Training method and device of consumption scene preference prediction model, equipment and medium

Abstract

The embodiment of the invention provides a consumption scene preference prediction model training method and device, equipment and a medium. The method comprises the following steps: acquiring historical multi-dimensional information of a user; according to the historical multi-dimensional information, determining a consumption scene preference score of the user in a predefined consumption scene; screening out a target consumption scene with the maximum consumption scene preference score from the consumption scenes; constructing a model training sample according to the historical multi-dimensional information and the target consumption scene; and training based on the model training sample to obtain a consumption scene preference prediction model for consumption scene preference prediction. According to the embodiment of the invention, the marketing effect and the user satisfaction are improved, operation and resource configuration are optimized, precise marketing is realized, and numerous key scenes of credit card customer operation and marketing are covered.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and particularly to a method, device, equipment and medium for training a consumption scenario preference prediction model. Background Art

[0002] With the increasingly fierce competition in the credit card market, the increasingly diverse consumption demands of customers, and the development requirements of the digital transformation of credit cards, how to accurately identify customer preferences and precisely match the rights and interests resources with customer needs is an important topic in the current customer operation and marketing of the credit card industry. Predicting the future consumption scenario preferences of users based on user characteristics and then carrying out refined marketing according to their preferences is a good solution.

[0003] The existing preference prediction methods usually include: predicting the preference of a user for a specific object, and marketing methods based on the user portrait and consumption portrait of the customer's history, etc.

[0004] Predicting the preference of a user for a specific object, including activities, goods, services, etc., is not applicable when the business scenario or goods are changed, and the universality is insufficient. The marketing methods based on the user portrait and consumption portrait of the customer's history rely more on expert experience. However, due to the variability of user behavior, the user portrait generated from past data only reflects a part of the user behavior, and there are limitations and biases, and it does not necessarily reflect future consumption behavior. And in marketing, it is often expected to accurately predict the future consumption behavior of customers. Therefore, the marketing scheme based on the user's historical preferences is not fully applicable, and there is a need to scientifically and accurately predict the future consumption preferences of customers. Summary of the Invention

[0005] The technical problem to be solved by the embodiments of the present application is to provide a method, device, equipment and medium for training a consumption scenario preference prediction model, so as to effectively improve the marketing effect and user satisfaction, optimize operation and resource allocation, achieve precise marketing, and cover many key scenarios of credit card customer operation and marketing.

[0006] In a first aspect, an embodiment of the present application provides a method for training a consumption scenario preference prediction model, the method comprising:

[0007] Obtaining historical multi-dimensional information of a user;

[0008] Determining a consumption scenario preference score of the user in a predefined consumption scenario according to the historical multi-dimensional information;

[0009] Screening out a target consumption scenario with the largest consumption scenario preference score from the consumption scenarios;

[0010] Constructing a model training sample according to the historical multi-dimensional information and the target consumption scenario;

[0011] Based on the model training samples, a consumption scenario preference prediction model for predicting consumption scenario preferences is trained.

[0012] In a second aspect, an embodiment of the present application provides a training device for a consumption scenario preference prediction model, and the device includes:

[0013] A historical multi-dimensional information acquisition module, configured to acquire historical multi-dimensional information of a user;

[0014] A consumption scenario preference determination module, configured to determine a consumption scenario preference score of the user in a predefined consumption scenario according to the historical multi-dimensional information;

[0015] A target consumption scenario screening module, configured to screen out a target consumption scenario with the largest consumption scenario preference score from the consumption scenarios;

[0016] A model training sample construction module, configured to construct model training samples according to the historical multi-dimensional information and the target consumption scenario;

[0017] A scenario preference prediction model training module, configured to train a consumption scenario preference prediction model for predicting consumption scenario preferences based on the model training samples.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0019] A processor, a memory, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, the training method of the consumption scenario preference prediction model described in any one of the above is implemented.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, and when instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the training method of the consumption scenario preference prediction model described in any one of the above.

[0021] Compared with the prior art, the embodiments of the present application have the following advantages:

[0022] In the embodiments of the present application, historical multi-dimensional information of a user is obtained. According to the historical multi-dimensional information, a consumption scenario preference score of the user in a predefined consumption scenario is determined. A target consumption scenario with the largest consumption scenario preference score is selected from the consumption scenarios. Based on the historical multi-dimensional information and the target consumption scenario, a model training sample is constructed. A consumption scenario preference prediction model for predicting consumption scenario preferences is trained based on the model training sample. In the embodiments of the present application, the preferences of the user for future consumption in various scenarios are predicted through user characteristics. Combining the scenario preference ranking results, a differentiated operation strategy is formulated, corresponding rights and interests, products and services are provided to the user, the marketing effect and user satisfaction are improved, the operation and resource allocation are optimized, so as to achieve precision marketing. At the same time, this solution can implement a complete personalized marketing operation strategy solution, and its application can cover many key scenarios of credit card customer operation and marketing.

[0023] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flowchart of the steps of a method for training a consumption scenario preference prediction model provided by an embodiment of the present application;

[0025] Figure 2 It is a flowchart of the steps of a method for determining a consumption scenario preference score provided by an embodiment of the present application;

[0026] Figure 3 It is a flowchart of the steps of another method for determining a consumption scenario preference score provided by an embodiment of the present application;

[0027] Figure 4 It is a flowchart of the steps of a method for matching consumption products provided by an embodiment of the present application;

[0028] Figure 5 It is a schematic diagram of a consumption scenario preference prediction and model strategy system process provided by an embodiment of the present application;

[0029] Figure 6 It is a schematic diagram of the structure of a training device for a consumption scenario preference prediction model provided by an embodiment of the present application;

[0030] Figure 7 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] The terms used in the embodiments of the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "the" and "said" used in the embodiments of the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0033] Referring to Figure 1 , a flowchart showing the steps of a method for training a consumption scenario preference prediction model provided by an embodiment of the present application is shown. As Figure 1 shown, the method for training the consumption scenario preference prediction model may include: Step 101, Step 102, Step 103, Step 104, and Step 105.

[0034] Step 101: Obtain the historical multi-dimensional information of the user.

[0035] The embodiments of the present application can be applied to scenarios where model training samples are constructed based on the scenario preferences of users in predefined consumption scenarios to train a consumption scenario preference prediction model.

[0036] When training a consumption scenario preference prediction model, the historical multi-dimensional information of the user can be obtained. Among them, the historical multi-dimensional information may include at least one of the following multi-dimensional information: the basic information of the user (such as gender, age, etc.), consumption behavior information, credit card type information used, etc. In a specific implementation, during the observation period of the user (i.e., a specific time range for collecting and analyzing user data), the basic attributes of the user (such as age, gender, occupation, etc.), card usage behavior (such as card type, card activation time, quick payment card binding, etc.), and consumption behavior information (such as consumption amount, consumption frequency, consumption type, consumption time, etc.) can be recorded and collected. These data are the basis for subsequent analysis and modeling. Based on the data collected during the observation period, various features of the user can be further mined and extracted. These features can be basic statistics (such as mean, median, mode, etc.), or more complex patterns or trends (such as continuous growth of user consumption, consumption habits, preferences, etc.).

[0037] It can be understood that before collecting user data, it is necessary to clearly inform the user of the purpose and scope of data collection and use, that is, obtain user authorization. If the user agrees to data collection, the browsing, purchasing, clicking and other behavior data of the user can be collected through the log recording function of the website or application.

[0038] After obtaining the historical multi-dimensional information of the user, Step 102 is executed.

[0039] Step 102: Determine the consumption scenario preference score of the user in a predefined consumption scenario according to the historical multi-dimensional information.

[0040] In this embodiment, consumption scenarios can be predefined. Specifically, by conducting a unified classification and inventory of various marketing benefit types and card products in the industry, NLP (Natural Language Processing) technology is applied to extract consumption keywords from consumption data. Through the processing of clustering algorithms, the mainstream user consumption scenarios in the credit card industry can be sorted out and summarized. The specific categories can include: offline dining (the consumption scenario where users consume at physical dining places such as restaurants and cafes), shopping malls and department stores (the consumption scenario where users consume at physical retail places such as supermarkets and department stores), e-commerce shopping (the consumption scenario where users purchase goods or services on online shopping platforms), recharge and payment (the scenario where users recharge and pay for living expenses such as phone bills, electricity bills, and water bills), overseas consumption (the consumption scenario where users consume overseas or use foreign currencies), and other user consumption scenarios.

[0041] Of course, in actual applications, user consumption scenarios not only include the consumption scenarios listed above, but may also include other consumption scenarios. Specifically, it can be determined according to the actual situation, and this embodiment does not limit this.

[0042] The above consumption scenarios not only cover the daily needs of users, but also reflect the wide application of credit cards in various payment scenarios. By deeply understanding these consumption scenarios, credit card issuers can formulate more accurate marketing strategies and product design plans to meet the diverse needs of users. At the same time, these consumption scenarios also provide rich data sources and analysis bases for financial institutions, helping to improve the risk management level and profitability of credit card business.

[0043] After obtaining the user's historical multi-dimensional information, the consumption scenario preference score of the user in the predefined consumption scenario can be determined according to the user's historical multi-dimensional information. Among them, the consumption scenario preference score can be used to indicate the consumption preference of the user in the predefined consumption scenario. The higher the consumption scenario preference score, the greater the consumption preference degree of the user in the corresponding consumption scenario.

[0044] In this embodiment, information such as the consumption amount and the number of consumption transactions of the user in the predefined consumption scenario can be obtained according to the user's historical multi-dimensional information, and the consumption scenario preference score of the user in the consumption scenario can be calculated. For this implementation process, it can be combined with Figure 2 to be described in detail as follows.

[0045] Referring to Figure 2 , a flowchart showing the steps of a method for determining a consumption scenario preference score provided by an embodiment of the present application is shown. As Figure 2 shown, the method for determining the consumption scenario preference score may include: step 201 and step 202.

[0046] Step 201: Determine the scene amount ranking information, scene consumption transaction count ranking information, personal amount ranking information, and personal consumption transaction count ranking information of the user in the consumption scenario based on the historical multi-dimensional information.

[0047] In this embodiment, the scene amount ranking information refers to the ranking of the user's consumption amount in the consumption scenario among the consumption amounts of all users in the consumption scenario. That is, in a specific consumption scenario (such as offline dining, shopping malls, etc.), the ranking of the user's consumption amount relative to the total consumption amount of all users in that scenario. The higher the ranking, the higher the user's consumption amount in that consumption scenario, indicating that the user has a higher consumption preference or demand for that scenario.

[0048] The scene consumption transaction count ranking information refers to the ranking of the user's consumption transaction count in the consumption scenario among the consumption transaction counts of all users in the consumption scenario. That is, in all the consumption scenarios of a user, the ranking of the consumption amount of a specific consumption scenario relative to the total consumption amount of all the user's consumption scenarios. The higher the ranking, the more important the position of that consumption scenario in the user's consumption habits, indicating the most frequent or most emphasized consumption area of the user.

[0049] The personal amount ranking information refers to the ranking of the user's consumption amount in the consumption scenario among the consumption amounts in all the consumption scenarios of the user. That is, in a specific consumption scenario (such as e-commerce shopping, life services, etc.), the ranking of the user's consumption transaction count relative to the total consumption transaction count of all users in that scenario. The higher the ranking, the higher the consumption activity of the user in that consumption scenario, indicating that the user has a higher usage frequency or dependence on that scenario.

[0050] The personal consumption transaction count ranking information refers to the ranking of the user's consumption transaction count in the consumption scenario among the consumption transaction counts in all the user's consumption scenarios. That is, in all the consumption scenarios of a user, the ranking of the consumption transaction count of a specific consumption scenario relative to the total consumption transaction count of all the user's consumption scenarios. The higher the ranking, the more important the position of that consumption scenario in the user's consumption habits, indicating the most commonly used or preferred consumption method of the user.

[0051] After obtaining the user's historical multi-dimensional information, the scene amount ranking information, scene consumption transaction count ranking information, personal amount ranking information, and personal consumption transaction count ranking information of the user in the consumption scenario can be obtained based on the historical multi-dimensional information.

[0052] Step 202: Determine the consumption scenario preference score of the user in the consumption scenario based on the scene amount ranking information, the scene consumption transaction count ranking information, the personal amount ranking information, and the personal consumption transaction count ranking information.

[0053] After obtaining the scene amount ranking information, the scene consumption transaction count ranking information, the individual amount ranking information, and the individual consumption transaction count ranking information, the consumption scene preference score of the user in the consumption scene can be determined according to the scene amount ranking information, the scene consumption transaction count ranking information, the individual amount ranking information, and the individual consumption transaction count ranking information. Specifically, the scene preference score of the user can be calculated according to the scene amount ranking information, the scene consumption transaction count ranking information, and the predefined amount importance coefficient and transaction count importance coefficient. And the individual preference score of the user can be calculated according to the individual amount ranking information, the individual consumption transaction count ranking information, and the predefined amount importance coefficient and transaction count importance coefficient. Then, the consumption scene preference score of the user in the consumption scene can be calculated according to the scene preference score and the individual preference score. For this implementation process, it can be combined with Figure 3 to be described in detail as follows.

[0054] Refer to Figure 3 , which shows the flowchart of steps of another method for determining the consumption scene preference score provided by an embodiment of the present application. As Figure 3 shown, the method for determining the consumption scene preference score may include: Step 301, Step 302, and Step 303.

[0055] Step 301: Determine the scene preference score of the user in the consumption scene according to the scene amount ranking information, the scene consumption transaction count ranking information, and the predefined amount importance coefficient and transaction count importance coefficient.

[0056] In this embodiment, the scene preference score is an index used to quantify the preference degree of the user in a specific consumption scene. It can be used to reflect the level of the user's consumption situation in a certain scene compared to the consumption situation of the overall users in that scene.

[0057] The amount importance coefficient can be used to measure the importance of the consumption amount in evaluating the user's scene preference, and this coefficient can be adjusted according to business requirements and user characteristics.

[0058] The transaction count importance coefficient can be used to measure the importance of the consumption transaction count in evaluating the user's scene preference. Similarly, this coefficient can be adjusted according to business requirements and user characteristics.

[0059] Wherein, the sum value of the amount importance coefficient and the transaction count importance coefficient is 1.

[0060] After obtaining the scene amount ranking information, the scene consumption transaction count ranking information, and the predefined amount importance coefficient and transaction count importance coefficient, the scene preference score of the user in the consumption scene can be calculated according to the scene amount ranking information, the scene consumption transaction count ranking information, the amount importance coefficient, and the transaction count importance coefficient.

[0061] The calculation method can be: Scenario preference score (compared to the entire customer group) = the ranking of the customer's consumption amount in this scenario among the consumption amounts of the entire customer group in this scenario (i.e., scenario amount ranking information) × amount importance coefficient + the ranking of the customer's consumption frequency in this scenario among the consumption frequencies of the entire customer group in this scenario (i.e., scenario consumption frequency ranking information) × frequency importance coefficient.

[0062] Step 302: Determine the personal preference score of the user in the consumption scenario according to the personal amount ranking information, the personal consumption frequency ranking information, the amount importance coefficient, and the frequency importance coefficient.

[0063] The personal preference score can be an indicator for measuring the preference degree of the user in a specific consumption scenario relative to all his consumption scenarios, which reflects the level of the user's consumption situation in a certain scenario among his own consumption scenarios.

[0064] After obtaining the personal amount ranking information, the personal consumption frequency ranking information, and the pre-set amount importance coefficient and frequency importance coefficient, the personal preference score of the user in the consumption scenario can be calculated according to the personal amount ranking information, the personal consumption frequency ranking information, the amount importance coefficient, and the frequency importance coefficient.

[0065] The calculation method can be: Personal preference score = the ranking of the customer's consumption amount in this scenario among the consumption amounts in all existing consumption scenarios of his own (i.e., personal amount ranking information) × amount importance coefficient + the ranking of the customer's consumption frequency in this scenario among the consumption frequencies in all existing consumption scenarios of his own (i.e., personal consumption frequency ranking information) × frequency importance coefficient.

[0066] Step 303: Determine the consumption scenario preference score of the user in the consumption scenario based on the scenario preference score, the personal preference score, and the pre-set scenario dimension importance coefficient and personal dimension importance coefficient.

[0067] The scenario dimension importance coefficient is a pre-set coefficient, which can be used to measure the importance of the scenario preference score in evaluating the user's consumption scenario preference. The coefficient value can be adjusted according to business requirements, market trends, and user characteristics.

[0068] The personal dimension importance coefficient is also a pre-set coefficient, which can be used to measure the importance of the personal preference score in evaluating the user's consumption scenario preference. Similarly, the coefficient value can be adjusted according to factors such as business requirements, the user's historical behavior, and preferences.

[0069] Among them, the sum of the scenario dimension importance coefficient and the personal dimension importance coefficient is 1.

[0070] After obtaining the scene preference score and the personal preference score, the consumption scene preference score of the user in the consumption scene can be calculated according to the scene preference score, the personal preference score, and the pre-set scene dimension importance coefficient and personal dimension importance coefficient.

[0071] The calculation method can be: consumption scene preference score = scene preference score × scene dimension importance coefficient + personal preference score × personal dimension importance coefficient. The consumption scene with a higher consumption scene preference score is the consumption scene that the user prefers more.

[0072] By calculating the consumption scene preference score in the embodiments of the present application, the overall preference degree of the user in a specific consumption scene can be understood more accurately, so as to train the consumption scene preference prediction model more accurately.

[0073] After determining the consumption scene preference score of the user in the pre-defined consumption scene according to the historical multi-dimensional information, step 103 is executed.

[0074] Step 103: Screen out the target consumption scene with the largest consumption scene preference score from the consumption scenes.

[0075] After determining the consumption scene preference score of the user in the pre-defined consumption scene according to the historical multi-dimensional information, the target consumption scene with the largest consumption scene preference score can be screened out from the consumption scenes. For example, by calculating in the above manner, the consumption scene preference scores of the user in the scenes of offline dining, tea and dessert, takeout, shopping malls and department stores, e-commerce shopping, and life services are 2, 5, 3, 8, 7, and 10 respectively. At this time, the life service scene can be used as the target consumption scene, etc.

[0076] It can be understood that the above examples are only examples listed for better understanding of the technical solutions of the embodiments of the present application, and do not serve as the only limitation to this embodiment.

[0077] After screening out the target consumption scene with the largest consumption scene preference score from the consumption scenes, step 104 is executed.

[0078] Step 104: Construct a model training sample according to the historical multi-dimensional information and the target consumption scene.

[0079] After screening out the target consumption scene with the largest consumption scene preference score from the consumption scenes, a model training sample can be constructed according to the historical multi-dimensional information and the target consumption scene. Specifically, user basic attribute information, card usage behavior information, transaction information, etc. can be extracted from the historical multi-dimensional information, and the target consumption scene is used as the training label to jointly form a model training sample for model training.

[0080] After constructing the model training samples based on historical multi-dimensional information and the target consumption scenario, step 105 is executed.

[0081] Step 105: Train a consumption scenario preference prediction model for predicting consumption scenario preferences based on the model training samples.

[0082] After constructing the model training samples based on historical multi-dimensional information and the target consumption scenario, a consumption scenario preference prediction model for predicting consumption scenario preferences can be trained based on the model training samples.

[0083] Specifically, after obtaining the model training samples, stratified sampling can be performed for customer groups according to the pre-set model training sample sampling method, and stratified sampling can be carried out based on dimensions such as the length of the user's account age and whether there is consumption, so as to improve the learning and prediction capabilities of the model for various customer groups. By comparing various machine learning algorithms, the Xgboost multi-classification algorithm is finally selected for modeling. Through the integrated learning technology and combined with the gradient boosting framework, the complex relationship between the user's various dimensions of information and the future consumption scenario is fully fitted, and the consumption scenario preference ranking of the user in the future predetermined time period is scientifically and efficiently predicted. The model has good generalization ability and stability.

[0084] In practical applications, stratification can be performed based on two dimensions: the length of the user's account age (i.e., the length of time the user has become a customer) and whether there is a consumption record. The length of the account age can reflect the loyalty and stability of the user, while the presence or absence of a consumption record can reflect the activity and purchase intention of the user. According to the length of the account age, users are divided into different levels such as new customers and old customers. Within each account age level, further subdivision into active users and inactive users is carried out according to whether there is a consumption record, so that multiple sub-customer groups can be formed, and each sub-customer group has unique characteristics and behavior patterns.

[0085] In each sub-customer group, samples are randomly selected according to a certain proportion to ensure that each sub-customer group can be fully represented. The sampling proportion can be adjusted according to the size and importance of the sub-customer group to balance the diversity and representativeness of the samples.

[0086] Xgboost is an integrated learning algorithm based on gradient boosting. It gradually approaches the optimal solution by training multiple weak classifiers through multiple iterations. It supports a variety of loss functions and regularization terms and can handle high-dimensional data, missing data, and imbalanced data.

[0087] The model training process can include the following steps:

[0088] 1. Data preparation: According to the sample set obtained by stratified sampling, extract the user's basic attribute information, behavior information, transaction information, etc. as input features. Use the target consumption scenario as the output label.

[0089] 2. Data preprocessing: Preprocess the input features, such as standardization, normalization, encoding, etc., to improve the training efficiency and prediction accuracy of the model. According to business requirements and algorithm characteristics, select and construct new features to capture changes and trends in user behavior patterns.

[0090] 3. Model training: Use the Xgboost multi-classification algorithm to train the sample set to obtain an initial model. During the training process, optimize the performance of the model by adjusting algorithm parameters (such as learning rate, maximum number of iterations, number and depth of trees, etc.). Methods such as cross-validation can be used to evaluate the stability and generalization ability of the model, and the model can be tuned according to the evaluation results.

[0091] 4. Model evaluation: Use the test set to evaluate the trained model, and calculate metrics such as accuracy, recall rate, and F1 score. By comparing the evaluation results of different models, select the model with the best performance as the final model.

[0092] 5. Model deployment: Deploy the final model to the production environment for real-time prediction of users' consumption scenario preferences in a future predetermined time period. The model can be updated and optimized regularly according to business requirements to maintain its accuracy and stability.

[0093] By combining customer group stratified sampling and the Xgboost multi-classification algorithm in the embodiments of this application, a prediction model with good generalization ability and stability can be constructed. This model can fully learn and capture the behavior patterns and feature differences of users in different customer groups, so as to accurately predict users' consumption scenario preferences in a future predetermined time period.

[0094] After training the consumption scenario preference prediction model, the model can be used to predict users' consumption scenario preferences and perform consumption product matching. The process of model prediction and consumption product matching can be combined Figure 4 and described in detail as follows.

[0095] Refer to Figure 4 , which shows the step flowchart of a consumption product matching method provided by the embodiments of this application. As Figure 4 shown, the consumption product matching method may include: Step 401, Step 402, and Step 403.

[0096] Step 401: Obtain the historical multi-dimensional data of the target user within a historical time period.

[0097] In this embodiment, when predicting the consumption scenario preferences of a target user in a future period, historical multi-dimensional data of the target user within a historical time can be obtained. The historical multi-dimensional data can be sourced from the enterprise's transaction system, log system, etc. Such data can include multi-dimensional data such as the user's basic attributes, card usage behavior, purchase records, etc.

[0098] After obtaining the historical multi-dimensional data of the target user within a historical time, step 402 is executed.

[0099] Step 402: Input the historical multi-dimensional data into the consumption scenario preference prediction model, and obtain the consumption preference probabilities of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model.

[0100] After obtaining the historical multi-dimensional data of the target user within a historical time, the historical multi-dimensional data can be input into the consumption scenario preference prediction model, so that the consumption scenario preference prediction model processes the historical multi-dimensional data and outputs the consumption preference probabilities of the target user in multiple consumption scenarios.

[0101] Specifically, the historical multi-dimensional data of the target user can be preprocessed first, such as cleaning and organizing the collected data, removing duplicate, incorrect, or invalid data, and using the preprocessed historical multi-dimensional data as input features and inputting them into the consumption scenario preference prediction model. The input features can include the user's purchase history, consumption time, consumption amount, etc. The model will calculate the consumption preference probabilities of the target user in multiple consumption scenarios based on the input features. These probability values represent the purchase willingness or preference degree of the target user in different consumption scenarios. The consumption preference probabilities output by the model are as shown in Table 1 below:

[0102] Table 1 Consumption Preference Probabilities

[0103]

[0104] As shown in Table 1 above, User A and User B are the target users. The consumption preference probabilities output by the model for User A are: Hypermarkets and Department Stores: 20.1%, Urban Transportation: 19.1%, Business and Travel: 13.2%, Overseas Consumption: 9.5%, Health Care: 4.2%. The consumption preference probabilities output by the model for User B are: Offline Dining: 31.5%, E-commerce Shopping: 12.4%, Recharge and Payment: 8.9%, Food Delivery: 5.7%, Automobile Consumption: 5.3%, etc.

[0105] It can be understood that the above examples are only examples listed for better understanding the technical solutions of the embodiments of the present application and do not serve as the sole limitation of this embodiment.

[0106] After obtaining the consumption preference probabilities of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model, step 403 is executed.

[0107] Step 403: Perform consumption product matching for the target user according to the consumption preference probabilities.

[0108] After obtaining the consumption preference probabilities of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model, consumption product matching can be performed for the target user according to the consumption preference probabilities. Specifically, multiple consumption scenarios can be sorted in descending order of consumption preference probability to obtain a sorting result. The top N consumption scenarios are screened out from the consumption scenarios according to the sorting result, where N is a positive integer greater than or equal to 1. Consumption products in the top N consumption scenarios are matched for the target user. The matching method of consumption products can be:

[0109] 1. Obtain the consumption products in each of the top N consumption scenarios to provide to the target user. That is, obtain the top N consumption scenarios from the consumption scenario preference prediction model, and for each consumption scenario, screen out the consumption products related to this scenario from the product library. Then, the screened consumption products can be presented to the target user in the form of a list or a grid. Of course, brief descriptions, prices, evaluations, and other information can be attached to each product to help the user make a choice.

[0110] 2. Match scenario combination consumption products for the target user according to the top N consumption scenarios. That is, for each scenario combination, screen out the consumption products related to this combination from the product library, and present the screened scenario combination consumption products to the target user in the form of a package or a gift package. The preferential nature, convenience, or uniqueness of the combination products can be emphasized during matching to attract the user's attention. For example, for a customer group with multiple scenario preferences, when the top 2 scenarios preferred by the customer are dining and overseas, match similar combined benefit packages such as overseas restaurant coupons for them.

[0111] Through the above two matching strategies in the embodiments of the present application, personalized consumption product configuration for the target user can be achieved, improving the user's purchase intention and satisfaction, and at the same time promoting the sales and brand development of the enterprise.

[0112] After obtaining the consumption preference probabilities of the target user in multiple consumption scenarios, the user groups sensitive and insensitive to rights and interests preferences can be divided according to whether the difference in scenario preference probability values reaches a set threshold (overall difference quantile). Specifically, the multiple consumption scenarios can be sorted in descending order of consumption preference probability to obtain a sorting result. According to the sorting result, the preference probability difference between two adjacent consumption scenarios among the multiple consumption scenarios is determined. The target user is divided into consumption preference user groups according to the preference probability difference. For example, taking Table 1 above as an example, calculate the difference between the consumption preference probabilities of all consumption scenarios of User A pairwise. For example, the difference between the supermarket and entertainment is 20.1% - 19.1% = 1.0%. Repeat the calculation to obtain all possible differences. Similarly, calculate the difference between the consumption preference probabilities of all consumption scenarios of User B pairwise. For each user, if the maximum difference (or a certain key difference, such as the difference between the maximum and the minimum) of its internal scenario preference probabilities is greater than or equal to the threshold, it is divided into the user group sensitive to rights and interests preferences; otherwise, it is divided into the user group insensitive to rights and interests preferences. Assume that the median of the overall difference has been determined to be 3.0% through calculation (this value needs to be calculated based on actual data and is only an example here). For User A: the difference between the supermarket and entertainment is 1.0%, the difference between the supermarket and business travel is 6.9%,..., the maximum difference is 6.9%, which is less than twice the threshold of 3.0% (considering a certain buffer), so User A belongs to the user group insensitive to rights and interests preferences (other factors, such as the distribution of differences, also need to be considered specifically). For User B: the difference between catering and e-commerce is 19.1%, the difference between catering and payment is 22.6%,..., the maximum difference is 22.6%, which is much greater than the threshold of 3.0%, so User B belongs to the user group sensitive to rights and interests preferences, etc.

[0113] It can be understood that the above examples are only examples listed for better understanding of the technical solutions of the embodiments of the present application and do not serve as the sole limitation of this embodiment.

[0114] The embodiments of the present application divide the user groups sensitive and insensitive to rights and interests preferences according to the difference in scenario preference probability values, which helps the enterprise achieve beneficial effects in many aspects such as precision marketing, personalized service, resource optimization, risk management, enhancing user value, and market insight. This helps the enterprise maintain a competitive advantage in the highly competitive market environment and achieve sustainable development.

[0115] Meanwhile, the model results can match multiple strategic solutions. In actual business applications, consumption scenarios can be mapped to various business means such as gifts, marketing campaigns, and credit card products. The application is extensive and flexible. This mapping relationship not only improves business efficiency but also enhances the customer experience. It can also be applied to various customer groups, covering all nodes in the customer's entire life cycle (such as from new customer acquisition, old customer maintenance to customer churn recovery, etc.), and can be applied to customers with / without consumption data, high / low account age customers. Among them, customers without consumption data refer to those who have no credit card consumption behavior during the observation period and have credit card consumption behavior during the performance period. Both the observation period and the performance period can be predefined time periods.

[0116] It can be understood that the above examples are only examples listed for better understanding the technical solutions of the embodiments of the present application and do not serve as the sole limitation of this embodiment.

[0117] In this implementation, the consumption scenarios of users can be predicted so that they can match multiple business means and be applied to various marketing scenarios. The implementation process can be described in detail in combination with the following specific implementation methods.

[0118] In a specific implementation of the present application, after the above step 105, the following may further be included:

[0119] Step S1: Obtain the historical multi-dimensional data of the target user within the historical time.

[0120] In this embodiment, when predicting the consumption scenario preferences of the target user in the future period, the historical multi-dimensional data of the target user within the historical time can be obtained. The historical multi-dimensional data can be sourced from the enterprise's transaction system, log system, etc. These data can include multi-dimensional data such as the user's basic attributes, card usage behavior, and purchase records.

[0121] After obtaining the historical multi-dimensional data of the target user within the historical time, execute step S2.

[0122] Step S2: Input the historical multi-dimensional data into the consumption scenario preference prediction model and obtain the consumption scenario preferences of the target user output by the consumption scenario preference prediction model.

[0123] After obtaining the historical multi-dimensional data of the target user within the historical time, the historical multi-dimensional data can be input into the consumption scenario preference prediction model so that the consumption scenario preference prediction model processes the historical multi-dimensional data and outputs the consumption scenario preferences of the target user, such as overseas consumption, health care, etc.

[0124] After obtaining the consumption scenario preferences of the target user, execute step S3.

[0125] Step S3: Obtain at least one marketing scenario that matches the consumption scenario preference.

[0126] After obtaining the consumption scenario preference of the target user, at least one marketing scenario that matches the consumption scenario preference can be obtained. Among them, the marketing scenario can include: full reduction consumption scenario, coupon giving scenario, etc.

[0127] It can be understood that the marketing scenario that matches the consumption scenario preference can be one marketing scenario or multiple marketing scenarios, and this embodiment does not limit this.

[0128] After obtaining at least one marketing scenario that matches the consumption scenario preference, perform Step S4.

[0129] Step S4: Push the marketing products corresponding to the at least one marketing scenario to the target user.

[0130] After obtaining at least one marketing scenario that matches the consumption scenario preference, at least one marketing product corresponding to the marketing scenario can be pushed to the target user. For example, if the matching marketing scenario is a full reduction activity for credit card payments in the catering and e-commerce categories, etc., products for the full reduction activity can be pushed to the target user. For the marketing scenario of rights and interests coupons, marketing products such as car fuel coupons and Meituan Takeaway coupons can be pushed to the target user. For the marketing scenario of credit cards, marketing products such as business travel cards, Sinopec cards, and Sam's Club cards can be pushed to the target user, etc.

[0131] It can be understood that the above examples are only examples listed for better understanding the technical solutions of the embodiments of the present application and do not serve as the sole limitation of this embodiment.

[0132] The embodiments of the present application have scientifically and clearly defined the "consumption scenario preference" of customers, and sorted out and summarized multiple mainstream user consumption scenarios in the credit card industry. Because current credit card marketing includes various business operation means, such as distributing gifts, pushing marketing activities, configuring debit and credit card products, etc., this embodiment extracts a generalization method - namely, the "consumption scenario", to predict the user's consumption scenario so that it can match various business operation means and be applied to various marketing scenarios.

[0133] In this embodiment, for users without historical consumption behavior, a consumption scenario preference prediction model can be used to process the user characteristic information of this user, predict the user's consumption preference, and perform matching of consumption products. The implementation process for this can be described in detail in combination with the following specific implementation methods.

[0134] In another specific implementation of the present application, after the above step 105, it may further include:

[0135] Step N1: Obtain the user characteristic information of users without historical consumption behavior.

[0136] In this embodiment, for users without historical consumption behavior, the user characteristic information of users without historical consumption behavior can be obtained. The user characteristic information may include: information such as the user's age, gender, city where they are located, and type of credit card handled.

[0137] After obtaining the user characteristic information of users without historical consumption behavior, step N2 is executed.

[0138] Step N2: Input the user characteristic information into the consumption scenario preference prediction model, and obtain the consumption preference probabilities of the users without historical consumption behavior output by the consumption scenario preference prediction model in multiple consumption scenarios.

[0139] After obtaining the user characteristic information of users without historical consumption behavior, the user characteristic information can be input into the consumption scenario preference prediction model, and the consumption preference probabilities of the users without historical consumption behavior output by the consumption scenario preference prediction model in multiple consumption scenarios can be obtained. The consumption preference probability can represent the purchase willingness or preference degree of users without historical consumption behavior in different consumption scenarios.

[0140] After obtaining the consumption preference probabilities of the users without historical consumption behavior in multiple consumption scenarios, step N3 is executed.

[0141] Step N3: Perform consumption product matching for the users without historical consumption behavior according to the consumption preference probabilities.

[0142] After obtaining the consumption preference probabilities of the users without historical consumption behavior in multiple consumption scenarios, consumption product matching can be performed for the users without historical consumption behavior according to the consumption preference probabilities. Specifically, multiple consumption scenarios can be sorted in descending order of the consumption preference probabilities to obtain a sorting result. The top N (N is a positive integer) consumption scenarios are screened out from the consumption scenarios according to the sorting result, where N is a positive integer greater than or equal to 1. Consumption products in the top N consumption scenarios are matched for the users without historical consumption behavior.

[0143] Next, in combination with Figure 5 The consumption scenario preference prediction and product matching process is described as follows.

[0144] As Figure 5As shown in the figure, the process may include: 1. Consumption scenario classification, that is, through unified classification and inventory of various marketing right types and card products in the industry, and through clustering algorithms, a total of 18 mainstream user consumption scenarios in the credit card industry are sorted out and summarized. The specific categories include: offline dining, tea and desserts, takeaways, shopping malls and department stores, e-commerce shopping, life services, health care, recharge and payment, urban transportation, business travel, automobile consumption, overseas consumption, etc.

[0145] 2. Definition of consumption scenario preferences: When defining users' preferences for various consumption scenarios, this embodiment comprehensively considers multi-dimensional information such as scenario dimension ranking, customer dimension, consumption amount, and consumption frequency, and calculates users' consumption scenario preferences through weighted combination. The specific calculation method can refer to the calculation of consumption scenario preference scores in the above steps, and this embodiment will not elaborate here.

[0146] 3. Data acquisition: The independent variables of the model are to obtain users' basic attribute information and various card-using behavior information during the observation period based on historical data, and extract various features of users from them. The features included are as follows: basic attributes, card-using behavior, transaction information, etc.

[0147] 4. Preference calculation: The target variable of the model is the preference scores of each consumption scenario calculated above, and the consumption scenario with the largest consumption scenario preference score is selected as the target variable.

[0148] 5. Model training: The model sample sampling method is population stratified sampling, stratified sampling based on dimensions such as users' account ages and whether there is consumption, so as to improve the learning and prediction capabilities of the model in various populations. By comparing various machine learning algorithms, the Xgboost multi-classification algorithm is finally selected for modeling. Through ensemble learning technology, combined with the gradient boosting framework, it fully fits the complex relationship between users' various dimensions of information and future consumption scenarios, and scientifically and efficiently predicts the preference ranking of users' consumption scenarios in the future predetermined time period. The model has good generalization ability and stability.

[0149] 6. Model prediction: The model output is the preference ranking of the predicted user in each scenario during the performance period.

[0150] 7. Rights and interests product matching: The model outputs the ranking of users' consumption preferences in each scenario, and the top N preferred scenarios can be flexibly selected for corresponding product and service matching. For example, when the marketing right is "taxi coupon", if the corresponding consumption scenario "urban transportation" is hit among the top three of the user's preferred consumption scenarios and the preference probability value is greater than the set threshold, then "taxi coupon" is matched for it.

[0151] 8. Refined population operation: Derive more refined populations, and divide rights and interests preference-sensitive and insensitive populations according to whether the difference in scenario preference probability values reaches the set threshold (overall difference quantile).

[0152] In the embodiments of the present application, the concept of "consumption scenario" can match various marketing means, such as marketing activities, coupon rights and interests, card products, etc., so it can be applied to promoting customers at different business nodes. At the same time, through various customer group stratified sampling trainings, the model can cover the nodes in the entire customer life cycle and can be applied to high / low account age customers, etc.

[0153] The training method of the consumption scenario preference prediction model provided by the embodiments of the present application includes obtaining the historical multi-dimensional information of users. According to the historical multi-dimensional information, determine the consumption scenario preference score of the user under the pre-defined consumption scenario. Screen out the target consumption scenario with the largest consumption scenario preference score from the consumption scenarios. Construct a model training sample according to the historical multi-dimensional information and the target consumption scenario. Train a consumption scenario preference prediction model for consumption scenario preference prediction based on the model training sample. The embodiments of the present application predict the preference ranking of users' future consumption in various scenarios through user characteristics, combine the scenario preference ranking results, formulate differentiated business strategies, match corresponding rights and interests, products and services to users, improve marketing effects and user satisfaction, optimize operations and resource allocation to achieve precision marketing. At the same time, this solution can implement a complete personalized marketing business strategy plan, and its application can cover many key scenarios in credit card customer operation and marketing.

[0154] Refer to Figure 6 , which shows a schematic structural diagram of a training device for a consumption scenario preference prediction model provided by the embodiments of the present application, as Figure 6 shown, the training device 600 for the consumption scenario preference prediction model may include the following modules:

[0155] A historical multi-dimensional information acquisition module 610, configured to acquire the historical multi-dimensional information of users;

[0156] A consumption scenario preference determination module 620, configured to determine the consumption scenario preference score of the user under the pre-defined consumption scenario according to the historical multi-dimensional information;

[0157] A target consumption scenario screening module 630, configured to screen out the target consumption scenario with the largest consumption scenario preference score from the consumption scenarios;

[0158] A model training sample construction module 640, configured to construct a model training sample according to the historical multi-dimensional information and the target consumption scenario;

[0159] A scenario preference prediction model training module 650, configured to train a consumption scenario preference prediction model for consumption scenario preference prediction based on the model training sample.

[0160] Optionally, the consumption scenario preference determination module includes:

[0161] A consumption information determination unit, configured to determine, according to the historical multi-dimensional information, the scenario amount ranking information, the scenario consumption times ranking information, the personal amount ranking information, and the personal consumption times ranking information of the user in the consumption scenario; wherein, the scenario amount ranking information refers to the ranking of the consumption amount of the user in the consumption scenario among the consumption amounts of all users in the consumption scenario, the scenario consumption times ranking information refers to the ranking of the consumption times of the user in the consumption scenario among the consumption times of all users in the consumption scenario, the personal amount ranking information refers to the ranking of the consumption amount of the user in the consumption scenario among the consumption amounts in all existing consumption scenarios of the user, and the personal consumption times ranking information refers to the ranking of the consumption times of the user in the consumption scenario among the consumption times in all consumption scenarios of the user;

[0162] A consumption scenario preference determination unit, configured to determine the consumption scenario preference score of the user in the consumption scenario according to the scenario amount ranking information, the scenario consumption times ranking information, the personal amount ranking information, and the personal consumption times ranking information.

[0163] Optionally, the consumption scenario preference determination unit includes:

[0164] A scenario preference determination subunit, configured to determine the scenario preference score of the user in the consumption scenario according to the scenario amount ranking information, the scenario consumption times ranking information, and a preset amount importance coefficient and a preset times importance coefficient;

[0165] A personal preference determination subunit, configured to determine the personal preference score of the user in the consumption scenario according to the personal amount ranking information, the personal consumption times ranking information, the amount importance coefficient, and the times importance coefficient;

[0166] A consumption scenario preference determination subunit, configured to determine the consumption scenario preference score of the user in the consumption scenario based on the scenario preference score, the personal preference score, and a preset scenario dimension importance coefficient and a preset personal dimension importance coefficient.

[0167] Optionally, the device further includes:

[0168] A multi-dimensional data acquisition module, configured to acquire historical multi-dimensional data of a target user within a historical time period;

[0169] A preference probability acquisition module, configured to input the historical multi-dimensional data into the consumption scenario preference prediction model, and acquire the consumption preference probabilities of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model;

[0170] A consumer product matching module, configured to perform consumer product matching for the target user according to the consumer preference probability.

[0171] Optionally, the consumer product matching module includes:

[0172] A sorting result acquisition unit, configured to sort the multiple consumption scenarios in descending order according to the consumer preference probability to obtain a sorting result;

[0173] A consumption scenario screening unit, configured to screen out the top N consumption scenarios from the consumption scenarios according to the sorting result, where N is a positive integer greater than or equal to 1;

[0174] A consumer product matching unit, configured to match consumer products for the target user in the top N consumption scenarios.

[0175] Optionally, the consumer product matching unit includes:

[0176] A first matching subunit, configured to obtain the consumer products in each of the top N consumption scenarios to provide to the target user;

[0177] A second matching subunit, configured to match scenario combination consumer products for the target user according to the top N consumption scenarios.

[0178] Optionally, the apparatus further includes:

[0179] A sorting result acquisition module, configured to sort the multiple consumption scenarios in descending order according to the consumer preference probability to obtain a sorting result;

[0180] A probability difference determination module, configured to determine the preference probability difference between two adjacent consumption scenarios among the multiple consumption scenarios according to the sorting result;

[0181] A preference customer group division module, configured to perform consumer preference customer group division for the target user according to the preference probability difference.

[0182] Optionally, the apparatus further includes:

[0183] A historical data acquisition module, configured to acquire historical multi-dimensional data of the target user within a historical time;

[0184] A consumption scenario preference acquisition module, configured to input the historical multi-dimensional data into the consumption scenario preference prediction model and acquire the consumption scenario preference of the target user output by the consumption scenario preference prediction model;

[0185] A marketing scenario acquisition module, configured to acquire at least one marketing scenario that matches the consumption scenario preference;

[0186] A marketing product push module, configured to push marketing products corresponding to the at least one marketing scenario to the target user.

[0187] Optionally, the device further includes:

[0188] A user feature acquisition module, configured to acquire user feature information of a user without historical consumption behavior;

[0189] A preference probability acquisition module, configured to input the user feature information into the consumption scenario preference prediction model, and acquire the consumption preference probabilities of the user without historical consumption behavior in multiple consumption scenarios output by the consumption scenario preference prediction model;

[0190] A product matching module, configured to perform consumption product matching for the user without historical consumption behavior according to the consumption preference probabilities.

[0191] The training device of the consumption scenario preference prediction model provided by the embodiments of the present application acquires the historical multi-dimensional information of the user. According to the historical multi-dimensional information, determine the consumption scenario preference score of the user in the predefined consumption scenarios. Screen out the target consumption scenario with the largest consumption scenario preference score from the consumption scenarios. Construct a model training sample according to the historical multi-dimensional information and the target consumption scenario. Train a consumption scenario preference prediction model for consumption scenario preference prediction based on the model training sample. The embodiments of the present application predict the preference ranking of the user's future consumption in each scenario through user features, combine the scenario preference ranking results, formulate a differentiated operation strategy, configure corresponding rights and interests, products and services for the user, improve the marketing effect and user satisfaction, optimize the operation and resource allocation, so as to achieve precision marketing. At the same time, this solution can implement a complete personalized marketing operation strategy plan, and its application can cover many key scenarios of credit card customer operation and marketing.

[0192] The embodiments of the present application further provide an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the computer program is executed by the processor, the training method of the consumption scenario preference prediction model described above is implemented.

[0193] Figure 7 The structural schematic diagram of an electronic device 700 according to an embodiment of the present invention is shown. As Figure 7As shown, the electronic device 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) 702 or computer program instructions loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 can also be stored. The CPU 701, ROM 702, and RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0194] Multiple components in the electronic device 700 are connected to the I / O interface 705, including: an input unit 806, such as a keyboard, mouse, microphone, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, optical disk, etc.; and a communication unit 709, such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows the electronic device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0195] Each of the processes and processes described above can be executed by the processing unit 701. For example, the method of any of the above embodiments can be implemented as a computer software program, which is tangibly included in a computer-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the CPU 701, one or more actions in the method described above can be executed.

[0196] Additionally, the embodiments of the present application also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the training method of the above consumption scenario preference prediction model is implemented.

[0197] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0198] Those skilled in the art should understand that the embodiments of the embodiments of the present application can be provided as a method, apparatus, or computer program product. Therefore, the embodiments of the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0199] Embodiments of the present application are described with reference to the flowcharts and / or block diagrams of methods, terminals (systems), and computer program products according to embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminals to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing terminal generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0200] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0201] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal, such that a series of operation steps are executed on the computer or other programmable terminal to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable terminal provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0202] Although the preferred embodiments of the embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.

[0203] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or terminal. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or terminal including the said element.

[0204] The above has introduced in detail a training method for a consumption scenario preference prediction model, a training device for a consumption scenario preference prediction model, an electronic device and a computer-readable storage medium provided by the present application. Specific examples are used in this text to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for training a consumption scenario preference prediction model, characterized in that: The method comprises: Obtain historical multi-dimensional information of users; Determining a consumption scenario preference score of the user in a predefined consumption scenario based on the historical multi-dimensional information; Filtering out a target consumption scenario with the largest consumption scenario preference score from the consumption scenarios; Constructing a model training sample according to the historical multi-dimensional information and the target consumption scenario; A consumption scenario preference prediction model for consumption scenario preference prediction is obtained based on the model training sample training.

2. The method according to claim 1, characterized in that Determining the consumption scenario preference score of the user in a predefined consumption scenario according to the historical multi-dimensional information includes: According to the historical multi-dimensional information, the scene amount ranking information, scene consumption number ranking information, personal amount ranking information and personal consumption number ranking information of the user in the consumption scene are determined; wherein the scene amount ranking information refers to the ranking of the consumption amount of the user in the consumption scene among the consumption amounts of all users in the consumption scene, the scene consumption number ranking information refers to the ranking of the number of consumption transactions of the user in the consumption scene among the number of consumption transactions of all users in the consumption scene, the personal amount ranking information refers to the ranking of the consumption amount of the user in the consumption scene among all existing consumption scenes of the user, and the personal consumption number ranking information refers to the ranking of the number of consumption transactions of the user in the consumption scene among all consumption scenes of the user; The consumption scenario preference score of the user in the consumption scenario is determined based on the scenario amount ranking information, the scenario consumption number ranking information, the personal amount ranking information and the personal consumption number ranking information.

3. The method according to claim 2, characterized in that The determining, according to the scene amount ranking information, the scene consumption number ranking information, the personal amount ranking information, and the personal consumption number ranking information, the consumption scene preference score of the user in the consumption scene includes: Determine the scenario preference score of the user in the consumption scenario according to the scenario amount ranking information, the scenario consumption number ranking information, and the preset amount importance coefficient and number importance coefficient; Determine the user's personal preference score in the consumption scenario according to the personal amount ranking information, the personal consumption number ranking information, the amount importance coefficient and the number importance coefficient; Based on the scenario preference score, the personal preference score, and the preset scenario dimension importance coefficient and personal dimension importance coefficient, the consumption scenario preference score of the user in the consumption scenario is determined.

4. The method according to claim 1, characterized in that: After the consumption scenario preference prediction model for consumption scenario preference prediction is obtained based on the model training sample training, the method further includes: Obtain historical multi-dimensional data of target users over a historical period of time; Inputting the historical multi-dimensional data into the consumption scenario preference prediction model, and obtaining the consumption preference probability of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model; Consumer products are matched to the target user according to the consumption preference probability.

5. The method according to claim 4, characterized in that The matching of consumer products to the target user according to the consumption preference probability includes: Sorting the multiple consumption scenarios in descending order of the consumption preference probability to obtain a sorting result; Filtering out the top N consumption scenarios from the consumption scenarios according to the ranking results, where N is a positive integer greater than or equal to 1; The consumer products in the consumption scenarios ranked in the top N positions are provided to the target users.

6. The method according to claim 5, characterized in that The providing the target user with the consumer products in the consumption scenarios ranked in the top N places includes: Obtaining consumer products in each of the consumer scenarios ranked in the top N positions, and matching them to the target user; or According to the consumption scenarios ranked in the top N positions, scenario combination consumption products are matched to the target user.

7. The method according to claim 4, characterized in that After acquiring the consumption preference probabilities of the target user in multiple consumption scenarios output by the consumption scenario preference prediction model, the method further includes: Sorting the multiple consumption scenarios in descending order of the consumption preference probability to obtain a sorting result; Determining, according to the ranking result, a preference probability difference between two adjacent consumption scenarios in the multiple consumption scenarios; The target users are divided into consumer preference groups according to the preference probability differences.

8. The method according to claim 1, characterized in that After the consumption scenario preference prediction model for consumption scenario preference prediction is obtained based on the model training sample training, the method further includes: Obtain historical multi-dimensional data of target users over a historical period of time; Inputting the historical multi-dimensional data into the consumption scenario preference prediction model, and obtaining the consumption scenario preference of the target user output by the consumption scenario preference prediction model; Acquire at least one marketing scenario that matches the consumption scenario preference; Pushing a marketing product corresponding to the at least one marketing scenario to the target user.

9. The method according to claim 1, characterized in that: After the consumption scenario preference prediction model for consumption scenario preference prediction is obtained based on the model training sample training, the method further includes: Obtain user feature information of users with no historical consumption behavior; Inputting the user feature information into the consumption scenario preference prediction model, and obtaining the consumption preference probabilities of the user without historical consumption behavior in multiple consumption scenarios output by the consumption scenario preference prediction model; Consumer products are matched to the user without historical consumption behavior according to the consumption preference probability.

10. A training device for a consumption scenario preference prediction model, characterized in that: The device comprises: A historical multi-dimensional information acquisition module is used to acquire the user's historical multi-dimensional information; A consumption scenario preference determination module, used to determine the consumption scenario preference score of the user in a predefined consumption scenario based on the historical multi-dimensional information; A target consumption scenario screening module, used to screen out the target consumption scenario with the largest consumption scenario preference score from the consumption scenarios; A model training sample construction module, used to construct a model training sample according to the historical multi-dimensional information and the target consumption scenario; The scenario preference prediction model training module is used to train the consumption scenario preference prediction model for consumption scenario preference prediction based on the model training samples.

Images