user characteristics that are suitable for each recommendation scheme through Pearson similarity calculation.

[0060] In this way, when the customer service system provides a recommendation scheme of recommended content combinations for new users, it can determine an adapted type of recommendation scheme based on user characteristics, and provide analytical support for the customer service system's recommended content decision-making, so as to improve user satisfaction with the customer service system's services.

[0061] Based on the above steps S101-S104, this application finds that the characteristics of the same type of recommended content combination represent a recommendation scheme through cluster analysis of users and recommended content, and one recommendation scheme corresponds to one user group. The user groups with the same recommendation scheme characteristics are screened out to facilitate the subsequent provision of intervention variables for user satisfaction evaluation to form a training set, and then, through propensity score matching, the bias and influence of confounding variables in the study are reduced, so that the effective recommendation schemes in each recommendation scheme are evaluated through the average causal effect, and finally, through Pearson similarity calculation, the user characteristics adapted to each recommendation scheme are determined, and analytical support is provided for the customer service system's recommendation content decision-making, so as to improve user satisfaction with the customer service system's services.

[0062] Based on the above embodiment, before step S101, the method further includes: preprocessing the user data; the preprocessing of the user data includes: data cleaning, missing value processing and feature extraction of the user data.

[0063] User data is structured through user data preprocessing and can be represented by a user matrix.

[0064] In the embodiment of the present application, the above Figure 1 The possible implementations of step S101 are described in detail below. It should be noted that the implementations given in the following description are only for illustrative purposes and do not represent all implementations of the embodiments of the present application.

[0065] In the above step S101, cluster analysis is performed on users and recommended contents in the user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme. The specific implementation steps may be:

[0066] Step A1: Create a bipartite graph of user recommendation content features based on user features and recommended content features.

[0067] Specifically, a user recommendation content feature bipartite graph G = (P, S, E) is established based on the user features and recommended content features in the preprocessed user data;

[0068] Where P = {p1, ..., p m}, represents the user node set; S = {s1,...,s n}, indicating the recommended content node set; E = {{p i ,s j}:p i ∈P,s j ∈S,}, represents the edge set of G.

[0069] Step A2: Establish an adjacency matrix M of the bipartite graph of user recommendation content features.

[0070]

[0071] Among them, the matrix A has m rows and n columns. If the i-th row and j-th column of A ij =1 indicates user P i Received recommended content S j , if A in row i and column j ij =0 means user P i No recommended content received j .

[0072] Step A3: Solve the generalized eigenvector of the adjacency matrix to obtain a characteristic matrix.

[0073] In a specific implementation, a Laplace matrix can be constructed based on an adjacency matrix, and the generalized eigenvector can be solved by the Laplace matrix. The formula for constructing a Laplace matrix based on an adjacency matrix is:

[0074]

[0075] Where D1 and D2 are diagonal matrices, D1(i,i)=∑ j A ij , D2(i,i)=∑ j A ji .

[0076] Step A4: clustering the feature matrix using a K-means algorithm to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

[0077] In the above steps A1-A4, through cluster analysis of a large number of users and recommended contents, user groups with similar recommended content combinations are found, and the correspondence between the user group division and the recommended content combination division results is achieved, and the characteristics of the divided user groups and the characteristics of the recommended solutions (recommended content combination division results) are achieved. This is convenient for providing cause variables, i.e. intervention, for subsequent user satisfaction evaluation. Furthermore, the characteristics of different user groups and recommended solutions are better represented, and user groups with recommended solution characteristics are screened out.

[0078] In the embodiment of the present application, the above Figure 1 The possible implementations of step S102 are described in detail below. It should be noted that the implementations described below are only for illustrative purposes and do not represent all implementations of the embodiments of the present application.

[0079] See also Figure 2 The schematic diagram of a training set generation process shown in the figure, in the above step S102, based on each recommendation scheme and its corresponding user group, generating a training set of the recommendation scheme satisfaction estimation, includes:

[0080] S201: Obtain a recommendation plan and its corresponding user group.

[0081] S202: Select a recommendation scheme for which no training set has been generated.

[0082] S203. Set the intervention variable representing the satisfaction estimate configured in the first user data to 1 (W=1), and set the intervention variable configured in the second user data to 0 (W=0). The first user data and the second user data are used as training sets of the recommendation scheme. The first user data is user data included in the user group corresponding to the recommendation scheme, and the second user data is user data included in the user group not corresponding to the recommendation scheme.

[0083] S204, determining whether there are any recommended solutions for which no training set has been generated in the recommended solutions obtained in step S201, if yes, continuing to execute step S202, if no, ending.

[0084] Through the above steps S201 - S204 , based on each recommendation scheme, corresponding intervention variables are configured for the user data of each user in all user groups to form a training set of the recommendation scheme.

[0085] In the embodiment of the present application, the above Figure 1The possible implementations of step S103 are described in detail below. It should be noted that the implementations described below are only for illustrative purposes and do not represent all implementations of the embodiments of the present application.

[0086] See also Figure 3 A schematic diagram of a satisfaction estimation process is shown, in step S103, based on the training set of each recommended solution, using propensity score matching and average causal effect evaluation to determine the evaluation value of the recommended solution, including:

[0087] S301: Obtain a training set of unevaluated recommendation solutions.

[0088] S302: Based on the user data of each user, a logistic regression calculation is performed with the user characteristics in the user data as the independent variable and the acceptance of the recommendation scheme as the dependent variable to obtain a propensity score of the user corresponding to the user data to accept the recommendation scheme.

[0089] S303. Based on the propensity score, a one-to-one nearest neighbor matching method is used to match users in the training set who accept the recommended solution and users who do not accept the recommended solution, and the users who accept the recommended solution are formed into a recommended solution group, and the users who do not accept the recommended solution are formed into a non-recommended solution group.

[0090] Nearest Neighbor Matching is a propensity score matching method. Propensity score is the probability of a user accepting a recommendation. In a one-to-one matching, for each user who accepts the recommendation, a user with the closest propensity score is found among the users who did not accept the recommendation for matching. This forms two relatively "similar" groups of users, one group of users who accept the recommendation and the other group of users who do not accept the recommendation but are similar to them in other user characteristics. In this way, the distribution of user characteristics between user groups is balanced through propensity scores and matching, thereby reducing the impact of bias and confounding variables.

[0091] S304, calculating the average causal effect evaluation value between user acceptance of the recommendation plan and user satisfaction.

[0092] S305. The recommended solution whose average causal effect estimate is higher than the preset value is a valid recommended solution.

[0093] S306: Determine whether there is any training set of recommended solutions that have not been evaluated. If so, continue to execute step S301; if not, end.

[0094] The specific formula for calculating the average causal effect value between user acceptance of the recommendation plan and user satisfaction can be:

[0095] Among them, N w＝1 is the number of users who accept the recommended solution, N w＝0 is the number of users who did not accept the recommended solution, Y w＝1 is the actual satisfaction result of the user who accepts the recommended solution W=1 after matching (the sum of the user experience scores in the user data of the user who receives the recommended solution); Y w＝1 It is the real satisfaction result of users who did not accept the recommendation solution W=0 after matching (the sum of the user experience scores in the user data of users who did not receive the recommendation solution).

[0096] Specifically, the average causal effect estimate corresponding to each recommended plan is calculated, and the recommended plan with an average causal effect estimate higher than the preset value is a valid recommended plan (the first recommended plan). The plan with higher customer satisfaction can be used as a valid recommended plan for subsequent customer service systems to push to adapted users.

[0097] Then, the user characteristics of the recommended solutions adapted by Pearson similarity calculation evaluation values ​​exceeding the preset values ​​may be specifically performed as follows:

[0098] Calculate the Pearson similarity value of the user feature of each user in the user data between the recommendation solution group and the non-recommendation solution group;

[0099] User features with similarity values ​​less than a threshold are selected as user features adapted by the recommendation scheme.

[0100] The above are some specific implementations of a method for generating a recommendation solution for a customer service system provided by the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided by the embodiment of the present application will be introduced from the perspective of functional modularization.

[0101] See also Figure 4 The structure diagram of a recommendation solution generating device of a customer service system is shown, and the recommendation solution generating device of the customer service system comprises:

[0102] A division unit 401 is used to perform cluster analysis on users and recommended contents in the user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme;

[0103] A training set generating unit 402 is used to generate a training set of satisfaction estimation of the recommendation schemes based on each recommendation scheme and its corresponding user group;

[0104] An evaluation unit 403 is used to determine the evaluation value of the recommended solution based on the training set of each recommended solution by using propensity score matching and average causal effect evaluation;

[0105] The analyzing unit 404 is used to calculate the user characteristics of the recommendation schemes adapted to the user whose evaluation value exceeds a preset value through Pearson similarity calculation.

[0106] Based on the above device, the division unit 401 screens out user groups with the same recommendation scheme characteristics, so that the subsequent training set generation unit 402 provides intervention variables for user satisfaction evaluation to form a training set. Furthermore, the evaluation unit 403 reduces the bias and influence of confounding variables in the study through propensity score matching, thereby evaluating the effective recommendation schemes among the recommendation schemes through the average causal effect. Finally, the analysis unit 404 determines the user characteristics adapted to each recommendation scheme through Pearson similarity calculation, and provides analytical support for the customer service system's recommendation content decision-making, so as to improve user satisfaction with the customer service system's services.

[0107] In a possible implementation manner, the above-mentioned device may further include a preprocessing unit, which is used to preprocess the user data.

[0108] The user data is preprocessed, including: data cleaning, missing value processing and feature extraction of the user data.

[0109] Optionally, the above-mentioned division unit is specifically used to establish a bipartite graph of user recommendation content features based on user characteristics and recommended content characteristics; establish an adjacency matrix of the user recommendation content feature bipartite graph; solve the generalized eigenvector of the adjacency matrix to obtain a feature matrix; use the K-means algorithm to cluster the feature matrix to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

[0110] In a specific implementation, see Figure 5 The schematic diagram of the structure of the training set generation unit, evaluation unit and analysis unit is shown.

[0111] The training set generating unit 402 is specifically used to obtain a recommendation scheme for which a training set has not been generated; an intervention variable indicating a satisfaction estimate configured in the first user data is set to 1, and an intervention variable configured in the second user data is set to 0, the first user data and the second user data are used as training sets for the recommendation scheme, the first user data are user data included in a user group corresponding to the recommendation scheme, and the second user data are user data included in a user group not corresponding to the recommendation scheme; the method of obtaining a recommendation scheme for which a training set has not been generated is continued until there is no recommendation scheme for which a training set has not been generated.

[0112] The evaluation unit 403 is specifically used to perform a logistic regression calculation based on each user data, with the user characteristics as the independent variable and whether to accept the recommended solution as the dependent variable, to obtain the propensity score of the user corresponding to the user data to accept the recommended solution; based on the propensity score, a one-to-one nearest neighbor matching method is used to match the users who accept the recommended solution and the users who do not accept the recommended solution in the training set, and the users who accept the recommended solution form a recommended solution group, and the users who do not accept the recommended solution are regarded as the group that does not accept the recommended solution; the average causal effect evaluation value between the user's acceptance of the recommended solution and the user's satisfaction is calculated as the evaluation value of the recommended solution. Among them, the recommended solution with an evaluation value exceeding the preset value is a valid recommended solution (the first recommended solution).

[0113] The analyzing unit 404 is specifically configured to calculate a similarity value of each user feature between the recommendation solution group and the unaccepted recommendation solution group; and select user features with similarity values ​​less than a threshold as user features adapted to the recommendation solution.

[0114] For details, see Figure 5 , input each user group, the recommendation scheme corresponding to the user group, and the user characteristics in the user data into the training set generation unit 402, obtain the training set generated by the training set generation unit 402 corresponding to each recommendation scheme, input each training set into the evaluation unit 403, and obtain an effective evaluation scheme. Then, the effective recommendation scheme is determined by the analysis unit 404 to determine the user characteristics adapted to the recommendation scheme. In this way, the user characteristics adapted to each recommendation scheme are determined, and analytical support is provided for the recommendation content decision of the customer service system, so as to improve the user satisfaction with the customer service system service.

[0115] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0116] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes a method for generating a recommendation solution for a customer service system described in any embodiment of the present application.

[0117] The computer storage medium stores codes, and when the codes are executed, the device executing the codes implements a method for generating a recommendation solution for a customer service system as described in any embodiment of the present application.

[0118] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0119] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0120] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0121] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A method for generating a recommendation plan for a customer service system, characterized in that: include: Perform cluster analysis on users and recommended contents in user data to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme; Based on each recommendation scheme and its corresponding user group, generating a training set of satisfaction estimation of the recommendation scheme; Based on the training set of each recommended solution, the evaluation value of the recommended solution is determined by using propensity score matching and average causal effect evaluation; The user characteristics that are adapted to the recommendation scheme when the evaluation value exceeds the preset value through Pearson similarity calculation.

2. The method according to claim 1, characterized in that The cluster analysis of users and recommended contents in the user data is performed to obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme, including: Based on the characteristics of the user and the characteristics of the recommended content, a bipartite graph of user recommended content characteristics is established; Establish the adjacency matrix of the bipartite graph of user recommendation content features; Solving the generalized eigenvector of the adjacency matrix to obtain a characteristic matrix; The feature matrix is ​​clustered using a K-means algorithm to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

3. The method according to claim 1, characterized in that: The step of generating a training set of satisfaction estimation of the recommendation schemes based on the recommendation schemes and their corresponding user groups includes: Get a recommendation solution without generating a training set; The intervention variable indicating the satisfaction estimate configured in the first user data is set to 1, and the intervention variable configured in the second user data is set to 0, the first user data and the second user data are used as training sets of the recommendation scheme, the first user data is user data included in the user group corresponding to the recommendation scheme, and the second user data is user data included in the user group not corresponding to the recommendation scheme; the step of obtaining a recommendation scheme for which a training set has not been generated is continued until there is no recommendation scheme for which a training set has not been generated.

4. The method according to claim 1, characterized in that The training set based on each recommended solution, using propensity score matching and average causal effect evaluation to determine the evaluation value of the recommended solution, includes: Based on each user data, a logistic regression calculation is performed with the user characteristics as the independent variable and whether the recommended solution is accepted as the dependent variable to obtain the propensity score of the user corresponding to the user data to accept the recommended solution; Based on the propensity score, a one-to-one nearest neighbor matching method is used to match users in the training set who accept the recommended solution and users who do not accept the recommended solution, and the users who accept the recommended solution are formed into a recommended solution group, and the users who do not accept the recommended solution are formed into a non-recommended solution group; The average causal effect evaluation value between user acceptance of the recommended solution and user satisfaction is calculated as the evaluation value of the recommended solution.

5. The method according to claim 4, characterized in that The user characteristics adapted by the recommendation scheme whose evaluation value calculated by Pearson similarity exceeds the preset value include: Calculating a similarity value of each user feature between the recommendation solution group and the unaccepted recommendation solution group; User features with similarity values ​​less than a threshold are selected as user features adapted by the recommendation scheme.

6. The method according to claim 1, characterized in that Before clustering the users and recommended contents in the user data, the method further includes preprocessing the user data; The preprocessing of the user data includes: data cleaning, missing value processing and feature extraction of the user data.

7. A recommendation solution generating device for a customer service system, characterized in that: The device comprises: A division unit, used to perform cluster analysis on users and recommended contents in the user data, and obtain multiple recommendation schemes and user groups corresponding to each recommendation scheme; A training set generating unit, used for generating a training set of satisfaction estimation of the recommendation schemes based on each recommendation scheme and its corresponding user group; An evaluation unit, used to determine the evaluation value of the recommended scheme based on the training set of each recommended scheme by using propensity score matching and average causal effect evaluation; The analyzing unit is used to calculate the user characteristics of the recommendation scheme adapted by the evaluation value exceeding the preset value through Pearson similarity calculation.

8. The device according to claim 7, characterized in that The partitioning unit is specifically used to establish a user recommendation content feature bipartite graph based on the user's features and the features of the recommended content; establish an adjacency matrix of the user recommendation content feature bipartite graph; and solve the generalized eigenvector of the adjacency matrix to obtain a feature matrix; The feature matrix is ​​clustered using a K-means algorithm to obtain recommendation schemes for recommended content combinations and user groups corresponding to each recommendation scheme.

9. The device according to claim 7, characterized in that The evaluation unit is specifically used to perform a logistic regression calculation based on each user data, taking the user characteristics as the independent variable and whether the recommended solution is accepted as the dependent variable, to obtain a tendency score of the user corresponding to the user data to accept the recommended solution; Based on the propensity score, a one-to-one nearest neighbor matching method is used to match users in the training set who accept the recommended solution and users who do not accept the recommended solution, and the users who accept the recommended solution are formed into a recommended solution group, and the users who do not accept the recommended solution are formed into a non-recommended solution group; The average causal effect evaluation value between user acceptance of the recommended solution and user satisfaction is calculated as the evaluation value of the recommended solution.

10. An electronic device, characterized in that: The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes, so that the device executes the method for generating a recommendation solution for a customer service system as described in any one of claims 1-6.

Images