Training of a prediction model, commodity recommendation method and device, electronic equipment and medium

Abstract

The present disclosure provides a training method of a prediction model, a commodity recommendation method and device, an electronic device and a medium, relates to the technical field of data processing, and particularly to the technical field of artificial intelligence, intelligent search and deep learning. The specific implementation scheme of the commodity recommendation method is as follows: determining a target commodity purchased within a fifth preset time period before a current time and a target user who purchases the target commodity; in response to a target lowest-level category to which the target commodity belongs being a repeat purchase category, extracting a target repeat purchase behavior feature of the target user repeating purchasing the target lowest-level category; inputting the target repeat purchase behavior feature into a trained prediction model to obtain a first repeat purchase probability of the target user; and based on the first repeat purchase probability, recommending a commodity under the target lowest-level category to the target user. By applying the technical scheme provided by the present disclosure, the accuracy and recall rate of the repeat purchase commodity recommendation for the user can be considered simultaneously.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and particularly relates to the technical field of artificial intelligence, intelligent search and deep learning. BACKGROUND

[0002] In e-commerce transactions, repeat purchase is an important concept. The more times a user repeats a purchase, the higher the user's recognition of the e-commerce platform, the higher the contribution rate of the e-commerce platform's commercial value, and the higher the user's stickiness to the e-commerce platform. SUMMARY

[0003] The present disclosure provides a training method of a prediction model, a commodity recommendation method and device, an electronic device and a medium.

[0004] According to a first aspect of the present disclosure, a training method of a prediction model is provided, and the method comprises:

[0005] determining a sample user of each lowest-level category;

[0006] extracting a repeat purchase behavior feature of a category corresponding to a repeat purchase of each sample user, the category corresponding to the sample user including a lowest-level category corresponding to a repeat purchase of the sample user and a superior category of the lowest-level category;

[0007] training a prediction model based on the repeat purchase behavior feature of each sample user.

[0008] In some embodiments, the step of determining a sample user of each lowest-level category comprises:

[0009] determining a user who purchases a commodity within a first preset time period after a first specified time as a candidate user;

[0010] determining a candidate user who purchases each lowest-level category within a second preset time period after a second specified time as a positive sample user of the lowest-level category, and determining a candidate user who does not purchase each lowest-level category within the second preset time period after the second specified time as a negative sample user of the lowest-level category, the second specified time being after the first specified time, and a time period between the second specified time and the first specified time being a third preset time.

[0011] In some embodiments, the step of determining a candidate user who does not purchase each lowest-level category within a second preset time period after a second specified time as a negative sample user of the lowest-level category comprises:

[0012] For each lowest-level category, determine, as first negative sample users of the lowest-level category, alternative users who have not purchased any lowest-level category within a second preset time period after the second specified time, and determine, as second negative sample users of the lowest-level category, alternative users who have not purchased the lowest-level category and have purchased other lowest-level categories within the second preset time period after the second specified time.

[0013] In some embodiments, the step of training the prediction model based on the repurchase behavior features of each sample user comprises:

[0014] Combining the positive sample users, the first negative sample users and the second negative sample users in various proportions to obtain a plurality of sample user sets;

[0015] Training the prediction model using the repurchase behavior features of the users in each sample user set;

[0016] From the plurality of trained prediction models, selecting a prediction model with the smallest loss as the prediction model for predicting the repurchase probability.

[0017] In some embodiments, the repurchase behavior features include recent behavior features of the user in the lowest-level category and recent behavior features of the user in the superior category of the lowest-level category.

[0018] In some embodiments, the recent behavior features include one or more of the number of clicks, the number of views, the number of collections and the number of comments in a plurality of preset time scales within a fourth preset time period before the third specified time.

[0019] In some embodiments, the repurchase behavior features further include at least one dimension feature: a repurchase attribute feature of the lowest-level category, a purchased time feature of the user in the lowest-level category, and a category feature of the lowest-level category.

[0020] In some embodiments, the repurchase attribute feature includes one or more of a user layer repurchase indicator, an order layer repurchase indicator and a comprehensive repurchase indicator.

[0021] The purchased time feature includes one or more of the time interval between the time of first purchase of a product under the lowest-level category and the third specified time, the time interval between the time of last purchase of a product under the lowest-level category and the third specified time, the average time interval of purchase of a product under the lowest-level category, and the number of times of purchase of a product under the lowest-level category within a fourth preset time period before the third specified time.

[0022] The category feature includes one or more of the number of products under the lowest-level category, the average price of products, and the click-through rate and conversion rate in a plurality of preset time scales.

[0023] In some embodiments, the user layer repeat purchase indicator is a proportion of a number of repeat purchase users of the lowest level category to a number of all purchase users.

[0024] The order layer repeat purchase indicator is a proportion of a number of repeat purchase times of the lowest level category to a number of all purchase times.

[0025] The comprehensive repeat purchase indicator is a product value of the user layer repeat purchase indicator and the order layer repeat purchase indicator, or the comprehensive repeat purchase indicator is a weighted average value of a plurality of repeat purchase indicators of the lowest level category.

[0026] According to a second aspect of the present disclosure, a commodity recommendation method is provided, comprising:

[0027] determining a target commodity purchased within a fifth preset time period before a current time and a target user who purchases the target commodity;

[0028] in response to a target lowest level category to which the target commodity belongs being a repeat purchase category, extracting a target repeat purchase behavior feature of the target user who repeats purchases the target lowest level category;

[0029] inputting the target repeat purchase behavior feature into a trained prediction model to obtain a first repeat purchase probability of the target user;

[0030] based on the first repeat purchase probability, recommending a commodity under the target lowest level category to the target user.

[0031] In some embodiments, the repeat purchase category is determined by the following steps:

[0032] determining a second repeat purchase probability of each lowest level category according to a repeat purchase indicator of each lowest level category;

[0033] determining a repeat purchase category according to the second repeat purchase probability of each lowest level category.

[0034] In some embodiments, the number of repeat purchase indicators is a plurality;

[0035] The step of determining a second repeat purchase probability of each lowest level category according to a repeat purchase indicator of each lowest level category comprises:

[0036] performing product processing on a plurality of repeat purchase indicators of each lowest level category to obtain a second repeat purchase probability of each lowest level category; or,

[0037] performing weighted average on a plurality of repeat purchase indicators of each lowest level category to obtain a second repeat purchase probability of each lowest level category.

[0038] In some embodiments, the repeat purchase indicator includes one or more of a user layer repeat purchase indicator and an order layer repeat purchase indicator.

[0039] In some embodiments, the user layer repeat purchase index is a proportion of the number of repeat purchase users of the lowest level category to the number of all purchase users.

[0040] The order layer repeat purchase index is a proportion of the number of repeat purchase times of the lowest level category to the number of all purchase times.

[0041] In some embodiments, the step of determining the repeat purchase category according to the second repeat purchase probability of each lowest level category comprises:

[0042] taking a preset number of lowest level categories with the highest second repeat purchase probability as the repeat purchase category; or

[0043] taking the lowest level category with a second repeat purchase probability greater than a preset probability as the repeat purchase category.

[0044] In some embodiments, the preset number is determined by the following steps:

[0045] counting the total number of lowest level categories; and taking a preset proportion of the total number as the preset number.

[0046] In some embodiments, the step of recommending the goods under the target lowest level category to the target user based on the first repeat purchase probability comprises:

[0047] establishing a recall channel of the target lowest level category for the target user in response to the first repeat purchase probability of the target lowest level category being greater than a preset threshold value;

[0048] integrating the recall channel into a recommendation system, the recommendation system being configured to recommend the goods under the target lowest level category to the target user.

[0049] According to a third aspect of the present disclosure, there is also provided a training device of a prediction model, the device comprising:

[0050] a first determination module configured to determine sample users of each lowest level category;

[0051] a first extraction module configured to extract repeat purchase behavior features of categories corresponding to repeat purchase of each sample user, the categories corresponding to the sample user including the lowest level category corresponding to repeat purchase of the sample user and the superior category of the lowest level category;

[0052] a training module configured to train a prediction model based on the repeat purchase behavior features of each sample user.

[0053] In some embodiments, the first determination module comprises:

[0054] The first determining sub-module is configured to determine users who purchase the goods within a first preset time period after a first specified time as candidate users;

[0055] The second determining sub-module is configured to determine candidate users who purchase each lowest-level category within a second preset time period after a second specified time as positive sample users of the lowest-level category, and determine candidate users who do not purchase each lowest-level category within the second preset time period after the second specified time as negative sample users of the lowest-level category, wherein the second specified time is after the first specified time, and a time period between the second specified time and the first specified time is a third preset time.

[0056] In some embodiments, the second determining sub-module is specifically configured to:

[0057] For each lowest-level category, the second determining sub-module is specifically configured to determine a candidate user who does not purchase any lowest-level category within a second preset time period after a second specified time as a first negative sample user of the lowest-level category, and determine a candidate user who does not purchase the lowest-level category and purchases other lowest-level categories within the second preset time period after the second specified time as a second negative sample user of the lowest-level category.

[0058] In some embodiments, the training module is specifically configured to:

[0059] The training module is specifically configured to combine the positive sample users, the first negative sample users and the second negative sample users in multiple proportions to obtain multiple sample user sets;

[0060] The training module is specifically configured to train the prediction model by using the repurchase behavior features of the users in each sample user set.

[0061] The training module is specifically configured to select a prediction model with the smallest loss from the multiple prediction models obtained by training as the prediction model for predicting the repurchase probability.

[0062] In some embodiments, the repurchase behavior features include recent behavior features of the user in the lowest-level category and recent behavior features of the user in a superior category of the lowest-level category.

[0063] In some embodiments, the recent behavior features include one or more of the following in multiple preset time scales within a fourth preset time period before a third specified time: click volume, browsing volume, collection volume and comment volume.

[0064] In some embodiments, the repurchase behavior features further include at least one dimensional feature: repurchase attribute features of the lowest-level category, purchased time features of the user in the lowest-level category, and category features of the lowest-level category.

[0065] In some embodiments, the repeat purchase attribute feature comprises one or more of a user layer repeat purchase indicator, an order layer repeat purchase indicator, and a comprehensive repeat purchase indicator.

[0066] The purchased time feature comprises one or more of a time length from a time of first purchase of a product under a lowest level category to the third specified time, a time length from a time of last purchase of a product under the lowest level category to the third specified time, an average time interval of purchase of a product under the lowest level category, and a number of times of purchase of a product under the lowest level category.

[0067] The category feature comprises one or more of a number of products under the lowest level category, a mean product price, a click rate, and a conversion rate in a plurality of preset time scales.

[0068] In some embodiments, the user layer repeat purchase indicator is a proportion of a number of repeat purchase users of the lowest level category to a total number of purchase users.

[0069] The order layer repeat purchase indicator is a proportion of a number of repeat purchase times of the lowest level category to a total number of purchase times.

[0070] The comprehensive repeat purchase indicator is a product of the user layer repeat purchase indicator and the order layer repeat purchase indicator, or the comprehensive repeat purchase indicator is a weighted average of a plurality of repeat purchase indicators of the lowest level category.

[0071] According to a fourth aspect of the present disclosure, a product recommendation device is also provided, comprising:

[0072] A second determination module is configured to determine a target product that is purchased within a fifth preset time length before a current time and a target user who purchases the target product.

[0073] A second extraction module is configured to, in response to a target lowest level category to which the target product belongs being a repeat purchase category, extract a target repeat purchase behavior feature of the target user who repeats purchase of the target lowest level category.

[0074] An input module is configured to input the target repeat purchase behavior feature into a trained prediction model to obtain a first repeat purchase probability of the target user.

[0075] A recommendation module is configured to recommend, based on the first repeat purchase probability, a product under the target lowest level category to the target user.

[0076] In some embodiments, the device further comprises a third determination module configured to determine the repeat purchase category, comprising:

[0077] A third determination sub-module is configured to determine, according to a repeat purchase indicator of each lowest level category, a second repeat purchase probability of each lowest level category.

[0078] a fourth determining sub-module, configured to determine a repurchase category according to a second repurchase probability of each lowest-level category.

[0079] In some embodiments, the number of the repurchase indicators is multiple;

[0080] The third determining sub-module is specifically configured to:

[0081] multiply the multiple repurchase indicators of each lowest-level category to obtain the second repurchase probability of each lowest-level category; or

[0082] weight-average the multiple repurchase indicators of each lowest-level category to obtain the second repurchase probability of each lowest-level category.

[0083] In some embodiments, the repurchase indicators include one or more of a user layer repurchase indicator and an order layer repurchase indicator.

[0084] In some embodiments, the user layer repurchase indicator is a proportion of repurchase times of the lowest-level category to all purchase times.

[0085] The order layer repurchase indicator is a proportion of repurchase times of the lowest-level category to all purchase times.

[0086] In some embodiments, the fourth determining sub-module is specifically configured to:

[0087] take a preset number of lowest-level categories with the highest second repurchase probability as the repurchase categories; or

[0088] take the lowest-level categories with a second repurchase probability greater than a preset probability as the repurchase categories.

[0089] In some embodiments, the fourth determining sub-module is further configured to:

[0090] count a total number of the lowest-level categories; and take a preset proportion of the total number as the preset number.

[0091] In some embodiments, the recommendation module is specifically configured to:

[0092] in response to the first repurchase probability of the target lowest-level category being greater than a preset threshold, establish a recall path of the target lowest-level category for the target user;

[0093] integrate the recall path into a recommendation system, the recommendation system being configured to recommend goods under the target lowest-level category to the target user.

[0094] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising:

[0095] at least one processor; and

[0096] a memory in communication with the at least one processor; wherein

[0097] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the methods provided in the first aspect or any of the methods provided in the second aspect.

[0098] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is further provided, wherein the computer instructions are used to enable the computer to perform any of the methods provided in the first aspect or any of the methods provided in the second aspect.

[0099] According to a seventh aspect of the present disclosure, a computer program product is further provided, comprising a computer program which, when executed by a processor, implements any of the methods provided in the first aspect or any of the methods provided in the second aspect.

[0100] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0101] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0102] Figure 1 The first flowchart of the training method of the prediction model provided by the embodiments of the present disclosure;

[0103] Figure 2 The classification system schematic diagram of a commodity category provided by the embodiments of the present disclosure;

[0104] Figure 3 The second flowchart of the training method of the prediction model provided by the embodiments of the present disclosure;

[0105] Figure 4 The positive and negative sample user selection schematic diagram provided by the embodiments of the present disclosure;

[0106] Figure 5 The third flowchart of the training method of the prediction model provided by the embodiments of the present disclosure;

[0107] Figure 6 The fourth flowchart of the training method of the prediction model provided by the embodiments of the present disclosure;

[0108] Figure 7 An example schematic diagram of a training method of a prediction model provided by an embodiment of the present disclosure;

[0109] Figure 8 A first flow schematic diagram of a commodity recommendation method provided by an embodiment of the present disclosure;

[0110] Figure 9 A flow schematic diagram of a repeat purchase category determination method provided by an embodiment of the present disclosure;

[0111] Figure 10 A second flow schematic diagram of a commodity recommendation method provided by an embodiment of the present disclosure;

[0112] Figure 11 A structure schematic diagram of a training device of a prediction model provided by an embodiment of the present disclosure;

[0113] Figure 12 A structure schematic diagram of a commodity recommendation device provided by an embodiment of the present disclosure;

[0114] Figure 13 A block diagram of an electronic device for implementing a training method of a prediction model or a commodity recommendation method provided by an embodiment of the present disclosure;

[0115] Figure 14 A structure schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0116] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to help in understanding, which should be considered in their context only. Thus, those of ordinary skill in the art should recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0117] Repeat purchase means the repeated purchase of a user on a certain commodity or service. In e-commerce transactions, repeat purchase is an important concept. The more a user repeats a purchase, the higher the user's recognition of the e-commerce platform, the higher the contribution rate of the e-commerce platform's business value, and the higher the user's stickiness of the e-commerce platform. In existing commodity recommendation scenarios, the methods of commodity recommendation mainly include the following two:

[0118] Method one, using a statistical scheme to mine the re-purchase category and re-purchase time. Specifically, based on a large number of user historical purchase of goods or goods belonging to the category and other historical purchase behaviors, a user re-purchase probability function is constructed, and the probability function is used to infer the user's purchase probability for each category; based on the historical user's re-purchase interval time distribution, the re-purchase time probability of each category is constructed; the re-purchase probability and the re-purchase time probability of the user to the category are multiplied to obtain the comprehensive re-purchase probability of the user to each category; and the user is recalled the goods under the category with high re-purchase probability.

[0119] This method is simple and easy to understand, but this method is based on the re-purchase behavior and re-purchase time of the group in the category, and does not consider the recent behavior of a single user. When solving the re-purchase problem, the recall rate is high, and the accuracy is insufficient.

[0120] Method two, using a model to predict the re-purchase probability of the user to the goods under the category. Specifically, a prediction model based on machine learning is constructed; the re-purchase features corresponding to each category are extracted from the historical purchase behavior of a single user, and the extracted re-purchase features are input into the prediction model to obtain the re-purchase probability of the single user to each category; and the single user is recalled the goods under the category with high re-purchase probability.

[0121] In this method, the re-purchase probability of the user is predicted by mining the recent behavior features of the user, the re-purchase behavior of the category, and the purchase time. The accuracy is high. However, when a user has no recent behavior for the purchased category, the prediction model cannot accurately predict the re-purchase probability, and the recall rate of the prediction model is obviously insufficient. For example, a user will purchase toilet paper on an e-commerce platform every two months, but in the last two months, the user has no click and browse behavior of toilet paper on the e-commerce platform, and has no click and browse behavior under the household cleaning category. At this time, the prediction model will not recommend toilet paper or goods under the household cleaning category to the user, but in fact, the user may have a purchase demand for toilet paper.

[0122] In order to improve the recall rate and accuracy of the goods, the embodiments of the present disclosure provide a training method of a prediction model, a re-purchase category determination method and a goods recommendation method. For ease of description, the following describes an electronic device as the execution subject, which is not limited. In the embodiments of the present disclosure, the electronic device executing the training method of the prediction model and the device executing the goods recommendation method can be the same or different. In the embodiments of the present disclosure, the prediction model is a machine learning model, such as an eXtreme Gradient Boosting (XGboost) model and a Deep Neural Networks (DNN) model.

[0123] As shown in Figure 1 The embodiments of the present disclosure provide a training method of a prediction model, including the following steps:

[0124] Step S11, determine the sample user of each lowest level category;

[0125] Step S12, extract the repurchase behavior features of the category corresponding to the repurchase of each sample user, the category corresponding to the sample user includes the lowest level category corresponding to the repurchase of the sample user and the superior category of the lowest level category;

[0126] Step S13, train the prediction model based on the repurchase behavior features of each sample user.

[0127] The training method of the prediction model provided by the embodiments of the present disclosure extracts the repurchase behavior features of the lowest level category corresponding to the repurchase of each sample user, and extracts the repurchase behavior features of the superior category of the lowest level category corresponding to the repurchase of each sample user, and comprehensively trains the repurchase behavior features of the lowest level category and the superior category to train the prediction model for predicting the repurchase probability. This expands the dimension of the repurchase behavior features, increases the repurchase behavior data of a single user for training the prediction model, and further trains the prediction model using larger repurchase behavior data of a single user to more accurately predict the repurchase probability. Based on the accurately predicted repurchase probability, the accuracy of the commodity recommendation is improved, and the recall rate of the commodity is also improved.

[0128] In the above step S11, the lowest level category is the smallest category to which the commodity belongs. In the embodiments of the present disclosure, the total commodity can be divided into several level categories according to the selected classification mark, and arranged into a hierarchical and gradually expanding classification system. For example, as shown in the category diagram, Figure 2 the total commodity is divided into four level categories, the first level category is the total category, and the total category includes daily necessities, food, clothing, furniture, household appliances, hardware and electrical materials, and several second level categories. Taking the daily necessities category as an example, the daily necessities category includes cleaning, beauty, protection, and several third level categories. Taking the cleaning category as an example, the cleaning category includes laundry, face washing, hair washing, and dish washing, and several fourth level categories. Figure 2 In the embodiments of the present disclosure, the rectangular box with "-" represents that the corresponding category includes the lower level category or the commodity has been expanded, such as Figure 2 the total category in Figure 2 the food category and the laundry category, etc. Taking Figure 2 as an example, if the laundry category no longer includes the lower level category, but includes a plurality of commodities, then the laundry category is the lowest level category.

[0129] In the embodiments of the present disclosure, when training the prediction model, the electronic device determines sample users of each lowest-level category, and the determined sample users include positive sample users and negative sample users. The positive sample user of one lowest-level category can be a user who has purchased a commodity included in the lowest-level category, and the negative sample user of one lowest-level category can be a user who has not purchased a commodity included in the lowest-level category.

[0130] In the step S12, the superior category of the lowest-level category is a category whose level is higher than that of the lowest-level category and which contains the lowest-level category. The superior category of the lowest-level category can be one or more. As shown in the table, the superior categories of the laundry category include the cleaning category in the third-level category, the daily necessities category in the second-level category, and the total category in the first-level category, and the like. The category corresponding to the repeat purchase represents one or more commodities included under the category corresponding to the repeat purchase. Figure 2

[0131] For each lowest-level category, the electronic device extracts the repeat purchase behavior feature of each sample user who repeats purchases of the lowest-level category, and the repeat purchase behavior feature of each superior category of the lowest-level category.

[0132] In the embodiments of the present disclosure, the repeat purchase behavior feature can include the recent behavior feature of the user in the lowest-level category and the recent behavior feature of the user in the superior category of the lowest-level category. The recent behavior feature can include one or more of the click volume, the browse volume, the collection volume, and the comment volume of multiple preset time scales within a fourth preset time length before a third specified time.

[0133] The third specified time can be set according to actual needs. For example, the third specified time can be the current time when training the prediction model, or can be a historical time before training the prediction model, which is not limited. The fourth preset time length can be set according to actual needs. For example, the fourth preset time length can be 6 months, 1 year, or 2 years, and the like. The preset time scale can be set according to actual needs. For example, the preset time scale can be 1 day, 7 days, 1 month, 3 months, 6 months, and 12 months, and the like.

[0134] For example, the recent behavior feature of the user in the lowest-level category can include the click volume, the browse volume, the collection volume, and the comment volume of the user in the lowest-level category within the fourth preset time length before the third specified time. Figure 2 ​Taking the laundry category as an example, the third specified moment is the current moment, the fourth preset time length is 1 year, the preset time scale can include 7 days, 1 month, 3 months, 6 months and 12 months, and the recent behavior characteristics include click volume, browse volume, collection volume and comments. The electronic device can extract, from 1 year before the current moment, the click volume, browse volume, collection volume and comment volume of the sample user in the laundry category within 7 days, 1 month, 3 months, 6 months and 12 months, extract the click volume, browse volume, collection volume and comment volume of the sample user in the cleaning category (i.e., the upper level category of the laundry category) within 7 days, 1 month, 3 months, 6 months and 12 months, extract the click volume, browse volume, collection volume and comment volume of the sample user in the daily necessities category (i.e., the upper two-level category of the laundry category, the upper level category of the cleaning category) within 7 days, 1 month, 3 months, 6 months and 12 months, and extract the click volume, browse volume, collection volume and comment volume of the sample user in the total category (i.e., the upper three-level category of the laundry category, the upper two-level category of the cleaning category, the upper level category of the daily necessities category) within 7 days, 1 month, 3 months, 6 months and 12 months.

[0135] In the embodiments of the present disclosure, the electronic device extracts the recent behavior characteristics of the user in the multiple preset time scales of the lowest level category, and extracts the recent behavior characteristics of the user in the multiple preset time scales of the related category of the lowest level category, and then the prediction model can be trained by using the recent behavior characteristics of the user in the lowest level category and the related category. In this way, even if the recent behavior characteristics of the user in the lowest level category are less, the training data of the prediction model can be increased by means of the recent behavior characteristics of the user in the related category, and the accuracy of the prediction model in predicting the repurchase probability is further improved.

[0136] In addition, in the embodiments of the present disclosure, the recent behavior characteristics are subdivided into characteristics in multiple preset time scales, which further increases the data amount of the characteristics, and further improves the accuracy of the prediction model in predicting the repurchase probability.

[0137] In some embodiments, the repurchase behavior characteristics can further include at least one dimension characteristic: repurchase attribute characteristics of the lowest level category, purchase time characteristics of the user in the lowest level category, and category characteristics of the lowest level category.

[0138] 1) The repurchase attribute characteristics can include one or more of user layer repurchase indicators, order layer repurchase indicators and comprehensive repurchase indicators.

[0139] a, the user layer repurchase indicator can be the proportion of the number of repurchase users in the lowest level category to the number of all purchase users. The user layer repurchase indicator of one lowest level category can be expressed by the following formula (1):

[0140] p repeat-uv =u1 / u2 (1)

[0141] In formula (1), p repeat-uv represents the user layer repurchase index of the lowest level category, and can also be understood as the user layer repurchase probability of the lowest level category, u1represents the repurchase user quantity of the lowest level category, that is, the quantity of users who purchase the goods under the lowest level category more than or equal to 2 times, and u2represents the total purchase user quantity of the lowest level category, that is, the total user quantity who purchases the goods under the lowest level category.

[0142] For example, the lowest level category 1 includes goods a, b and c. User 1 purchases the goods a once and the goods b once; user 2 purchases the goods a once; user 3 purchases the goods a once, the goods b once and the goods c three times. At this time, the repurchase users of the lowest level category 1 include user 1 and user 3, that is, u1=2, the total purchase users of the lowest level category 1 include user 1, user 2 and user 3, that is, u2=3, and at this time, the user layer repurchase index p repeat-uv of the lowest level category 1 is 2 / 3.

[0143] In the embodiments of the present disclosure, the user layer repurchase index can be expressed in other forms. For example, the user layer repurchase index can be the proportion of the repurchase user quantity of the lowest level category to the total purchase user quantity of all the lowest level categories, and no limitation is made as long as the user layer repurchase index is determined from the user perspective.

[0144] b, the order layer repurchase index can be the proportion of the repurchase times of the lowest level category to the total purchase times. The order layer repurchase index of one lowest level category can be expressed by the following formula (2):

[0145] p repeat-pv = c1 / c2 (2)

[0146] In formula (2), p repeat-pv represents the order layer repurchase index of the lowest level category, and can also be understood as the order layer repurchase probability of the lowest level category, c1represents the repurchase times of the lowest level category, that is, the non-first order times of the goods under the lowest level category, and c2represents the total purchase times of the lowest level category, that is, the total order times of the goods under the lowest level category.

[0147] For example, the lowest level category 1 includes commodities a, b, c. User 1 has purchased commodity a once and commodity b once; user 2 has purchased commodity a once; user 3 has purchased commodity a once, commodity b once, and commodity c three times. At this time, the repurchase times of the lowest level category 1 include 1 repurchase of user 1 and 2 repurchases of user 3, that is, c1=1+2=3, the total purchase times of the lowest level category 1 include 2 repurchases of user 1, 1 repurchase of user 2, and 3 repurchases of user 3, that is, c2=2+1+3=6, at this time, the order layer repurchase index p repeat-pv of the lowest level category 1 is 3 / 6.

[0148] In the embodiments of the present disclosure, the order layer repurchase index can be expressed in other forms, for example, the order layer repurchase index can be the proportion of the repurchase times of the lowest level category to the total purchase times of all the lowest level categories, and this is not limited, as long as the order layer repurchase index is determined from the order angle.

[0149] c, the comprehensive repurchase index is a repurchase index obtained by the comprehensive user layer repurchase index and the order layer repurchase index.

[0150] In the embodiments of the present disclosure, the comprehensive repurchase index can be the product value of the user layer repurchase index and the order layer repurchase index. At this time, the comprehensive repurchase index can be expressed by the following formula (3):

[0151] p repeat = p repeat-uv * p repeat-pv (3)

[0152] In formula (3), p repeat represents the comprehensive repurchase index of the lowest level category, p repeat-uv represents the user layer repurchase index of the lowest level category, and p repeat-pv represents the order layer repurchase index of the lowest level category.

[0153] The comprehensive repurchase index can also be the weighted average of multiple repurchase indexes of the lowest level category. At this time, the comprehensive repurchase index can be expressed by the following formula (4):

[0154] p repeat = (w1p repeat-uv +w2p repeat-pv ) / (w1+w2) (4)

[0155] In formula (3), p repeat represents the comprehensive repurchase index of the lowest level category, p repeat-uv represents the user layer repurchase index of the lowest level category, p repeat-pv represents the order layer repurchase index of the lowest level category, and w1 represents prepeat-uv the weight of p repeat-pv the weight of p

[0156] In the embodiments of the present disclosure, the repeat purchase attribute features are determined from multiple perspectives, the diversity of the repeat purchase attribute features is improved, and the accuracy of the prediction model trained based on the repeat purchase attribute features is further improved.

[0157] 2) The purchased time features can include one or more of a time length from a time of first purchase of a product under the lowest-level category to the third specified time, a time length from a time of last purchase of a product under the lowest-level category to the third specified time, an average time interval of purchase of a product under the lowest-level category, and a number of times of purchase of a product under the lowest-level category.

[0158] The purchased time features are features obtained by statistics within the fourth preset time length before the third specified time.

[0159] For example, for each sample user and each lowest-level category, the electronic device extracts the purchased time features such as a time length from a time of first purchase of a product under the lowest-level category to the third specified time, a time length from a time of last purchase of a product under the lowest-level category to the third specified time, an average time interval of purchase of a product under the lowest-level category, and a number of times of purchase of a product under the lowest-level category within the fourth preset time length before the third specified time.

[0160] In the embodiments of the present disclosure, the repeat purchase behavior features for training the prediction model increase the time factor, i.e., the purchased time features, which further improves the diversity of the repeat purchase attribute features, and further improves the accuracy of the prediction model trained based on the repeat purchase attribute features.

[0161] 3) The category features include one or more of a number of products under the lowest-level category, a product price average, and a click rate and a conversion rate within a plurality of preset time scales.

[0162] The click rate is a ratio of a number of clicks of the products under the lowest-level category to a number of views of the products under the lowest-level category, and the conversion rate is a ratio of a number of purchases of the products under the lowest-level category to a number of views of the products under the lowest-level category.

[0163] In the embodiments of the present disclosure, the repeat purchase behavior features for training the prediction model increase the features of the category itself, which further improves the diversity of the repeat purchase attribute features, and further improves the accuracy of the prediction model trained based on the repeat purchase attribute features.

[0164] In the embodiments of the present disclosure, the electronic device describes the repeat purchase behavior through four-dimensional features, i.e., a recent behavior feature, a repeat purchase attribute feature of a lowest-level category, a purchase time feature of a user in the lowest-level category, and a category feature of the lowest-level category, so that the features of the training prediction model are more suitable for the repeat purchase scenario, and the accuracy of the prediction model trained based on the repeat purchase attribute feature is improved.

[0165] In the step S13, the electronic device inputs the repeat purchase behavior feature of each sample user into the prediction model to train the prediction model.

[0166] For example, the sample users are divided into positive sample users and negative sample users. The electronic device inputs the repeat purchase behavior feature of each sample user into the prediction model, the prediction model outputs the repeat purchase probability of each sample user, the parameters of the prediction model are adjusted, the repeat purchase behavior feature of each sample user is re-input into the prediction model, and iterative training is performed, so that the repeat purchase probability of the positive sample user is close to 1 and the repeat purchase probability of the negative sample user is close to 0. The iterative training is performed until a training end condition is reached.

[0167] In some embodiments, as shown in Figure 3 The embodiments of the present disclosure also provide a training method of a prediction model, which can include the following steps:

[0168] In step S31, a user who purchases a product within a first preset time period after a first specified time is determined as a candidate user.

[0169] In the embodiments of the present disclosure, the first specified time can be any historical time. The first preset time period can be set according to actual needs. For example, the first preset time period can be 1 day, 2 days, or 7 days, etc.

[0170] When training the prediction model, the electronic device selects a first specified time, and a user who purchases a product within a first preset time period after the first specified time is selected as a candidate user. For example Figure 4 As shown in the positive and negative sample user selection diagram, Figure 4 In the diagram, time point 1 is the first specified time, the first preset time period is 7 days, and the electronic device selects a user who purchases a product within 7 days after the time point 1 as a candidate user, i.e. Figure 4 In the diagram, the user who places an order within the time period 1 is a candidate user. The number of candidate users can be one or more.

[0171] Step S32, determining the candidate users who purchase each lowest-level category within a second preset time period after the second specified time as positive sample users of the lowest-level category, and determining the candidate users who do not purchase each lowest-level category within the second preset time period after the second specified time as negative sample users of the lowest-level category, the second specified time being after the first specified time, and the time period between the second specified time and the first specified time being a third preset time period.

[0172] In the embodiments of the present disclosure, the electronic device determines the second specified time after determining the first specified time. The second specified time can be a historical time after the first specified time, and the time period between the second specified time and the first specified time is a third preset time period. The second preset time period can be set according to actual needs. For example, the second preset time period can be 1 month, 6 months, or 1 year, etc. The third preset time period is set according to actual needs. For example, the third preset time period can be 1 day, 7 days, or 14 days, etc.

[0173] For each lowest-level category and each candidate user, if the candidate user purchases the goods under the lowest-level category within the second preset time period after the second specified time, the candidate user is a positive sample user of the lowest-level category; otherwise, the candidate user is a negative sample user of the lowest-level category.

[0174] Still taking Figure 4 as an example for description, Figure 4 In the third preset time period is 14 days, that is, the time point 2 is the second specified time, and the second preset time period is 1 year; for the lowest-level category 1, the electronic device regards the users who purchase the goods under the lowest-level category 1 within 1 year after the time point 2 as positive sample users, that is, Figure 4 In the third preset time period is 14 days, that is, the time point 2 is the second specified time, and the second preset time period is 1 year; for the lowest-level category 1, the electronic device regards the users who purchase the goods under the lowest-level category 1 within 1 year after the time point 2 as positive sample users, that is, Figure 4 In the third preset time period is 14 days, that is, the time point 2 is the second specified time, and the second preset time period is 1 year; for the lowest-level category 1, the electronic device regards the users who purchase the goods under the lowest-level category 1 within 1 year after the time point 2 as positive sample users, that is,

[0175] Step S33, extracting the repurchase behavior features of each sample user corresponding to the category, the category corresponding to the sample user including the lowest-level category corresponding to the repurchase of the sample user and the superior category of the lowest-level category. The same as the above step S12.

[0176] Step S34, training the prediction model based on the repurchase behavior features of each sample user. The same as the above step S13.

[0177] By using the technical solution provided in the embodiments of the present disclosure, the electronic device divides the user into positive sample users and negative sample users based on the user's purchase behavior in different time periods, thereby ensuring the diversity of the samples; and the prediction model is trained by using the diverse positive sample users and negative sample users, thereby improving the accuracy of the prediction model.

[0178] In some embodiments, as shown in FIG. 1, Figure 5 The training method of the prediction model provided in the embodiments of the present disclosure can include the following steps:

[0179] In step S51, the user who purchases a product within a first preset time period after a first specified time point is determined as a candidate user. The same as step S31.

[0180] In step S52, the candidate user who purchases each lowest-level category within a second preset time period after a second specified time point is determined as a positive sample user of the lowest-level category. The same as the step of determining the positive sample user in step S32.

[0181] In step S53, for each lowest-level category, the candidate user who does not purchase any lowest-level category within the second preset time period after the second specified time point is determined as a first negative sample user of the lowest-level category, and the candidate user who does not purchase the lowest-level category but purchases other lowest-level categories within the second preset time period after the second specified time point is determined as a second negative sample user of the lowest-level category.

[0182] Still taking Figure 4 as an example for illustration. Figure 4 In the example, the third preset time period is 14 days, i.e., the time point 2 is the second specified time point, and the second preset time period is 1 year; for the lowest-level category 1, the electronic device determines the user who does not purchase any lowest-level category within 1 year after the time point 2 as the first negative sample user of the lowest-level category 1, i.e. Figure 4 In the example, the user who does not place an order for any product under the lowest-level category within the time period 2 is the first negative sample user of the lowest-level category 1; the user who does not purchase the lowest-level category 1 but purchases other lowest-level categories within 1 year after the time point 2 is the second negative sample user of the lowest-level category 1, i.e. Figure 4 In the example, the user who does not place an order for the product under the lowest-level category 1 within the time period 2 but places an order for the product under other lowest-level categories is the second negative sample user of the lowest-level category 1.

[0183] The embodiments of the present disclosure do not limit the execution order of steps S52 and S53.

[0184] Step S54, extract the repurchase behavior features of the corresponding category of each sample user repurchase. The corresponding category of the sample user includes the lowest level category corresponding to the sample user repurchase and the superior category of the lowest level category. The same as step S12 described above.

[0185] Step S55, train the prediction model based on the repurchase behavior features of each sample user. The same as step S13 described above.

[0186] By applying the technical solutions provided by the embodiments of the present disclosure, the electronic device refines the negative sample users, further enriches the diversity of the samples, and further improves the accuracy of the prediction model.

[0187] In addition, when determining the negative sample users for each lowest level category, the electronic device divides the negative sample users into first negative sample users and second negative sample users according to whether the candidate users have not purchased the goods of other lowest level categories in the case of not purchasing the goods of the lowest level category. The negative sample users are more carefully divided, the repurchase behavior features of the carefully divided negative sample users are input into the prediction model to train the prediction model, which can improve the discrimination of the prediction model and further improve the accuracy of the prediction model.

[0188] In some embodiments, as shown in Figure 6 The embodiments of the present disclosure also provide a training method of a prediction model, which can include the following steps:

[0189] Step S61, determine the users who purchase goods within a first preset time period after a first specified time as candidate users. The same as step S31 described above.

[0190] Step S62, determine the candidate users who purchase each lowest level category within a second preset time period after a second specified time as the positive sample users of the lowest level category. The same as the step of determining the positive sample users in step S32 described above.

[0191] Step S63, for each lowest level category, determine the candidate users who do not purchase any lowest level category within a second preset time period after a second specified time as the first negative sample users of the lowest level category, and determine the candidate users who do not purchase the lowest level category but purchase other lowest level categories within a second preset time period after a second specified time as the second negative sample users of the lowest level category. The same as the step of determining the negative sample users in step S53 described above.

[0192] Step S64, extract the repurchase behavior features of the corresponding category of each sample user repurchase. The corresponding category of the sample user includes the lowest level category corresponding to the sample user repurchase and the superior category of the lowest level category. The same as step S12 described above.

[0193] Step S65, combine the positive sample users, the first negative sample users and the second negative sample users according to multiple ratios to obtain multiple sample user sets.

[0194] In the embodiments of the present disclosure, the combination ratio of the sample users can be set according to actual needs. For example, the combination ratio of the positive sample users, the first negative sample users and the second negative sample users can be 5:4:1, 5:3:2, 5:2:3 or 5:1:4, and the like, as shown in the combination ratio of the sample users. Figure 7 Figure 7 In the sample user area as shown, the rectangular frame filled with right oblique lines represents a positive sample user, the rectangular frame filled with grids represents a first negative sample user, and the rectangular frame filled with vertical lines represents a second negative sample user.

[0195] After obtaining the positive sample users, the first negative sample users and the second negative sample users, the electronic device combines the positive sample users, the first negative sample users and the second negative sample users according to multiple ratios to obtain multiple sample user sets, as shown in Figure 7

[0196] Step S66, training a prediction model by using the repeat purchase behavior features of the users in each sample user set.

[0197] For each sample user set, the electronic device inputs the repeat purchase behavior features of each user in the sample user set into the prediction model to train the prediction model. The electronic device trains multiple prediction models by using multiple sample user sets, as shown in Figure 7

[0198] Step S67, selecting a prediction model with the smallest loss from the multiple trained prediction models as the prediction model for predicting the repeat purchase probability.

[0199] The smaller the loss of the trained prediction model is, the higher the accuracy and recall rate of the prediction model are. Therefore, after training the multiple prediction models, the electronic device selects a prediction model with the smallest loss from the multiple trained prediction models, and uses the selected prediction model as the prediction model for predicting the repeat purchase probability in the future.

[0200] In the embodiments of the present disclosure, the loss of the prediction model can be measured by the area (Area Under Roc, AUC) formed by the Roc curve and the coordinate axis. For example, the smaller the AUC is, the smaller the loss of the prediction model is, and the higher the accuracy of the prediction model is; the larger the AUC is, the larger the loss of the prediction model is, and the lower the accuracy of the prediction model is. At this time, the electronic device can select the prediction model with the largest AUC as the prediction model for predicting the repeat purchase probability in the future, as shown in Figure 7 ​​​The loss of the prediction model can also be measured by other parameters in the embodiments of the present disclosure, which are not limited.

[0201] By applying the technical solutions provided by the embodiments of the present disclosure, the electronic device combines the positive sample user, the first negative sample user and the second negative sample user according to multiple proportions, and then obtains multiple sample user sets. The prediction model is trained by using the features in the sample user sets, so as to select the prediction model with the highest accuracy, which is used for predicting the repurchase probability and recommending the repurchase commodity, thereby improving the accuracy and recall rate of commodity recommendation.

[0202] Based on the prediction model obtained by the above training, the embodiments of the present disclosure further provide a commodity recommendation method, which is described with reference to Figure 8 , Figure 8 The first flowchart of the commodity recommendation method provided by the embodiments of the present disclosure includes the following steps:

[0203] In step S81, the target commodity purchased within the fifth preset time period before the current time and the target user who purchased the target commodity are determined.

[0204] In step S82, in response to the target lowest-level category to which the target commodity belongs being a repurchase category, target repurchase behavior features of the target user for the target lowest-level category are extracted.

[0205] In step S83, the target repurchase behavior features are input into the trained prediction model to obtain the first repurchase probability of the target user.

[0206] In step S84, based on the first repurchase probability, the target user is recommended with commodities under the target lowest-level category.

[0207] In the technical solutions provided by the embodiments of the present disclosure, the electronic device recommends commodities to the target user by using the prediction model in the case that the target lowest-level category to which the target commodity purchased within the fifth preset time period before the current time belongs is a repurchase category, thereby improving the accuracy and recall rate of commodity recommendation. In addition, when recommending commodities to the target user by using the prediction model, the electronic device extracts target repurchase behavior features of the target user for the target lowest-level category. The target repurchase behavior features comprehensively include the repurchase behavior features of the lowest-level category and the upper-level category, which expands the dimension of the repurchase behavior features, improves the accuracy of the prediction model in predicting the repurchase probability, and further improves the accuracy and recall rate of commodity recommendation.

[0208] In the above step S81, the fifth preset time period can be set according to actual needs. For example, the fifth preset time period can be 7 days, 14 days or 1 month, etc. The current time is the time when commodity recommendation is performed.

[0209] The electronic device can determine, according to the purchase records of each user, a product purchased within a fifth preset time period before the current time as a target product, and a user who purchases the target product within the fifth preset time period before the current time as a target user. The number of target users can be one or more.

[0210] In step S82, the repeat purchase category is a lowest-level category with a relatively large repeat purchase probability, and the number of repeat purchase categories can be one or more. For ease of description, in the embodiments of the present disclosure, the lowest-level category to which the target product belongs is referred to as a target lowest-level category.

[0211] After determining the target product, the electronic device can determine whether the target lowest-level category is a repeat purchase category. If the target lowest-level category is a repeat purchase category, the electronic device extracts a repeat purchase behavior feature of the target user for the target lowest-level category, i.e., a target repeat purchase behavior feature. If the target lowest-level category is not a repeat purchase category, the electronic device can not perform any processing to save computing resources of the electronic device.

[0212] Here, the specific meaning and use of the target repeat purchase behavior feature can be referred to the related description in step S12, which will not be repeated here.

[0213] In step S83, the training process of the prediction model can be referred to the related description in the above Figures 1-7 After extracting the target repeat purchase behavior feature, the electronic device inputs the target repeat purchase behavior feature into the prediction model. After the prediction model processes the target repeat purchase behavior feature, the prediction model outputs a probability that the target user purchases a product under the target lowest-level category, i.e., a first repeat purchase probability of the target user. Then, based on the first repeat purchase probability, the electronic device performs step S84 to recommend a product under the target lowest-level category to the target user.

[0214] In some embodiments, as shown in Figure 9 The present disclosure also provides a repeat purchase category determination method, which can include the following steps:

[0215] Step S91: determining a second repeat purchase probability of each lowest-level category according to a repeat purchase index of each lowest-level category.

[0216] In the embodiments of the present disclosure, the repeat purchase index can include one or more of a user layer repeat purchase index and an order layer repeat purchase index. The user layer repeat purchase index can be a proportion of the number of repeat purchase users of the lowest-level category to the total number of purchase users. The order layer repeat purchase index can be a proportion of the number of repeat purchase times of the lowest-level category to the total number of purchase times. The specific user layer repeat purchase index and order layer repeat purchase index can also be expressed in other forms, which can be referred to the related description in step S12, which will not be repeated here.

[0217] The electronic device obtains a repurchase index of each lowest-level category; for each lowest-level category, the electronic device determines a repurchase probability of the lowest-level category, that is, a second repurchase probability. In the embodiments of the present disclosure, the second repurchase probability is the above-mentioned comprehensive repurchase index. In this case, the second repurchase probability can be a product value of the user layer repurchase index and the order layer repurchase index, or a weighted average of the plurality of repurchase indexes of the lowest-level category, that is, the above-mentioned step S91 can be implemented in any one of the following two ways:

[0218] Way one: performing product processing on the plurality of repurchase indexes of each lowest-level category to obtain the second repurchase probability of each lowest-level category.

[0219] Way two: performing weighted average on the plurality of repurchase indexes of each lowest-level category to obtain the second repurchase probability of each lowest-level category.

[0220] Step S92: determining a repurchase category according to the second repurchase probability of each lowest-level category.

[0221] After obtaining the second repurchase probability of each lowest-level category, the electronic device determines a repurchase category from the plurality of lowest-level categories according to the second repurchase probability of each lowest-level category.

[0222] In some embodiments, step S92 can be: taking a preset number of lowest-level categories with the highest second repurchase probability as the repurchase category.

[0223] The preset number can be a specific value set in advance, such as 30, 50, 100, or 400, etc. The preset number can also be determined according to the total number of lowest-level categories and a preset ratio, for example, the electronic device can count the total number of lowest-level categories; and take the total number of the preset ratio as the preset number. The size of the preset ratio can be set according to actual needs, such as 20%, 30%, or 35%, etc. Taking the preset ratio of 30% as an example, if the electronic device counts that the total number of lowest-level categories is 1000, then the preset number is: 1000*30% = 300.

[0224] In other embodiments, step S92 can be: taking the lowest-level category with a second repurchase probability greater than a preset probability as the repurchase category.

[0225] The preset probability can be set according to actual needs, such as 0.70, 0.75, 0.8, or 0.9, etc. Taking the preset probability of 0.7 as an example, the electronic device can take the lowest-level category with a second repurchase probability greater than 0.7 as the repurchase category.

[0226] In the embodiments of the present disclosure, the electronic device can also determine the repurchase category in other ways, which are not limited.

[0227] By applying the technical solution provided in the embodiments of the present disclosure, the electronic device, from the perspective of a group, counts the repurchase indexes of each lowest-level category, determines the second repurchase probability of each lowest-level category, and further screens out repurchase categories. Subsequently, only the user repurchase behaviors under the repurchase categories are predicted to determine whether to recommend the goods under the repurchase categories to the user, thereby reducing the burden of the electronic device and improving the accuracy and recall rate of the goods recommendation.

[0228] In some embodiments, as shown in Figure 10 The embodiments of the present disclosure also provide a goods recommendation method, which can include the following steps:

[0229] Step S101, determining a target good that is purchased within a fifth preset time period before the current time and a target user who purchases the target good. The same as step S81.

[0230] Step S102, in response to the target lowest-level category to which the target good belongs being a repurchase category, extracting a target repurchase behavior feature of the target user repurchasing the target lowest-level category. The same as step S82.

[0231] Step S103, inputting the target repurchase behavior feature into the trained prediction model to obtain a first repurchase probability of the target user. The same as step S83.

[0232] Step S104, in response to the first repurchase probability of the target lowest-level category being greater than a preset threshold, establishing a recall path of the target lowest-level category for the target user.

[0233] In the embodiments of the present disclosure, the preset threshold can be set according to actual needs, such as 0.6, 0.7, or 0.8. After the electronic device obtains the first repurchase probability of the target lowest-level category, it determines whether the first repurchase probability of the target lowest-level category is greater than the preset threshold. If it is greater than the preset threshold, the electronic device establishes a recall path of the target lowest-level category for the target user. The recall path is used to recall the goods under the target lowest-level category from the resource pool of the target lowest-level category.

[0234] Step S105, integrating the recall path into a recommendation system, and the recommendation system is used to recommend the goods under the target lowest-level category to the target user.

[0235] In the embodiments of the present disclosure, the recommendation system is used to call the recall path to recall the goods under the target lowest-level category from the resource pool of the target lowest-level category, and recommend the recalled goods to the user corresponding to the recall path.

[0236] After establishing the recall path of the target lowest level category, the electronic device integrates the recall path into the recommendation system. In this case, the recommendation system can call the recall path of the target lowest level category, recall the goods under the target lowest level category, and then recommend the recalled goods to the target user.

[0237] The technical solution provided by the embodiments of the present disclosure can accurately complete the recommendation of goods, i.e., improve the accuracy of the recommendation of goods, by establishing a recall path for a target user to achieve the purpose of recommending goods under a target lowest level category.

[0238] In addition, with the improvement of the accuracy of the recommendation of goods, the user can obtain the information of the goods of interest, which helps to improve the business scale of the recommendation of goods, the gross merchandise volume (GMV) of the business, the contribution rate of the commercial value of the e-commerce platform, and the user stickiness.

[0239] Based on the same inventive concept, corresponding to the training method of the prediction model, the embodiments of the present disclosure also provide a training device of a prediction model, which is described with reference to Figure 11 , Figure 11 A structural schematic diagram of the training device of the prediction model provided by the embodiments of the present disclosure is shown in FIG. 1. The device includes:

[0240] The first determination module 111 is configured to determine sample users of each lowest level category.

[0241] The first extraction module 112 is configured to extract the repurchase behavior features of the categories corresponding to the repurchases of each sample user. The categories corresponding to the sample user include the lowest level category corresponding to the repurchases of the sample user and the superior category of the lowest level category.

[0242] The training module 113 is configured to train the prediction model based on the repurchase behavior features of each sample user.

[0243] In the training device of the prediction model provided by the embodiments of the present disclosure, the electronic device extracts the repurchase behavior features of the lowest level category corresponding to the repurchases of each sample user and extracts the repurchase behavior features of the superior category of the lowest level category corresponding to the repurchases of each sample user. The repurchase behavior features of the lowest level category and the superior category are combined to train the prediction model for predicting the repurchase probability. This expands the dimension of the repurchase behavior features, increases the repurchase behavior data of a single user for training the prediction model, and then uses the larger repurchase behavior data of a single user to train the prediction model to more accurately predict the repurchase probability. Based on the accurately predicted repurchase probability, the accuracy of the recommendation of goods is improved, and the recall rate of the goods is also improved.

[0244] In some embodiments, the first determination module 111 can include:

[0245] a first determining sub-module, configured to determine, as candidate users, users who purchase the goods within a first preset time period after a first specified time point;

[0246] a second determining sub-module, configured to determine, as positive sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within a second preset time period after a second specified time point, and determine, as negative sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within the second preset time period after the second specified time point, the second specified time point being after the first specified time point, and a time period between the second specified time point and the first specified time point being a third preset time period.

[0247] In some embodiments, the second determining sub-module can be specifically configured to:

[0248] for each lowest-level category, determine, as a first negative sample user of the lowest-level category, a candidate user who does not purchase any lowest-level category within a second preset time period after a second specified time point, and determine, as a second negative sample user of the lowest-level category, a candidate user who does not purchase the lowest-level category and purchases other lowest-level categories within the second preset time period after the second specified time point.

[0249] In some embodiments, the training module 113 can be specifically configured to:

[0250] combine the positive sample users, the first negative sample users and the second negative sample users in multiple proportions to obtain multiple sample user sets;

[0251] train the prediction model respectively by using the repeat-purchase behavior features of the users in each sample user set;

[0252] select, from the multiple trained prediction models, a prediction model with the smallest loss as the prediction model for predicting the repeat-purchase probability.

[0253] In some embodiments, the repeat-purchase behavior features can include: recent behavior features of the user in the lowest-level category and recent behavior features of the user in a superior category of the lowest-level category.

[0254] In some embodiments, the recent behavior features can include: one or more of the click volume, the browse volume, the collection volume and the comment volume of multiple preset time scales within a fourth preset time period before a third specified time point.

[0255] In some embodiments, the repeat-purchase behavior features can further include at least one dimensional feature: a repeat-purchase attribute feature of the lowest-level category, a purchased time feature of the user in the lowest-level category, and a category feature of the lowest-level category.

[0256] In some embodiments, the repurchase attribute feature can include one or more of a user layer repurchase indicator, an order layer repurchase indicator, and a comprehensive repurchase indicator.

[0257] The purchased time feature includes one or more of a time length from a time of first purchase of a commodity under a lowest level category to the third specified time, a time length from a time of last purchase of the commodity under the lowest level category to the third specified time, an average time interval of purchase of the commodity under the lowest level category, and a number of times of purchase of the commodity under the lowest level category.

[0258] The category feature includes one or more of a number of commodities under the lowest level category, a commodity price average, a click rate, and a conversion rate in a plurality of preset time scales.

[0259] In some embodiments, the user layer repurchase indicator can be a proportion of a number of repurchase users of the lowest level category to a number of all purchase users.

[0260] The order layer repurchase indicator is a proportion of a number of times of repurchase of the lowest level category to a number of all purchase times.

[0261] The comprehensive repurchase indicator is a product value of the user layer repurchase indicator and the order layer repurchase indicator, or the comprehensive repurchase indicator is a weighted average value of a plurality of repurchase indicators of the lowest level category.

[0262] Based on the same inventive concept, corresponding to the above-mentioned commodity recommendation method embodiments, the embodiments of the present disclosure also provide a commodity recommendation device, which is described with reference to Figure 12 , Figure 12 A structural schematic diagram of a commodity recommendation device provided by the embodiments of the present disclosure is shown in the figure. The device includes:

[0263] A second determination module 121 is configured to determine a target commodity that is purchased within a sixth preset time length before a current time and a target user who purchases the target commodity;

[0264] A second extraction module 122 is configured to, in response to a target lowest level category to which the target commodity belongs being a repurchase category, extract a target repurchase behavior feature of the target user repurchasing the target lowest level category;

[0265] An input module 123 is configured to input the target repurchase behavior feature into a trained prediction model to obtain a first repurchase probability of the target user;

[0266] A recommendation module 124 is configured to recommend, based on the first repurchase probability, a commodity under the target lowest level category to the target user.

[0267] In the technical scheme provided by the embodiments of the present disclosure, the electronic device utilizes the prediction model to recommend the target user with the target commodity in the case that the target lowest-level category to which the target commodity belongs and which is purchased by the electronic device within the fifth preset time length before the current time is a repeat-purchase category, thereby improving the accuracy and recall rate of the commodity recommendation. In addition, when the prediction model is utilized to recommend the target user with the target commodity, the electronic device extracts a target repeat-purchase behavior feature of the target user for the target lowest-level category, the target repeat-purchase behavior feature comprehensively includes the repeat-purchase behavior features of the lowest-level category and the upper-level category, the dimension of the repeat-purchase behavior feature is widened, the accuracy of the prediction model in predicting the repeat-purchase probability is improved, and the accuracy and recall rate of the commodity recommendation are further improved.

[0268] In some embodiments, the apparatus described above can further include a third determination module configured to determine a repeat-purchase category, including:

[0269] A third determination sub-module configured to determine a second repeat-purchase probability of each lowest-level category according to a repeat-purchase index of each lowest-level category.

[0270] A fourth determination sub-module configured to determine a repeat-purchase category according to the second repeat-purchase probability of each lowest-level category.

[0271] In some embodiments, the number of repeat-purchase indexes can be multiple.

[0272] The third determination sub-module is specifically configured to:

[0273] perform product processing on the multiple repeat-purchase indexes of each lowest-level category to obtain the second repeat-purchase probability of each lowest-level category; or

[0274] perform weighted average processing on the multiple repeat-purchase indexes of each lowest-level category to obtain the second repeat-purchase probability of each lowest-level category.

[0275] In some embodiments, the repeat-purchase index can include one or more of a user layer repeat-purchase index and an order layer repeat-purchase index.

[0276] In some embodiments, the user layer repeat-purchase index can be a proportion of the number of repeat purchases of the lowest-level category to the total number of purchases.

[0277] The order layer repeat-purchase index is a proportion of the number of repeat purchases of the lowest-level category to the total number of purchases.

[0278] In some embodiments, the fourth determination sub-module can be specifically configured to:

[0279] take a preset number of lowest-level categories with the highest second repeat-purchase probability as the repeat-purchase category; or

[0280] take the lowest-level category with a second repeat-purchase probability greater than a preset probability as the repeat-purchase category.

[0281] In some embodiments, the fourth determination submodule is further configured to:

[0282] count the total number of the lowest-level categories; and set the preset number as a preset proportion of the total number.

[0283] In some embodiments, the recommendation module can be specifically configured to:

[0284] in response to the first repeat purchase probability of the target lowest-level category being greater than a preset threshold, establish a recall path of the target lowest-level category for the target user;

[0285] integrate the recall path into a recommendation system, and the recommendation system is configured to recommend a commodity under the target lowest-level category to the target user.

[0286] In the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the technical solutions comply with relevant laws and regulations and do not violate public order and good customs.

[0287] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0288] Figure 13 A schematic block diagram of an example electronic device 130 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the present disclosure described and / or claimed in this document.

[0289] As shown in Figure 13 The device 130 includes a computing unit 131 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 132 or a computer program loaded from a storage unit 138 into a random access memory (RAM) 133. Various programs and data required for the operation of the device 130 can also be stored in the RAM 133. The computing unit 131, the ROM 132, and the RAM 133 are connected to each other through a bus 134. An input / output (I / O) interface 135 is also connected to the bus 134.

[0290] A plurality of components in the device 130 are connected to the I / O interface 135, including: an input unit 136, such as a keyboard, a mouse, etc.; an output unit 137, such as various types of displays, speakers, etc.; a storage unit 138, such as a magnetic disk, an optical disk, etc.; and a communication unit 139, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 139 allows the device 130 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0291] The computing unit 131 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 131 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 131 performs various methods and processes described above, such as the prediction model training method or the item recommendation method. For example, in some embodiments, the prediction model training method or the item recommendation method can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 138. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 130 via the ROM 132 and / or the communication unit 139. When the computer program is loaded to the RAM 133 and executed by the computing unit 131, one or more steps of the prediction model training method or the item recommendation method described above can be performed. Alternatively, in other embodiments, the computing unit 131 can be configured to perform the prediction model training method or the item recommendation method by any other appropriate means, such as by means of firmware.

[0292] The embodiments of the present disclosure further provide an electronic device, such as Figure 14 As shown, the electronic device includes at least one processor 141; and a memory 142 connected to the at least one processor in communication; wherein

[0293] The memory 142 stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the prediction model training method or the item recommendation method described above.

[0294] The embodiments of the present disclosure further provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the prediction model training method or the item recommendation method described above.

[0295] The embodiments of the present disclosure further provide a computer program product comprising a computer program which, when executed by a processor, implements the above-mentioned prediction model training method or the commodity recommendation method.

[0296] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0297] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0298] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0299] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0300] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0301] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0302] It should be understood that various forms of flow shown above can be used, with steps reordered, added, or removed. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, without limitation herein, so long as the desired results of the technology disclosed in the present disclosure are achieved.

[0303] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and principles of the disclosure. Accordingly, the disclosure is not limited to the specific embodiments described above, but only by the scope of the appended claims.

Claims

1. A method for training a prediction model, the method comprising: determining sample users of each lowest-level category; extracting a repurchase behavior feature of a category corresponding to a repurchase of each sample user, the category corresponding to the sample user including a lowest-level category corresponding to a repurchase of the sample user and a superior category of the lowest-level category; training a prediction model based on the repurchase behavior feature of each sample user, an output result of the prediction model being a repurchase probability of each sample user, a training target of the prediction model being that the prediction model outputs a repurchase probability of a positive sample user close to 1 and a repurchase probability of a negative sample user close to 0; wherein the step of determining the sample users of each lowest-level category comprises: determining users who purchase goods within a first preset time period after a first specified time as candidate users; determining, as positive sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within a second preset time period after a second specified time, and determining, as negative sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within the second preset time period after the second specified time, the second specified time being after the first specified time, and a time period between the second specified time and the first specified time being a third preset time.

2. The method of claim 1, wherein, The step of determining, as negative sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within a second preset time period after a second specified time comprises: for each lowest-level category, determining, as a first negative sample user of the lowest-level category, a candidate user who does not purchase any lowest-level category within a second preset time period after a second specified time, and determining, as a second negative sample user of the lowest-level category, a candidate user who does not purchase the lowest-level category and purchases other lowest-level categories within the second preset time period after the second specified time.

3. The method of claim 2, wherein, The step of training a prediction model based on the repurchase behavior feature of each sample user comprises: combining positive sample users, first negative sample users and second negative sample users in multiple proportions to obtain multiple sample user sets; training prediction models respectively using repurchase behavior features of users in each sample user set; selecting, from the multiple prediction models obtained by training, a prediction model with the smallest loss as a prediction model for predicting a repurchase probability.

4. The method according to any one of claims 1 to 3, wherein, The repurchase behavior feature includes a recent behavior feature of a user in a lowest-level category and a recent behavior feature of a user in a superior category of the lowest-level category.

5. The method of claim 4, wherein, The recent behavior feature includes one or more of a click volume, a browse volume, a collection volume and a comment volume in multiple preset time scales within a fourth preset time period before a third specified time.

6. The method of claim 4, wherein, The repurchase behavior feature further includes at least one dimension feature: a repurchase attribute feature of a lowest-level category, a purchased time feature of a user in a lowest-level category, and a category feature of a lowest-level category.

7. The method of claim 6, wherein, The repurchase attribute feature includes one or more of a user layer repurchase indicator, an order layer repurchase indicator and a comprehensive repurchase indicator. The purchased time feature includes one or more of a time length from a time of first purchase of a product under the lowest-level category to the third specified time, a time length from a time of last purchase of a product under the lowest-level category to the third specified time, an average time interval of purchase of a product under the lowest-level category, and a number of times of purchase of a product under the lowest-level category. The category feature includes one or more of a number of products under the lowest-level category, a mean product price, and a click rate and a conversion rate in a plurality of preset time scales.

8. The method of claim 7, wherein, The user-level repeat purchase indicator is a proportion of a number of repeat purchase users of the lowest-level category to a total number of purchase users. The order placement level repeat purchase indicator is a proportion of a number of repeat purchases of the lowest-level category to a total number of purchases. The comprehensive repeat purchase indicator is a product of the user-level repeat purchase indicator and the order placement level repeat purchase indicator, or the comprehensive repeat purchase indicator is a weighted average of a plurality of repeat purchase indicators of the lowest-level category.

9. A product recommendation method, comprising: determining a target product that is purchased within a fifth preset time length before a current time and a target user who purchases the target product; in response to a target lowest-level category to which the target product belongs being a repeat purchase category, extracting a target repeat purchase behavior feature of the target user who repeats purchase of the target lowest-level category; inputting the target repeat purchase behavior feature into a prediction model trained by the method of any one of claims 1-8 to obtain a first repeat purchase probability of the target user; based on the first repeat purchase probability, recommending a product under the target lowest-level category to the target user.

10. The method of claim 9, wherein, The repeat purchase category is determined by the following steps: determining a second repeat purchase probability of each lowest-level category according to a repeat purchase indicator of each lowest-level category; determining a repeat purchase category according to the second repeat purchase probability of each lowest-level category.

11. The method of claim 10, wherein, The number of repeat purchase indicators is a plurality; The step of determining a second repeat purchase probability of each lowest-level category according to a repeat purchase indicator of each lowest-level category includes: performing product processing on a plurality of repeat purchase indicators of each lowest-level category to obtain a second repeat purchase probability of each lowest-level category; or performing weighted average on a plurality of repeat purchase indicators of each lowest-level category to obtain a second repeat purchase probability of each lowest-level category.

12. The method of claim 10 or 11, wherein, The repeat purchase indicator includes one or more of a user-level repeat purchase indicator and an order placement level repeat purchase indicator.

13. The method of claim 12, wherein, The user-level repeat purchase indicator is a proportion of a number of repeat purchase users of the lowest-level category to a total number of purchase users. The order placement level repeat purchase indicator is a proportion of a number of repeat purchases of the lowest-level category to a total number of purchases.

14. The method of claim 10, wherein, The step of determining a repeat purchase category according to the second repeat purchase probability of each lowest-level category includes: taking a preset number of lowest-level categories with the highest second repeat purchase probability as the repeat purchase category; or taking a lowest-level category with a second repeat purchase probability greater than a preset probability as the repeat purchase category.

15. The method of claim 14, wherein, The preset number is determined by the following steps: counting a total number of lowest-level categories; and taking a preset proportion of the total number as the preset number.

16. The method of any one of claims 9-11 and 13-15, wherein, The step of recommending a product under the target lowest-level category to the target user based on the first repeat purchase probability includes: in response to the first repeat purchase probability of the target lowest-level category being greater than a preset threshold, establishing a recall channel of the target lowest-level category for the target user; integrating the recall channel into a recommendation system, the recommendation system being configured to recommend a commodity under the target lowest-level category to the target user.

17. A device for training a prediction model, the device comprising: a first determination module configured to determine sample users of each lowest-level category; a first extraction module configured to extract repeat purchase behavior features of a category corresponding to repeat purchases of each sample user, the category corresponding to the sample user including a lowest-level category corresponding to repeat purchases of the sample user and a superior category of the lowest-level category; a training module configured to train a prediction model based on the repeat purchase behavior features of each sample user, an output result of the prediction model being a repeat purchase probability of each sample user, and a training target of the prediction model being to make the repeat purchase probability of a positive sample user output by the prediction model close to 1 and the repeat purchase probability of a negative sample user close to 0; wherein the first determination module comprises: a first determination submodule configured to determine users who purchase commodities within a first preset time period after a first specified time as candidate users; a second determination submodule configured to determine, as positive sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within a second preset time period after a second specified time, and determine, as negative sample users of each lowest-level category, candidate users who do not purchase the lowest-level category within the second preset time period after the second specified time, the second specified time being after the first specified time, and a time period between the second specified time and the first specified time being a third preset time.

18. The apparatus of claim 17, wherein, The second determination submodule is specifically configured to: for each lowest-level category, determine, as a first negative sample user of the lowest-level category, a candidate user who does not purchase any lowest-level category within the second preset time period after the second specified time, and determine, as a second negative sample user of the lowest-level category, a candidate user who does not purchase the lowest-level category and purchases other lowest-level categories within the second preset time period after the second specified time.

19. The apparatus of claim 18, wherein, The training module is specifically configured to: combine the positive sample users, the first negative sample users and the second negative sample users in multiple proportions to obtain multiple sample user sets; train the prediction model based on the repeat purchase behavior features of the users in each sample user set; select, from the multiple prediction models obtained by training, a prediction model with the smallest loss as the prediction model for predicting the repeat purchase probability.

20. The apparatus of any of claims 17-19, wherein, The repeat purchase behavior features include recent behavior features of the user in the lowest-level category and recent behavior features of the user in the superior category of the lowest-level category.

21. The apparatus of claim 20, wherein, The recent behavior features include one or more of click volume, browse volume, collection volume and comment volume in multiple preset time scales within a fourth preset time period before a third specified time.

22. The apparatus of claim 20, wherein, The repeat purchase behavior features further include at least one dimensional feature of a repeat purchase attribute feature of the lowest-level category, a purchased time feature of the user in the lowest-level category and a category feature of the lowest-level category.

23. The apparatus of claim 22, wherein, The repeat purchase attribute feature comprises one or more of a user layer repeat purchase index, an order layer repeat purchase index, and a comprehensive repeat purchase index. The purchased time feature comprises one or more of a time interval between a time of first purchase of a commodity under a lowest level category and a third specified time, a time interval between a time of last purchase of the commodity under the lowest level category and the third specified time, an average time interval of purchase of the commodity under the lowest level category, and a number of times of purchase of the commodity under the lowest level category. The category feature comprises one or more of a number of commodities under the lowest level category, a commodity price average, a click rate, and a conversion rate in a plurality of preset time scales.

24. The apparatus of claim 23, wherein, The user layer repeat purchase index is a proportion of a number of repeat purchase users of the lowest level category to a total number of purchase users. The order layer repeat purchase index is a proportion of a number of repeat purchase times of the lowest level category to a total number of purchase times. The comprehensive repeat purchase index is a product of the user layer repeat purchase index and the order layer repeat purchase index, or the comprehensive repeat purchase index is a weighted average of a plurality of repeat purchase indexes of the lowest level category.

25. A commodity recommendation device, comprising: a second determination module configured to determine a target commodity that is purchased within a fifth preset time interval before a current time and a target user who purchases the target commodity; a second extraction module configured to, in response to a target lowest level category to which the target commodity belongs being a repeat purchase category, extract a target repeat purchase behavior feature of the target user who repeats purchase of the target lowest level category; an input module configured to input the target repeat purchase behavior feature into a prediction model trained by the device of any one of claims 17-24 to obtain a first repeat purchase probability of the target user; a recommendation module configured to recommend, based on the first repeat purchase probability, a commodity under the target lowest level category to the target user.

26. The device of claim 25, further comprising a third determination module configured to determine the repeat purchase category, comprising: a third determination submodule configured to determine, according to a repeat purchase index of each lowest level category, a second repeat purchase probability of each lowest level category; a fourth determination submodule configured to determine, according to the second repeat purchase probability of each lowest level category, a repeat purchase category.

27. The apparatus of claim 26, wherein, The number of repeat purchase indexes is a plurality; The third determination submodule is specifically configured to: perform product processing on the plurality of repeat purchase indexes of each lowest level category to obtain the second repeat purchase probability of each lowest level category; or perform weighted average on the plurality of repeat purchase indexes of each lowest level category to obtain the second repeat purchase probability of each lowest level category.

28. The apparatus of claim 26 or 27, wherein, The repeat purchase index comprises one or more of a user layer repeat purchase index and an order layer repeat purchase index.

29. The apparatus of claim 28, wherein, The user layer repeat purchase index is a proportion of a number of repeat purchase times of the lowest level category to a total number of purchase times. The order layer repeat purchase index is a proportion of a number of repeat purchase times of the lowest level category to a total number of purchase times.

30. The apparatus of claim 26, wherein, The fourth determination submodule is specifically configured to: select a preset number of lowest level categories with the highest second repeat purchase probability as the repeat purchase category; or select a lowest level category with a second repeat purchase probability greater than a preset probability as the repeat purchase category.

31. The apparatus of claim 30, wherein, The fourth determining sub-module is further configured to count a total number of the lowest-level categories; and set a preset number as a preset proportion of the total number.

32. The apparatus of any one of claims 25-27 and 29-31, wherein, The recommendation module is specifically configured to: in response to the first repeat purchase probability of the target lowest-level category being greater than a preset threshold, establish a recall path of the target lowest-level category for the target user; and integrate the recall path into a recommendation system, the recommendation system being configured to recommend goods under the target lowest-level category to the target user. 33.An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8 or 9-16.

34. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method of any one of claims 1-8 or 9-16. 35.A computer program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1-8 or 9-16.

Images