Problem solution recommendation method, system, electronic device, and storage medium

Abstract

The application provides a problem solution recommendation method and system, an electronic device and a storage medium, relates to the technical field of computers, solves the problem that the expression ability of a logic regression model is weak when the best solution of a problem customer is automatically recommended, and the method comprises the following steps: taking target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer as input to perform solution theme prediction, obtaining a plurality of to-be-recommended solution themes, and ranking the plurality of to-be-recommended solution themes according to an improvement probability; and taking a to-be-recommended solution theme ranked at a target position as a target solution theme to be recommended to a target recommendation terminal. Through the above recommendation method, the application can automatically recommend a solution with a high improvement rate according to a newly mined problem customer, effectively improves the problem solution efficiency of the problem customer, and achieves quick and comprehensive solution of the problem.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, in particular to a problem solution recommendation method and system, electronic equipment and storage medium. BACKGROUND

[0002] Large groups of enterprises generally mine the problem customers existing in the current operation process through many data mining models based on business rules, and then feed back the problem customers to experts skilled in operational decision-making. The experts propose corresponding solutions based on the problem customers and their corresponding data mining model ideas, and then the execution layer lands the solutions to form a task closed loop and achieve problem improvement. Finally, the corresponding output, i.e. the cost reduction or income increase on the financial level, is calculated for the completion of the problem improvement, so as to evaluate the effectiveness and value of the solutions provided by the experts.

[0003] However, the best solutions for each problem customer may not be consistent for each expert, and the time and energy of the experts are extremely valuable. If the best solutions for the problem customers can be automatically recommended, it can help the experts to make better judgments efficiently, which has great significance for integrating expert resources and expert knowledge and achieving accurate and efficient decision-making.

[0004] In order to achieve the above-mentioned needs, most platforms use the most classic collaborative filtering algorithm to achieve the recommendation function, i.e. collaborative history feedback, evaluation and opinion are used to filter massive information, and the target problem user may need the solutions are selected based on user similarity or solution similarity. However, the collaborative filtering model only uses the interaction of user and solution information for recommendation, and does not comprehensively use user, solution information, context and other different features, so it is difficult to form a more comprehensive recommendation result.

[0005] In addition, a logistic regression model (LR) can be used to sort information based on the predicted probability of positive samples, and recommend the top N samples in the ranking. The positive samples here are the solutions that are fed back by the experts and implemented, but LR only does simple weighting on a single feature and does not have the ability to generate high-dimensional combined features by crossing features, so the expression ability is weak. SUMMARY

[0006] The present application provides a problem solution recommendation method, system, electronic equipment and storage medium, which has the ability to generate high-dimensional combined features by crossing different features, can predict the target solution theme and improvement probability corresponding to the target problem customer based on the target customer attribute features and target cross multi-dimensional features of the target problem customer, and can more accurately recommend the target solution theme suitable for solving the problems existing in the target problem customer to the target recommendation terminal according to the improvement probability, thereby effectively improving the problem solving efficiency.

[0007] In an aspect, the present application provides a method for recommending a solution to a problem, comprising:

[0008] obtaining target customer attribute features and target cross-multiple dimension features corresponding to a target problem customer;

[0009] performing solution theme prediction according to the target customer attribute features and the target cross-multiple dimension features, to obtain a solution theme prediction result corresponding to the target customer information, the solution theme prediction result comprising a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively;

[0010] sorting the plurality of to-be-recommended solution themes according to the sizes of the plurality of solution theme improvement probabilities, to obtain a sorted solution theme set;

[0011] taking the to-be-recommended solution theme arranged at a target position in the sorted solution theme set as a target solution theme corresponding to the target problem customer, and recommending the target solution theme to a target recommendation terminal.

[0012] In a possible implementation of the present application, after the target customer attribute features and the target cross-multiple dimension features are determined according to the target customer information, the method comprises:

[0013] performing encoding processing on the target customer attribute features and the target cross-multiple dimension features respectively, to obtain a customer attribute feature vector matrix and a cross-multiple dimension feature vector matrix;

[0014] performing vector splicing on the customer attribute feature vector matrix and the cross-multiple dimension feature vector matrix, to obtain a target customer feature vector set.

[0015] In a possible implementation of the present application, the solution theme prediction according to the target customer attribute features and the target cross-multiple dimension features, to obtain a solution theme prediction result corresponding to the target customer information, comprises:

[0016] inputting the target customer feature vector set into a preset solution theme recommendation model to perform solution theme prediction, to obtain a solution theme prediction result corresponding to the target customer information.

[0017] In a possible implementation of the present application, before the target customer attribute features and the target cross-multiple dimension features corresponding to a target problem customer are obtained, the method comprises:

[0018] obtaining sample problem customer attribute information corresponding to a sample problem customer and sample solution information corresponding to the sample problem customer attribute information;

[0019] determining a sample solution topic according to the sample solution information;

[0020] determining sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution topic;

[0021] training a model by taking the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features as model inputs to obtain the solution topic recommendation model.

[0022] In a possible implementation of the present application, the determining of the sample solution topic according to the sample solution information comprises:

[0023] determining a related topic probability that the sample solution information belongs to a preset solution topic according to the sample solution information;

[0024] comparing the related topic probability with a preset topic probability threshold to obtain a topic probability comparison result;

[0025] judging whether the sample solution information belongs to the preset solution topic according to the topic probability comparison result;

[0026] when the sample solution information belongs to the preset solution topic, taking the preset solution topic as the sample solution topic corresponding to the sample solution information.

[0027] In a possible implementation of the present application, after the determining of the sample solution topic according to the sample solution information, the method comprises:

[0028] obtaining execution probability information and improvement probability information corresponding to the sample solution topic;

[0029] taking the sample solution topic, the execution probability information and the improvement probability information together as the sample solution related features.

[0030] In a possible implementation of the present application, after the determining of the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution topic, the method comprises:

[0031] The sample problem customer attribute feature, the sample solution related feature and the sample cross multi-dimensional feature are all encoded to obtain a sample customer attribute feature encoding matrix, a sample solution related feature encoding matrix and a sample other cross multi-dimensional feature encoding matrix respectively.

[0032] The sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix are spliced to obtain a spliced sample feature vector matrix, and the sample feature vector matrix includes a plurality of sample feature vectors.

[0033] In a possible implementation of the application, after the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix are spliced to obtain a spliced sample feature vector matrix, the method includes:

[0034] The spliced sample feature vector matrix is converted into a sample feature vector matrix in a zero-out storage format through a preset zero-out format, and the zero-out format is:

[0035] y index1: value1 index2: value2...index n : value n

[0036] y represents an improvement probability label in the sample feature vector matrix, index i represents a feature column index of a non-zero value in the sample feature vector matrix, value i represents a corresponding feature value in the sample feature vector matrix, index i i∈n.

[0037] In a possible implementation of the application, the sample problem customer attribute feature, the sample solution related feature and the sample cross multi-dimensional feature are used as model inputs to train a model to obtain the solution topic recommendation model, and the method includes:

[0038] A nonlinear prediction model is established.

[0039] All sample features in the sample feature vector matrix in the zero-out storage format are divided into positive sample features and negative sample features, the positive sample features include a sample feature vector corresponding to a sample problem customer whose problem is improved according to sample solution information, and the negative sample features include a sample feature vector corresponding to a sample problem customer whose problem is not improved according to sample solution information.

[0040] The positive sample features and the negative sample features are input into the nonlinear prediction model for model training, to obtain the solution theme recommendation model.

[0041] In another aspect, the present application provides a system, which comprises:

[0042] An acquisition module is configured to acquire target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer;

[0043] A prediction module is configured to perform solution theme prediction according to the target customer attribute features and the target cross multi-dimensional features, to obtain a solution theme prediction result corresponding to the target customer information, wherein the solution theme prediction result comprises a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively;

[0044] An ordering module is configured to order the plurality of to-be-recommended solution themes according to the sizes of the plurality of solution theme improvement probabilities, to obtain an ordered solution theme set;

[0045] A recommendation module is configured to recommend, as a target solution theme corresponding to the target problem customer, a to-be-recommended solution theme arranged at a target position in the ordered solution theme set, and to recommend the target solution theme to a target recommendation terminal.

[0046] In another aspect, the present application further provides an electronic device, which comprises:

[0047] One or more processors;

[0048] A memory; and

[0049] One or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to implement the problem solution recommendation method.

[0050] In another aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is loaded by a processor to execute steps in the problem solution recommendation method.

[0051] In the present application, by acquiring target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer, solution theme prediction is performed according to the target customer attribute features and the target cross multi-dimensional features, a plurality of to-be-recommended solution theme corresponding to the target customer information is obtained, a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes are obtained, the plurality of to-be-recommended solution themes are sorted according to the solution theme improvement probabilities, the to-be-recommended solution themes ranked in a specified position are taken as target solution themes corresponding to the target problem customer, and are recommended to a target recommendation terminal. Therefore, the present application has the ability to process target cross multi-dimensional features, can quickly and accurately generate a problem solution theme suitable for a target problem customer in combination with target customer attribute features and target cross multi-dimensional features, and effectively improves the problem solving efficiency of the target problem customer. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Problem solution recommendation method, system, electronic device and storage medium

[0053] Figure 1 is an embodiment flow diagram of a problem solution recommendation method provided in the embodiments of the present application;

[0054] Figure 2 is an embodiment structure diagram of a target customer feature vector set provided in the embodiments of the present application;

[0055] Figure 3 is an embodiment flow diagram of a problem solution recommendation method provided in the embodiments of the present application;

[0056] Figure 4 is an embodiment structure diagram of a sample feature vector matrix provided in the embodiments of the present application;

[0057] Figure 5 is an embodiment structure diagram of a factor decomposition dimension and hyperparameter relationship provided in the embodiments of the present application;

[0058] Figure 6 is an embodiment structure diagram of a pre-set solution theme and hyperparameter relationship provided in the embodiments of the present application;

[0059] Figure 7 is an embodiment structure diagram of a problem solution recommendation system provided in the embodiments of the present application;

[0060] Figure 8 FIG. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, any other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the present application.

[0062] In the description of the present application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0063] In the present application, the word "exemplary" is used to mean "serving as an example, instance, or illustration." Any implementation described as "exemplary" in the present application is not necessarily to be construed as preferred or advantageous over other implementations. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present application. It will be apparent, however, to one skilled in the art that the present application can be practiced without using these specific details. In other instances, well-known structures and processes are not elaborated in order not to obscure the description of the present application with unnecessary details. Thus, the present application is not intended to be limited by the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0064] The present application provides a problem solution recommendation method, system, electronic device and storage medium, which are described in detail below.

[0065] In the present application, for the convenience of understanding, the subject performing the problem solution recommendation method in the following text is referred to as the enterprise, but the enterprise does not specifically refer to any specific enterprise at present, and the enterprise is not more specifically limited.

[0066] As shown in FIG. 1, it is an embodiment flow diagram of the problem solution recommendation method in the embodiments of the present application, and the problem solution recommendation method includes steps 101-104: Figure 1

[0067] ​101、obtaining target customer attribute features and target cross multi-dimensional features corresponding to the target problem customer.

[0068] The target problem customer refers to a customer in all customer groups of the enterprise who has abnormal cooperation behavior.

[0069] That is, before obtaining the target customer attribute features and the target cross multi-dimensional features corresponding to the target problem customer, the method further comprises: discovering a customer having abnormal cooperation behavior in the current operation process by a business rule-based data mining model, and taking the customer having abnormal cooperation behavior as the target problem customer.

[0070] Specifically, the business rule-based data mining model can include a data model based on quotation estimation and a data model based on monitoring customer dynamics, etc. The business rule-based data mining model is trained based on the existing conventional business and customer group data of the enterprise, and different data mining models can be trained according to different target problem customers as needed. Here, the business rule-based data mining model is not specifically limited.

[0071] In the operation process of the enterprise, the data model based on quotation estimation can be used to monitor customers with unreasonable quotation estimation, or the data model based on monitoring customer dynamics can be used to monitor customers who have not cooperated with the enterprise for a set N days, etc. These "customers with unreasonable quotation estimation" and "customers who have not cooperated with the enterprise for a set N days" can all be target problem customers.

[0072] After the target problem customer is monitored, the target customer attribute features and the target cross multi-dimensional features corresponding to the target problem customer are extracted according to the attribute information of the target problem customer and in combination with the user portrait data of the customers accumulated by the enterprise. The target customer attribute features represent the basic attribute information of the target problem customer, and the target cross multi-dimensional features represent multi-dimensional features formed by the combination of two or more features of the target problem customer.

[0073] Exemplarily, the target customer attribute features can include customer name (ID), customer stratification, customer classification, cooperation agreement signing time, and other customer basic attribute features, and the target cross multi-dimensional features can include the hit of a certain business rule-based data model on the target problem customer, the importance of the business rule-based data model, or the historical improvement rate of the target problem customer signed for N years, and the like. For example, the target customer attribute features of a certain target problem customer include: the company name is A Company, the customer stratification is core customer, the customer classification is one-level customer, and the cooperation agreement signing time is 3 years; the target cross multi-dimensional features of a certain target customer include: hitting the data model based on monitoring customer dynamics, the importance of the pre-set data model based on monitoring customer dynamics is high, and the historical improvement rate of the target problem customer signed for 3 years is 60%, and the like.

[0074] In the embodiment, after determining the target customer attribute features and the target cross multi-dimensional features according to the target customer information, the method includes:

[0075] The target customer attribute features and the target cross multi-dimensional features are respectively encoded to obtain a customer attribute feature vector matrix and a cross multi-dimensional feature vector matrix; and the customer attribute feature vector matrix and the cross multi-dimensional feature vector matrix are spliced to obtain a target customer feature vector set.

[0076] After obtaining the target customer attribute features and the target cross multi-dimensional features corresponding to the target problem customer, in order to facilitate the calculation of each feature of the target problem customer as input, the target customer attribute features and the target cross multi-dimensional features need to be normalized and processed. In the embodiment, the customer attribute feature vector matrix and the cross multi-dimensional feature vector matrix are spliced by one-hot encoding the target customer attribute features and the target cross multi-dimensional features respectively to obtain the corresponding target customer feature vector set, and the subsequent calculation is performed by taking the target customer feature vector set as the target input to obtain the corresponding result.

[0077] Specifically, the target customer feature vector set can be as shown in the feature vector X as shown in Figure 2 . Wherein, x (1) ~ x (7) represent 7 different target problem customers. Exemplarily, the first row of x (1) is taken as an example, in the horizontal direction of the figure, for the customer ID feature part, x (1) is 1 in the column value of "whether it is an A customer", and the other columns are 0, for the customer level part, "whether it is a two-level customer" is 1 in the column value, and the other columns are 0, for the hit of the rule-based data mining model part, "whether it hits model 1" is 1 in the column value, and the other columns are 0, that is, the target problem customer x (1)The features include A customer, a two-level customer, being hit by Model 1, etc.

[0078] 102. Based on the target customer attribute features and the target cross multi-dimensional features, solution theme prediction is performed to obtain a solution theme prediction result corresponding to the target customer information, the solution theme prediction result including a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively.

[0079] In this embodiment, after the target customer attribute features and the target cross multi-dimensional features are processed to obtain the target customer feature vector set, the target customer feature vector set is directly input for solution theme prediction. Therefore, based on the target customer attribute features and the target cross multi-dimensional features, solution theme prediction is performed to obtain a solution theme prediction result corresponding to the target customer information, which specifically includes:

[0080] The target customer feature vector set is input into a preset solution theme recommendation model for solution theme prediction to obtain a solution theme prediction result corresponding to the target customer information.

[0081] In this embodiment, the preset solution theme recommendation model is trained based on a factorization machines (FM) model. Since the FM model can overcome the defects of the LR, can process large-scale sparse data, and can automatically learn the information brought by feature cross, after the solution theme recommendation model is trained based on the FM model, the target customer feature vector set is input into the preset solution theme recommendation model. The solution theme recommendation model automatically performs calculation according to the target customer feature vector set, and can also calculate the target cross multi-dimensional features in the target customer feature vector set, thereby generating a plurality of to-be-recommended solution themes suitable for the target problem customer. At the same time, the preset solution theme recommendation model has a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes, which facilitates more accurate and targeted recommendation of solution themes suitable for the target problem customer.

[0082] In the embodiment, in the solution theme recommendation model training process, a plurality of different to-be-recommended solution themes are extracted in advance according to a plurality of different types of solution texts provided, and the to-be-recommended solution themes are related to the pre-provided solution texts. For example, the to-be-recommended solution themes can include adjusting price, switching product, reducing claim, line optimization, and the like. Based on literal understanding, when the to-be-recommended solution theme output by the preset solution theme recommendation model is “adjusting price”, the to-be-recommended solution theme indicates that the solution generated by the solution theme recommendation model is suitable for adjusting the product price of the enterprise cooperating with the target problem customer, and the meanings of other to-be-recommended solution themes are similar, which will not be described herein.

[0083] 103. The plurality of to-be-recommended solution themes are sorted according to the sizes of the plurality of solution theme improvement probabilities, to obtain a sorted solution theme set.

[0084] 104. The to-be-recommended solution theme arranged at the target position in the solution theme set is taken as a target solution theme corresponding to the target problem customer, and the target solution theme is recommended to a target recommendation terminal.

[0085] In the embodiment, the plurality of solution theme improvement probabilities are arranged in descending order, and the to-be-recommended solution themes arranged at the top N positions in the solution theme set are taken as the to-be-recommended solution themes arranged at the target position in the solution theme set, where N is a natural number.

[0086] That is, after obtaining the solution theme prediction result, the to-be-recommended solution themes arranged at the top N positions in the sorted solution theme set are recommended to the target recommendation terminal as the target solution theme. For example, the to-be-recommended solution themes arranged at the top 3 positions in the solution theme set are recommended to the target recommendation terminal as the target solution theme.

[0087] In the embodiment, the target recommendation terminal can be an expert user terminal for auditing whether the target solution theme is actually suitable for the target problem customer. After the expert user terminal receives the target solution theme, it is judged whether the target solution theme can help solve the problem of the target problem customer. If the target solution theme can help solve the problem of the target problem customer, the target solution theme can be implemented to form a closed loop to improve the efficiency of solving the problem of the target problem customer. The target recommendation terminal can also be a sales terminal of the enterprise. The target recommendation terminal can be set according to the needs, and the embodiment is not limited in this regard.

[0088] In the present application, by acquiring target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer, solution theme prediction is performed according to the target customer attribute features and the target cross multi-dimensional features, a plurality of to-be-recommended solution theme corresponding to the target customer information is obtained, and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes are obtained. The plurality of to-be-recommended solution themes are sorted according to the solution theme improvement probabilities, the to-be-recommended solution themes ranked in a specified position are taken as target solution themes corresponding to the target problem customer, and are recommended to a target recommendation terminal. Therefore, the present application has the ability to process target cross multi-dimensional features, can quickly and accurately generate a problem solution theme suitable for a target problem customer in combination with target customer attribute features and target cross multi-dimensional features, and effectively improves the problem solving efficiency of the target problem customer.

[0089] According to the above content, the input target customer feature vector set is calculated by the preset solution theme recommendation model, the solution theme prediction result is output, and the appropriate target solution theme is recommended to the target recommendation terminal according to the to-be-recommended solution theme in the solution theme prediction result and the corresponding solution theme improvement probability.

[0090] Therefore, in the present application, before the recommendation work is performed, a plurality of different solution themes and a large amount of sample data are first collected as model input data, the solution theme prediction result is taken as output data, the model is trained based on a specific model, and the solution theme recommendation model proposed in the present application is obtained.

[0091] Therefore, in another embodiment of the present application, before acquiring the target customer attribute features and the target cross multi-dimensional features corresponding to the target problem customer, as shown in Figure 3 The method includes steps 201-204.

[0092] 201. Acquire sample problem customer attribute information corresponding to a sample problem customer and sample solution information corresponding to the sample problem customer attribute information.

[0093] The sample problem customer refers to a sample customer in all customer groups of the enterprise who has abnormal cooperation behavior and has proposed a solution to solve the abnormality.

[0094] That is, before acquiring the sample problem customer attribute information corresponding to the sample problem customer and the sample solution information corresponding to the sample problem customer attribute information, the method further includes:

[0095] The customer with the abnormal cooperation behavior is taken as an initial problem customer through the business rule-based data mining model to discover the customer with the abnormal cooperation behavior; initial solution information for solving the problem of the initial problem customer is obtained according to the initial problem customer, the initial solution information is taken as sample solution information, and the initial problem customer who has solved the problem by using the initial solution information is taken as a sample problem customer.

[0096] In the embodiment, the sample problem customer attribute information includes basic attribute information corresponding to the sample problem customer, such as enterprise portrait data of an enterprise, and related information of a business signed with the enterprise, the sample solution information is solution information proposed when solving the problem of the sample problem customer, such as a control opinion (for example, termination of cooperation, re-signing of a contract, etc.) for an external sample problem customer or an improvement strategy (for example, adjustment of a price scheme, change of an operation mode, etc.) for an internal enterprise of the enterprise.

[0097] The way of obtaining the sample problem customer attribute information of the sample problem customer by using the business rule-based data mining model in the embodiment can be as described in step 101, and details are not described herein again. The sample solution information is solution information proposed by a staff of the enterprise for solving the problem of the sample problem customer after monitoring the sample problem customer. The sample solution information usually exists in the form of text, and can also be in other forms, which are not limited herein.

[0098] 202. Determine a sample solution theme according to the sample solution information.

[0099] Since the sample solution information usually exists in the form of text, different sample solution information in different texts can be different in the number of words, the way of expression, or other differences. Some different sample solution information has differences in texts, but the solutions for solving the problem of the sample problem customer are similar. Therefore, the multiple different sample solution information needs to be normalized, that is, in the embodiment, the sample solution theme is determined according to different sample solution information.

[0100] Therefore, in the embodiment, the sample solution theme is determined according to the sample solution information, including:

[0101] According to the sample solution information, a relevant topic probability that the sample solution information belongs to a preset solution topic is determined; the relevant topic probability is compared with a preset topic probability threshold to obtain a topic probability comparison result; according to the topic probability comparison result, it is judged whether the sample solution information belongs to the preset solution topic; and when the sample solution information belongs to the preset solution topic, the preset solution topic is taken as a sample solution topic corresponding to the sample solution information.

[0102] In the embodiment, a Latent Dirichlet Allocation (LDA) can be used to extract a preset solution topic corresponding to the sample solution information according to a potential semantic of the sample solution information.

[0103] Specifically, the multiple sample solution information is modeled by LDA, and a Gibbs sampling method is used to solve model parameters to obtain a topic distribution θ of the sample solution information,

[0104]

[0105] wherein θ ατ represents a probability that the sample solution information α belongs to the preset solution topic τ, that is,

[0106]

[0107] In the embodiment, assuming that there are T preset solution topics τ, an average probability that one sample solution information α is associated with one preset solution topic τ is 1 / T, the preset topic probability threshold is set as σ = 1 / T, and in the application process, the association probability of one sample solution information α and one preset solution topic τ is compared with the preset topic probability threshold σ = 1 / T to obtain a topic probability comparison result, when the topic probability comparison result is that the association probability of one sample solution information α and one preset solution topic τ is greater than σ = 1 / T, it is considered that the sample solution information α belongs to the preset solution topic τ, that is, the preset solution topic τ is taken as a sample solution topic corresponding to the sample solution information α.

[0108] In the embodiment, it is noted that one sample solution information α is simultaneously associated with multiple preset solution topics τ, that is, there is a one-to-many relationship, that is, the sample solution information α can belong to both the preset solution topic τ1 and the preset solution topic τ2.

[0109] 203、determining sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution theme.

[0110] In the embodiment, before the model training, the obtained sample problem customer attribute information, sample solution information and sample solution theme need to be subjected to feature screening.

[0111] Specifically, according to the sample problem customer attribute information, the sample problem customer attribute features corresponding to the sample problem customer in the sample problem customer attribute information are extracted in combination with the user portrait data related to the sample problem customer accumulated by the enterprise. Exemplarily, the sample problem customer attribute features mainly include customer seasonal product order quantity, customer stratification, customer classification, cooperation agreement signing time, customer product and region profit amount and proportion and the like.

[0112] According to the sample solution information and the sample solution theme, the corresponding sample solution related features for solving the problem of the sample problem customer can be determined. Exemplarily, the sample solution related features include the sample solution theme, the execution rate and the improvement rate of the sample solution theme in the last N months and the like.

[0113] Therefore, in the embodiment, specifically, after the sample solution theme is determined according to the sample solution information, the method comprises: acquiring execution probability information and improvement probability information corresponding to the sample solution theme; and taking the sample solution theme, the execution probability information and the improvement probability information as the sample solution related features. In the embodiment, the application situation of the sample solution theme can be judged through the execution probability information and the improvement probability information corresponding to the sample solution theme, which is convenient for helping to score the problem improvement probability of the sample solution theme in the subsequent model training process.

[0114] According to the sample problem customer attribute information, the sample solution information and the sample solution theme, the sample cross multi-dimensional features are determined. The sample cross multi-dimensional features are combination features associated with multi-dimensional single features, and the features contain more feature information. Exemplarily, the sample cross multi-dimensional features can include combination features combining a business rule-based data mining model monitoring the problem of a certain sample problem customer and the improvement situation of sample solution information for solving the problem of the sample problem customer, such as the hit situation of the business rule-based data mining model on the certain sample problem customer, the importance of the data model, and the historical improvement rate of a certain sample solution information on a certain sample problem customer signed within three years and the like.

[0115] 204、obtaining a solution theme recommendation model by taking the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features as model inputs for model training.

[0116] After obtaining the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features, the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features are taken as model inputs for model training to obtain a solution theme recommendation model. The present application is based on an FM model for model training to obtain a solution theme recommendation model, and the specific training manner is as follows.

[0117] Firstly, the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features all need to be encoded. For continuous features such as the number of hits based on business rule data mining model, customer profit, customer contract term, etc., data binning needs to be performed first, and then the features are converted into category features before being input into the model to be trained.

[0118] Therefore, in another embodiment of the present application, after determining the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution theme, the method comprises:

[0119] The sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features are all encoded to obtain a sample customer attribute feature encoding matrix, a sample solution related feature encoding matrix and a sample other cross multi-dimensional feature encoding matrix, respectively.

[0120] The sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix are spliced to obtain a spliced sample feature vector matrix, and the sample feature vector matrix comprises a plurality of sample feature vectors.

[0121] In the present embodiment, the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features are all encoded in a one-hot encoding manner, and the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix are spliced to obtain a sample feature vector matrix as shown in FIG. 5, which comprises a feature vector matrix X and a label matrix Y spliced by the above-mentioned plurality of encoding matrices. Figure 4

[0122] wherein, x (1) ~ x (7) represent different sample problem customers, y (1) ~ y (7) ​The label matrix Y is a numerical value 1 indicating that the problem of the sample problem customer can be improved, and the label matrix Y is a numerical value 0 indicating that the problem of the sample problem customer cannot be improved. The sample feature vector matrix includes a sample customer attribute feature encoding matrix of “customer ID”, a sample solution related feature encoding matrix of “solution theme”, a sample customer attribute feature encoding matrix of “customer level”, a sample other cross multi-dimensional feature encoding matrix of “hit rule-based data mining model”, and the like. Here, only a part of the sample feature vector matrix is shown, and different sample feature vector matrices can be set according to needs, and the sample feature vector matrix is not specifically limited here.

[0123] For the convenience of understanding, the feature vector x (1) The sample feature vector matrix is described. Exemplarily, the feature vector x (1) The one-hot encoding of the sample customer attribute feature encoding matrix of “customer ID” is “10000”, that is, x (1) The value of “whether it is an A customer” in this column is 1, and the values of other columns are 0, indicating that the feature vector x (1) belongs to an A customer.

[0124] The feature vector x (1) The one-hot encoding of the sample solution related feature encoding matrix of “solution theme” is “100…”. Based on the matrix content extracted in the drawing, x (1) The value of “whether it is an A solution theme” in this column is 1, and the values of other columns are 0 (not combined with the matrix content not extracted in the drawing for evaluation), indicating that the feature vector x (1) belongs to an A solution theme.

[0125] The feature vector x (1) The one-hot encoding of the sample customer attribute feature encoding matrix of “customer level” is “01…”, that is, x (1) The value of “whether it is a two-level customer” in this column is 1, and the values of other columns are 0, indicating that the feature vector x (1) belongs to a two-level customer.

[0126] The feature vector x (1) The one-hot encoding of the sample other cross multi-dimensional feature encoding matrix of “hit rule-based data mining model” is “100…”, that is, x (1) The value of “whether it hits model 1” in this column is 1, and the values of other columns are 0, indicating that the sample problem customer in the feature vector x (1) has a problem hit by model 1.

[0127] In summary, in combination with the information of the sample feature vector matrix given in the drawings, i.e., the sample features of the feature vector x (1) include the customer ID A customer, the solution subject belongs to the A solution subject, the customer level is a two-level customer, and the sample problem customer in the feature vector x (1) The sample features hit by Model 1 are included.

[0128] After the sample problem customer attribute features, sample solution related features, and sample cross multi-dimensional features are encoded using one-hot encoding, the feature vector dimension of the obtained sample feature vector matrix is high, and the values of most features in the sample feature vector matrix are 0, and the feature matrix distribution is very sparse (data sparsity is greater than 90%). Using a dense matrix will greatly consume storage resources and computing resources, so it is necessary to adopt a zero-out storage mode to optimize the space complexity, and the specific optimization method is as follows.

[0129] In another embodiment of the present application, after the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix, and the sample other cross multi-dimensional feature encoding matrix are spliced to obtain the spliced sample feature vector matrix, the method comprises:

[0130] The spliced sample feature vector matrix is converted into a sample feature vector matrix in a zero-out storage format by a preset zero-out format, wherein the zero-out format is:

[0131] y index1:value1 index2:value2... index n :value n

[0132] y represents the improvement probability label in the sample feature vector matrix, index i represents the feature column index of the non-zero value in the sample feature vector matrix, value i represents the feature value corresponding to index i in the sample feature vector matrix, wherein i∈n.

[0133] In this embodiment, the sample feature vector matrix is converted into a sample feature vector matrix in a zero-out storage format according to the zero-out format, and the sample feature vector matrix in the zero-out storage format is obtained. The libsvm storage format is used as the zero-out storage format in this embodiment, and the sample feature vector matrix in the libsvm storage format can be as shown in Table 1.

[0134] Table 1 Sample feature vector matrix in libsvm storage format

[0135] 0 0:1 1:1 3:1 … 0 0:1 2:1 5:1 … 1 0:1 2:1 8:1 … 0 0:1 3:1 5:1 … 1 1:1 2:1 4:1 …

[0136] For the convenience of understanding, the first row of data in Table 1 and the accompanying Figure 4 The characteristic data in Table 1 is explained and described.

[0137] Specifically, the first characteristic data in the first row of Table 1 is "0", which corresponds to Figure 4 "Label Y" in "Label Y" is "0", which means that the sample problem customer corresponding to the feature vector cannot be improved;

[0138] The second characteristic data in the first row of Table 1 is "0:1", which corresponds to Figure 4 The sample customer attribute feature encoding matrix of "Customer ID", index1:value1=0:1, indicates that the feature column index of the non-zero value of the sample customer attribute feature encoding matrix corresponding to the feature vector of "Customer ID" is 0, that is, the feature vector belongs to the feature attribute of the 0th "Customer ID", that is, the feature vector belongs to "A customer";

[0139] The third characteristic data in the first row of Table 1 is "1:1", which corresponds to Figure 4 The sample solution related feature encoding matrix of "Solution Topic", index2:value2=1:1, indicates that the feature column index of the non-zero value of the sample solution related feature encoding matrix corresponding to the feature vector of "Solution Topic" is 1, that is, the feature vector belongs to the 1st "Solution Topic" feature attribute, that is, the feature vector belongs to "b solution topic";

[0140] The fourth characteristic data in the first row of Table 1 is "3:1", which corresponds to Figure 4 The sample customer attribute feature encoding matrix of "Customer Level", index3:value3=3:1, indicates that the feature column index of the non-zero value of the sample customer attribute feature encoding matrix corresponding to the feature vector of "Customer Level" is 3, that is, the feature vector belongs to the 3rd "Solution Topic" feature attribute, that is, the feature vector belongs to "four-level customer" Figure 4 (not shown in the figure), which is inferred according to Figure 4

[0141] Using this zero storage method to store the sample feature vector matrix can optimize the space complexity, reduce the storage cost, and has the advantages of easy deployment, etc.

[0142] In another embodiment of the present application, the sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features are used as model inputs for model training to obtain a solution topic recommendation model, comprising:​

[0143] establishing a nonlinear prediction model;

[0144] dividing all sample features in the sample feature vector matrix in the zero-out storage format into positive sample features and negative sample features, the positive sample features including sample feature vectors corresponding to sample problem customers whose problems are improved according to sample solution information, and the negative sample features including sample feature vectors corresponding to sample problem customers whose problems are not improved according to sample solution information;

[0145] training the nonlinear prediction model by taking the positive sample features and the negative sample features as model inputs to obtain a solution theme recommendation model.

[0146] In an embodiment, a sample feature vector matrix is obtained by converting sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features, and the sample feature vector matrix is divided into positive sample features and negative sample features. In the context of predicting whether sample solution information is adopted, the positive sample features and the negative sample features are taken as inputs, and the probability that a certain sample solution information can bring business improvement of the enterprise is taken as output for model training based on an FM model to obtain a solution theme recommendation model.

[0147] In this embodiment, the nonlinear prediction model uses a nonlinear prediction equation of an FM model. The nonlinear prediction equation of the FM model includes two formats. Based on the original nonlinear prediction equation, a ReduceSum function is used to represent the sum of elements at all positions of a vector to obtain a simplified nonlinear prediction equation. The two nonlinear prediction equations are described in detail below.

[0148] In another embodiment of the present application, the nonlinear prediction model includes:

[0149]

[0150] wherein y1^(x 1i ) is the probability that a certain sample solution information can improve the problem of a sample problem customer, x 1i is a sample feature vector in a sample feature vector matrix, each sample feature vector x 1i has n sample feature elements, ω 1i , v 1i and v 1j are parameters that need to be adjusted by the model, ω 10 is a global bias of the nonlinear prediction model, ω 1i is a feature influence intensity, v 1i and v 1j represent k-dimensional vectors of the i-th feature and the j-th feature, respectively, and <v1i ,v 1j >is v 1i and v 1j The inner product of two sample vectors in the sample feature vector matrix represents the interaction between the two sample feature vectors, i.e., the corresponding latent vectors of the two sample vectors in the sample feature vector matrix.

[0151] In another embodiment of the present application, based on the original nonlinear prediction model, the nonlinear prediction model can be simplified, i.e., the nonlinear prediction model can also include:

[0152]

[0153] where y2^(x 2i ) is the probability that the sample solution information can improve the problem of the sample problem customer, x 2i is a sample feature vector in the sample feature vector matrix, each sample feature vector x 2i has n sample features, ω 2i , v 2i and v 2j are parameters that need to be adjusted by the model, ω 20 is the global bias of the nonlinear prediction model, ω 2i is the feature influence strength, v 2i and v 2j represent the k-dimensional vectors of the i-th feature and the j-th feature, respectively, <v 2i , v 2j >is v 2i and v 2j The inner product of two sample vectors in the sample feature vector matrix represents the interaction between the two sample feature vectors, i.e., the corresponding latent vectors of the two sample vectors in the sample feature vector matrix, and the ReduceSum function represents the sum of the sample feature elements in all positions of a sample feature vector.

[0154] In this embodiment, the ReduceSum function represents the sum of the elements in all positions of a vector. Through this zero-padded calculation, the time complexity can be reduced from O(n 2 ) to O(n), where n represents the number of non-zero features in the sample feature vector matrix. Although the feature space of the algorithm is in the order of millions of dimensions, since the features are very sparse in the recommendation scenario, n of each sample is very limited, so the speed of model training and prediction can be improved.

[0155] In another embodiment of the present application, based on the simplified nonlinear prediction model, the training method of model training specifically includes the following steps 301-305:

[0156] 301、model input, determine the simplified nonlinear prediction model as the model to be trained.

[0157] 302、determine the training set and the validation set.

[0158] The positive sample features and the negative sample features obtained by the division are divided into the training set and the validation set according to the preset proportion, the training set obtained contains the positive sample training features and the negative sample training features, and the validation set obtained contains the positive sample validation features and the negative sample validation features.

[0159] In the subsequent model training process, the positive sample training features and the negative sample training features in the training set are used for model training to obtain a trained initial solution theme recommendation model, the positive sample validation features and the negative sample validation features in the validation set are used for verifying the initial solution theme recommendation model, and the related parameters in the model are adjusted according to the verification result.

[0160] 303、initialize the parameter ω 20 = 0, ω 2i = (0,..., 0), V ~ N(0, σ).

[0161] 304、use the stochastic gradient descent method to estimate the parameters in the model.

[0162] Since the FM model regards the recommendation problem as a classification problem (the label y2^(x 2i ) ∈ {0, 1}), the output of the initial model needs to be first subjected to a Sigmoid transformation, so that the output result of the label y2^(x 2i ) is mapped to 0 or 1.

[0163] In the model training process, a first loss function is used to calculate the model loss to obtain a plurality of first loss values, in this embodiment, the first loss function used is a Cross Entropy loss function. That is, the Cross Entropy loss function is specifically:

[0164]

[0165] wherein,

[0166] In the loss function, x 2i is the feature value of the i-th sample feature vector in the sample feature vector matrix, y2 represents the label value of the i-th sample feature vector in the sample feature vector matrix, and σ(y2^) is the probability that the problem of the customer corresponding to the i-th sample feature vector can be improved.

[0167] In this embodiment, the closer the value of y2 is to the value of y2^, the smaller the value of the loss function loss(y2^,y2). To prevent overfitting, L2 regularization is applied to this loss function, transforming the optimization objective of the FM model into:

[0168]

[0169] Where λ is the regularization coefficient, and ε is the parameter group, i.e., ω 20 ω 2i and <v 2i ,v 2j >

[0170] 305. Model hyperparameter adjustment.

[0171] In this embodiment, 10-fold cross-validation is used to test the FM model during training. The FM model is implemented based on the xlearn library, and the main parameters involved are: task (binary classification in this example), learning rate, metric (f1 score in this example), stop window (early stopping window length), epoch (number of iterations), k (dimension of each latent vector), etc. Bayesian optimization is used to adjust the above parameters.

[0172] Experiments show that as the factorization dimension k increases, the f1_score also increases, meaning the algorithm's accuracy improves. However, this also increases the model's computational space overhead and reduces computation speed. Figure 5 As shown, Figure 5 The horizontal axis of the coordinate plot represents the factorization dimension. Figure 5 The vertical axis of the coordinate graph represents the value of f1_score. When the factorization dimension increases from 2 to 12, f1_score shows a significant improvement, but when the factorization dimension is greater than 12, f1_score begins to decrease. Therefore, to improve the efficiency and accuracy of the algorithm, in this embodiment, the factorization dimension is set to k = 12.

[0173] In addition to adjusting the hyperparameters of FM itself, it is also necessary to adjust the number of preset solution topics τ involved in LDA. For example... Figure 6 As shown, Figure 6 The horizontal axis of the coordinate graph represents the number of preset solution topics τ. Figure 6The vertical axis of the coordinate diagram is the value of f1_score. As the number of preset solution topics τ increases, the value of f1_score gradually increases. When the number of solution topics τ increases from 5 to 15, f1_score increases more obviously, and when the number of solution topics τ is more than 15, the value of f1_score tends to be relatively stable. Therefore, the total number of solution topics τ is set to T = 15.

[0174] 305, model output.

[0175] During the FM model training process, the FM model continuously learns and adjusts the parameters ω 20 and the parameters ω 2i In the end, each parameter obtained by optimization is stored in a row, and all calculated parameters are saved. Based on the FM model and the parameters obtained by optimization, a solution topic recommendation model is finally obtained.

[0176] In the present application, since the business of the enterprise is running every day, the data mining model based on business rules can help the enterprise quickly mine users who are suspected to have abnormal cooperation behavior. The sample solution topic corresponding to the sample solution information is extracted by LDA, and the solution topic recommendation model is trained to automatically form the recommended target solution. The recommended solution is output to the enterprise professional for review and modification, and then the solution is implemented by the execution layer. The online, automated and intelligent recommendation of the target solution to the problem user can save energy, improve problem solving efficiency, achieve fast and comprehensive solution of the problem of the problem customer, and provide important decision information assistance for fine management and improvement of the problem customer. It helps companies to operate leanly and effectively reduce costs. Compared with existing problem solution recommendation algorithms, the LDA-FM algorithm framework proposed in the present application has the advantages of high efficiency, low cost and easy deployment, and can be widely promoted in a short time.

[0177] In order to better implement the problem solution recommendation method in the embodiments of the present application, on the basis of the problem solution recommendation method, the embodiments of the present application also provide a problem solution recommendation system. As shown in Figure 7 The problem solution recommendation system 400 includes:

[0178] The acquisition module 401 is configured to acquire target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer.

[0179] The prediction module 402 is configured to perform solution theme prediction according to the target customer attribute feature and the target cross multi-dimensional feature, to obtain a solution theme prediction result corresponding to the target customer information, and the solution theme prediction result includes a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively.

[0180] The sorting module 403 is configured to sort the plurality of to-be-recommended solution themes according to the sizes of the plurality of solution theme improvement probabilities, to obtain a sorted solution theme set.

[0181] The recommendation module 404 is configured to take a to-be-recommended solution theme arranged at a target position in the sorted solution theme set as a target solution theme corresponding to the target problem customer, and recommend the target solution theme to a target recommendation terminal.

[0182] The system further includes an encoding module 407, and the encoding module 407 is specifically configured to:

[0183] perform encoding processing on the target customer attribute feature and the target cross multi-dimensional feature, to obtain a customer attribute feature vector matrix and a cross multi-dimensional feature vector matrix.

[0184] perform vector splicing on the customer attribute feature vector matrix and the cross multi-dimensional feature vector matrix, to obtain a target customer feature vector set.

[0185] The prediction module 402 is further specifically configured to:

[0186] input the target customer feature vector set into a preset solution theme recommendation model to perform solution theme prediction, to obtain the solution theme prediction result corresponding to the target customer information.

[0187] The system further includes a training module 408, and the training module 408 is specifically configured to:

[0188] obtain sample problem customer attribute information corresponding to a sample problem customer and sample solution information corresponding to the sample problem customer attribute information;

[0189] determine a sample solution theme according to the sample solution information;

[0190] determine sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution theme;

[0191] perform model training by taking the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features as model input, to obtain the solution theme recommendation model.

[0192] The subject extraction module 408 is specifically configured to:

[0193] determine a relevant subject probability that the sample solution information belongs to the preset solution subject according to the sample solution information;

[0194] compare the relevant subject probability with a preset subject probability threshold to obtain a subject probability comparison result;

[0195] determine whether the sample solution information belongs to the preset solution subject according to the subject probability comparison result;

[0196] when the sample solution information belongs to the preset solution subject, take the preset solution subject as the sample solution subject corresponding to the sample solution information.

[0197] The training module 405 is specifically configured to:

[0198] obtain execution probability information and improvement probability information corresponding to the sample solution subject;

[0199] take the sample solution subject, the execution probability information and the improvement probability information as sample solution related features.

[0200] The encoding module 407 is specifically configured to:

[0201] perform encoding processing on the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features to obtain a sample customer attribute feature encoding matrix, a sample solution related feature encoding matrix and a sample other cross multi-dimensional feature encoding matrix, respectively;

[0202] splice the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix to obtain a spliced sample feature vector matrix.

[0203] The system further comprises a zero elimination module 406, which is specifically configured to:

[0204] convert the spliced sample feature vector matrix into a sample feature vector matrix in a zero elimination storage format through a preset zero elimination format, wherein the zero elimination format is:

[0205] y index1: value1 index2: value2... index n : value n

[0206] y represents an improvement probability label in the sample feature vector matrix, index iThe value represents the index of the feature column containing the non-zero value in the sample feature vector matrix. i The index represents the sample feature vector matrix. i The corresponding eigenvalues, where i∈n.

[0207] Training module 405 is further specified as follows:

[0208] Used to establish nonlinear prediction models;

[0209] This is used to train a nonlinear prediction model by using a zero-placing sample feature vector matrix as model input, resulting in a solution topic recommendation model.

[0210] In another embodiment of this application, such as Figure 8 As shown, this application also provides an electronic device 500, which illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application. Specifically:

[0211] The electronic device may include components such as a processor 501 with one or more processing cores, a memory 502 with one or more computer-readable storage media, a power supply 503, and an input unit 504. Those skilled in the art will understand that... Figure 8 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0212] The processor 501 is the control center of the electronic device, connects all parts of the electronic device through various interfaces and lines, executes various functions of the electronic device and processes data by running or executing software programs and / or modules stored in the memory 502 and calling data stored in the memory 502, thereby overall monitoring the electronic device. Optionally, the processor 501 can include one or more processing cores; the processor 501 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like, preferably, the processor 501 can integrate an application processor and a modem processor, wherein the application processor mainly processes operating systems, user interfaces, application programs and the like, and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 501.

[0213] The memory 502 can be used to store software programs and modules, and the processor 501 executes various functions and data processing by running the software programs and modules stored in the memory 502. The memory 502 can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, application programs required by at least one function (such as sound playing function, image playing function, etc.), etc.; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 502 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, flash memory device, or other volatile solid-state memory device. Accordingly, the memory 502 can also include a memory controller to provide access of the processor 501 to the memory 502.

[0214] The electronic device further includes a power supply 503 for supplying power to each component, and preferably, the power supply 503 can be logically connected to the processor 501 through a power management system, so as to realize functions such as management of charging, discharging and power consumption management through the power management system. The power supply 503 can also include one or more than one direct current or alternating current power supply, a recharging system, a power failure detection circuit, a power converter or inverter, a power state indicator and any other components.

[0215] The electronic device can further include an input unit 504, which can be used to receive inputted digital or character information, and to generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0216] Although not shown, the electronic device can further include a display unit or the like, which will not be described here. In particular, in the present embodiment, the processor 501 in the electronic device will load executable files corresponding to the processes of one or more application programs into the memory 502 according to the following instructions, and run the application programs stored in the memory 502 by the processor 501, thereby implementing various functions, as follows:

[0217] Obtaining target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer;

[0218] According to the target customer attribute features and the target cross multi-dimensional features, performing solution theme prediction to obtain a solution theme prediction result corresponding to the target customer information, the solution theme prediction result including a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities respectively corresponding to the plurality of to-be-recommended solution themes;

[0219] According to the sizes of the plurality of solution theme improvement probabilities, sorting the plurality of to-be-recommended solution themes to obtain a sorted solution theme set;

[0220] Arranging a to-be-recommended solution theme arranged in a target position in the solution theme set as a target solution theme corresponding to the target problem customer, and recommending the target solution theme to a target recommendation terminal.

[0221] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by instructions controlling relevant hardware, which can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0222] In some embodiments of the present application, the present application further provides a computer-readable storage medium, which can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in the problem solution recommendation method provided by the embodiments of the present application. For example, the computer program loaded by the processor can execute the following steps:

[0223] Obtaining target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer;

[0224] According to the target customer attribute feature and the target cross multi-dimensional feature, solution theme prediction is performed to obtain a solution theme prediction result corresponding to the target customer information, the solution theme prediction result including a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities respectively corresponding to the plurality of to-be-recommended solution themes;

[0225] The plurality of to-be-recommended solution themes are sorted according to the sizes of the plurality of solution theme improvement probabilities to obtain a sorted solution theme set;

[0226] A to-be-recommended solution theme arranged at a target position in the solution theme set is taken as a target solution theme corresponding to the target problem customer, and the target solution theme is recommended to a target recommendation terminal.

[0227] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the detailed description of other embodiments above, which will not be described here again.

[0228] The above describes in detail a problem solution recommendation method, system, electronic device and storage medium provided by the embodiments of the present application. The principles and implementation manners of the present application are described by applying specific examples. The above embodiment description is only used to help understand the method and core idea of the present application. Meanwhile, for those skilled in the art, the specific implementation manner and application range of the present application can be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for recommending a solution to a problem, characterized in that, target customer attribute features and target cross-multiple features corresponding to a target problem customer are obtained; the target customer attribute features include one or more of the following customer basic attribute features: customer identification, customer stratification, customer classification, or cooperation agreement signing time; the target cross-multiple features include one or more of the following features: a hit of the target problem customer based on a business rule data model, an importance of the business rule data model, or a historical improvement rate of the target problem customer who has signed an N-year cooperation agreement, the N being a positive integer; the target cross-multiple features represent a multi-dimensional feature formed by a combination of two or more features of the target problem customer; solution theme prediction is performed according to the target customer attribute features and the target cross-multiple features, to obtain a solution theme prediction result corresponding to the target problem customer, the solution theme prediction result including a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively; the solution theme improvement probability corresponding to the to-be-recommended solution theme represents a probability that the to-be-recommended solution theme can improve a problem existing in the target problem customer; the solution theme prediction result is obtained by a preset solution theme recommendation model performing operation on the target customer attribute features and the target cross-multiple features and learning the target cross-multiple features; the plurality of to-be-recommended solution themes are sorted according to the sizes of the plurality of solution theme improvement probabilities, to obtain a sorted solution theme set; a to-be-recommended solution theme arranged at a target position in the solution theme set is taken as a target solution theme corresponding to the target problem customer, and the target solution theme is recommended to a target recommendation terminal.

2. The problem solution recommendation method of claim 1, wherein, After the target customer attribute features and the target cross-multiple features corresponding to the target problem customer are obtained, the method includes: the target customer attribute features and the target cross-multiple features are respectively encoded to obtain a customer attribute feature vector matrix and a cross-multiple feature vector matrix; the customer attribute feature vector matrix and the cross-multiple feature vector matrix are spliced to obtain a target customer feature vector set.

3. The problem solution recommendation method of claim 2, wherein, The solution theme prediction according to the target customer attribute features and the target cross-multiple features to obtain the solution theme prediction result corresponding to the target problem customer includes: the target customer feature vector set is input into a preset solution theme recommendation model to perform solution theme prediction, to obtain the solution theme prediction result corresponding to the target problem customer.

4. The problem solution recommendation method of claim 3, wherein, Before the target customer attribute features and the target cross-multiple features corresponding to the target problem customer are obtained, the method includes: sample problem customer attribute information corresponding to a sample problem customer and sample solution information corresponding to the sample problem customer attribute information are obtained; sample solution themes are determined according to the sample solution information; According to the sample problem customer attribute information, the sample solution information and the sample solution topic, determine sample problem customer attribute features, sample solution related features and sample cross multi-dimensional features; Train a model by taking the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features as model inputs to obtain the solution topic recommendation model.

5. The problem solution recommendation method of claim 4, wherein, The determination of the sample solution topic according to the sample solution information comprises: According to the sample solution information, determine a related topic probability that the sample solution information belongs to a preset solution topic; Compare the related topic probability with a preset topic probability threshold to obtain a topic probability comparison result; According to the topic probability comparison result, determine whether the sample solution information belongs to a preset solution topic; When the sample solution information belongs to a preset solution topic, take the preset solution topic as the sample solution topic corresponding to the sample solution information.

6. The problem solution recommendation method of claim 5, wherein, After the determination of the sample solution topic according to the sample solution information, the method comprises: Obtain execution probability information and improvement probability information corresponding to the sample solution topic; Take the sample solution topic, the execution probability information and the improvement probability information as the sample solution related features.

7. The problem solution recommendation method of claim 4, wherein, After the determination of the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features according to the sample problem customer attribute information, the sample solution information and the sample solution topic, the method comprises: Encode the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features to obtain a sample customer attribute feature encoding matrix, a sample solution related feature encoding matrix and a sample other cross multi-dimensional feature encoding matrix respectively; Splice the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix to obtain a spliced sample feature vector matrix, and the sample feature vector matrix comprises a plurality of sample feature vectors.

8. The problem solution recommendation method of claim 7, wherein, After the splicing of the sample customer attribute feature encoding matrix, the sample solution related feature encoding matrix and the sample other cross multi-dimensional feature encoding matrix to obtain a spliced sample feature vector matrix, the method comprises: According to a preset zero-out format, convert the spliced sample feature vector matrix into a sample feature vector matrix in a zero-out storage format, wherein the zero-out format is: y represents the improvement probability label in the sample feature vector matrix, index i value represents the feature column index of the non-zero value in the sample feature vector matrix, value i value represents the feature column index of the non-zero value in the sample feature vector matrix, value i corresponding feature value, wherein, i∈n, n is a positive integer.

9. The problem solution recommendation method of claim 8, wherein, The training of the model by taking the sample problem customer attribute features, the sample solution related features and the sample cross multi-dimensional features as model inputs to obtain the solution topic recommendation model comprises: Establish a nonlinear prediction model; Divide all sample features in the sample feature vector matrix in the zero storage format into positive sample features and negative sample features, the positive sample features including sample feature vectors corresponding to sample problem customers whose problems are improved according to the sample solution information, and the negative sample features including sample feature vectors corresponding to sample problem customers whose problems are not improved according to the sample solution information; Model train the nonlinear prediction model by taking the positive sample features and the negative sample features as model inputs, to obtain the solution theme recommendation model.

10. A system, characterized by The system comprises: An acquisition module configured to acquire target customer attribute features and target cross multi-dimensional features corresponding to a target problem customer; the target customer attribute features include one or more of the following customer basic attribute features: customer identification, customer stratification, customer classification, or cooperation agreement signing time; the target cross multi-dimensional features include one or more of the following features: a hit of the target problem customer based on a business rule data model, an importance of the business rule data model, or a historical improvement rate of the target problem customer who has signed an N-year cooperation agreement, where N is a positive integer; the target cross multi-dimensional features represent a multi-dimensional feature formed by a combination of two or more features of the target problem customer; A prediction module configured to perform solution theme prediction based on the target customer attribute features and the target cross multi-dimensional features, to obtain a solution theme prediction result corresponding to the target problem customer, the solution theme prediction result including a plurality of to-be-recommended solution themes and a plurality of solution theme improvement probabilities corresponding to the plurality of to-be-recommended solution themes respectively; the solution theme improvement probability corresponding to a to-be-recommended solution theme represents a probability that the to-be-recommended solution theme can improve a problem existing in the target problem customer; the solution theme prediction result is obtained by performing operation on the target customer attribute features and the target cross multi-dimensional features by a preset solution theme recommendation model and learning the target cross multi-dimensional features; An ordering module configured to order the plurality of to-be-recommended solution themes according to the sizes of the plurality of solution theme improvement probabilities, to obtain an ordered solution theme set; A recommendation module configured to recommend, as a target solution theme corresponding to the target problem customer, a to-be-recommended solution theme arranged at a target position in the solution theme ordered set, and to recommend the target solution theme to a target recommendation terminal.

11. An electronic device, comprising: The electronic device comprises: one or more processors; a memory; and one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to implement the problem solution recommendation method in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, A computer program is stored thereon, and the computer program is loaded by a processor to perform steps in the problem solution recommendation method in any one of claims 1 to 9.

Images