An information processing method and device, and a storage medium

Abstract

The application discloses an information processing method, which comprises the following steps: inputting a first feature into a screening model to obtain a candidate score of each candidate user in at least one candidate user, wherein the first feature at least comprises description information of a target category and user features of each candidate user in the at least one candidate user, and the candidate score is used for representing the interested degree of the candidate user corresponding to the candidate score in the target category; determining at least one target user from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user; and outputting an object corresponding to the target category to the at least one target user. In addition, the application further discloses an information processing device and a storage medium.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information processing, and relates to but is not limited to an information processing method and device and a storage medium. BACKGROUND

[0002] In related technologies, a terminal device is installed with an application program (APP), and when the APP recommends a commodity to a user, the APP usually needs to recommend a favorite commodity to the user according to an analysis conclusion. Since the analysis conclusion is obtained by data analysts, the efficiency of the analysis is low, and since the experience of a person is limited, the analysis conclusion and scheme are greatly affected by human factors, thereby causing low accuracy and efficiency of commodity recommendation to the user. SUMMARY

[0003] Embodiments of the present application provide an information processing method and device and a storage medium to solve at least one problem in related technologies, and the accuracy and efficiency of screening a user and recommending a commodity to the screened user can be improved.

[0004] The technical solution of the present application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides an information processing method, and the method comprises the following steps:

[0006] inputting a first feature into a screening model to obtain a candidate score of each candidate user in at least one candidate user, the first feature at least comprising: description information of a target category, user features of each candidate user in the at least one candidate user, and the candidate score of the candidate user being used to represent an interested degree of the candidate user corresponding to the candidate score in the target category;

[0007] determining at least one target user from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user;

[0008] outputting an object corresponding to the target category to the at least one target user.

[0009] In a second aspect, an embodiment of the present application provides an information processing device, and the device comprises:

[0010] a first determination unit configured to input a first feature into a screening model to obtain a candidate score of each candidate user in at least one candidate user, the first feature at least comprising: description information of a target category, user features of each candidate user in the at least one candidate user, and the candidate score of the candidate user being used to represent an interested degree of the candidate user corresponding to the candidate score in the target category;

[0011] a second determining unit, configured to determine at least one target user from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user;

[0012] an output unit, configured to output an object corresponding to the target category to the at least one target user.

[0013] In a third aspect, an embodiment of the present application provides a storage medium, which stores a computer program. The computer program is executed by a processor to implement the information processing method.

[0014] The present application provides an information processing method and device, and a storage medium. The information processing device inputs a first feature into a screening model to obtain a candidate score of each candidate user in at least one candidate user. The first feature at least includes description information of a target category and user features of each candidate user in the at least one candidate user. The candidate score of the candidate user is used to represent the degree of interest of the candidate user corresponding to the candidate score in the target category. At least one target user is determined from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user. An object corresponding to the target category is output to the at least one target user. In this way, when screening target users, the screening model screens according to the first feature, so that the accuracy and efficiency of the screened target users can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 An optional structural schematic diagram of an information processing system provided by an embodiment of the present application;

[0016] Figure 2 An optional flowchart of an information processing method provided by an embodiment of the present application;

[0017] Figure 3 An optional flowchart of an information processing method provided by an embodiment of the present application;

[0018] Figure 4 An optional flowchart of an information processing method provided by an embodiment of the present application;

[0019] Figure 5 An optional flowchart of determining a potential commodity provided by an embodiment of the present application;

[0020] Figure 6 An optional flowchart of determining a data source provided by an embodiment of the present application;

[0021] Figure 7 An optional flowchart of determining a data source provided by an embodiment of the present application;

[0022] Figure 8 An optional flowchart for determining a data source provided by an embodiment of the present application;

[0023] Figure 9 An optional flowchart for mining festival commonality and difference provided by an embodiment of the present application;

[0024] Figure 10 An optional structural diagram of an information processing device provided by an embodiment of the present application;

[0025] Figure 11 An optional flowchart for mining festival commonality and difference provided by an embodiment of the present application;

[0026] Figure 12 An optional flowchart of an information processing method provided by an embodiment of the present application;

[0027] Figure 13 An optional flowchart of an information processing method provided by an embodiment of the present application;

[0028] Figure 14 An optional structural diagram of an information processing device provided by an embodiment of the present application;

[0029] Figure 15 An optional structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will further describe the specific technical solutions of the present application with reference to the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application, but not to limit the scope of the present application.

[0031] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict.

[0032] In the following description, the terms "first\second\third" are only used to distinguish different objects, and do not represent a specific order of the objects, and do not have a limitation of sequence. It can be understood that "first\second\third" can be interchanged in a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0034] It is understood that the embodiments of this application involve data related to user age, user gender, and user occupation. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and an option to disable the collection, use and processing of related data must be provided.

[0035] The information processing method in this application embodiment can be applied to... Figure 1 The information processing system 100 shown is as follows: Figure 1 As shown, the information processing system 100 includes a server 10 and a client 20. The client 20 is capable of interacting with the user based on input devices, including devices such as a monitor, mouse, and keyboard that can receive user input information.

[0036] In one example, such as Figure 1 As shown, server 10 and client 20 are located on different physical entities. At this time, server 10 communicates with client 20 through network 30.

[0037] The client 20 runs a browser or application (APP) capable of displaying the page.

[0038] The information processing method provided in this application embodiment can be applied to an information processing device, which can be a server 10 or a client 20.

[0039] The information processing device inputs a first feature into a screening model to obtain candidate scores for each candidate user among at least one candidate user. The first feature includes at least: descriptive information of the target category and user characteristics of each candidate user among the at least one candidate user. The candidate scores of the candidate users are used to characterize the degree of interest of the candidate user corresponding to the candidate score in the target category. Based on the candidate scores of each candidate user among the at least one candidate user, at least one target user is determined from the at least one candidate user. The object corresponding to the target category is output to the at least one target user.

[0040] When the information processing device is server 10, after determining at least one target user, server 10 sends the object corresponding to the target category to client 20, and client 20 displays the object corresponding to the target category to at least one target user.

[0041] In the case that the information processing device is the client 20, after the client 20 determines the at least one target object, the client 20 directly displays the objects corresponding to the target category to the at least one target user.

[0042] In the following, the various embodiments of the information processing method and device provided by the embodiments of the present application, the storage medium will be described in detail in combination with the schematic diagram of the information processing system shown in the drawings. Figure 1

[0043] Figure 2 The implementation flowchart of the information processing method provided by the embodiments of the present application is shown in the drawings, and the method comprises the following steps: Figure 2

[0044] S201, the information processing device inputs the first feature into the screening model to obtain the candidate score of each candidate user in the at least one candidate user.

[0045] Here, the first feature includes: the description information of the target category, and the user feature of each candidate user in the at least one candidate user. The description information of the target category is used to describe the target category.

[0046] In an example, the at least one candidate user includes: candidate user 1, candidate user 2 and candidate user 3, wherein the user feature of candidate user 1 is feature A, the user feature of candidate user 2 is feature B, the user feature of candidate user 3 is feature C, and the description information of the target category is fresh food. The information processing device inputs feature A, feature B, feature C and fresh food into the screening model to obtain the candidate score of candidate user 1 as 0.3, the candidate score of candidate user 2 as 0.8, and the candidate score of candidate user 3 as 0.9.

[0047] For the target category, it can be a target festival or a target commodity, and the embodiments of the present application do not make any limitation on this.

[0048] In the case that the target category is a target festival, the target festival is any one of the at least one festival, and the at least one festival includes: Qixi Festival, Valentine's Day, Spring Festival and Dragon Boat Festival, etc. For example, the target festival is Valentine's Day.

[0049] For the at least one festival, the at least one festival can be classified to determine at least one festival type, wherein each festival type in the at least one festival type corresponds to a festival attribute feature. After determining the at least one festival type, for each festival type in the at least one festival type, the festival attribute feature corresponding to the festival type is the attribute feature of the festival included in the festival type.

[0050] ​​In an example, the at least one festival includes: Qixi Festival, Valentine's Day, Spring Festival, and Dragon Boat Festival, and the at least one festival type includes: a love festival and a traditional festival, where the love festival corresponds to a love attribute, and the traditional festival corresponds to a traditional attribute. After classifying the Qixi Festival, Valentine's Day, Spring Festival, and Dragon Boat Festival, it can be determined that the festivals in the love festival include the Qixi Festival and Valentine's Day, and the festivals in the traditional festival include the Spring Festival and the Dragon Boat Festival. Therefore, it can be determined that the festival attribute characteristics of the Qixi Festival and Valentine's Day are love attribute characteristics, and the festival attribute characteristics of the Spring Festival and the Dragon Boat Festival are traditional attribute characteristics.

[0051] Here, the naming of the festival type is not limited by the embodiments of the present application. For example, for festivals such as the Spring Festival and the Dragon Boat Festival, the festival type of the Spring Festival and the Dragon Boat Festival can be named as a traditional festival, and the festival type of the Spring Festival and the Dragon Boat Festival can also be named as a general festival, which is not limited by the embodiments of the present application.

[0052] The festival attribute characteristics corresponding to the festival type are not limited by the embodiments of the present application. For example, for a traditional festival, the festival attribute characteristics corresponding to the traditional festival can be traditional attribute characteristics, and the festival attribute characteristics corresponding to the traditional festival can also be family attribute characteristics, which is not limited by the embodiments of the present application.

[0053] The at least one festival object corresponding to the festival attribute characteristics is not limited by the embodiments of the present application.

[0054] In an example, the festival attribute characteristics are love attribute characteristics, and the at least one festival object corresponding to the love attribute characteristics includes: roses, chocolates, and other goods that can express love.

[0055] The user characteristics of each candidate user in the at least one candidate user can include: age, occupation, gender, and other characteristics representing user attributes.

[0056] It should be noted that in the embodiments of the present application, the information processing device can obtain the private information such as the age, occupation, and gender of the user only after obtaining the authorization of the user. If the information processing device does not receive the authorization of the user, the information processing device cannot obtain the private information such as the age, occupation, and gender of the user.

[0057] For each candidate user in the at least one candidate user, the candidate score of the candidate user is used to represent the interest degree of the candidate user corresponding to the candidate score in the target category. If the candidate score of the candidate user is high, the interest degree of the candidate user corresponding to the candidate score in the target category is high; if the candidate score of the candidate user is low, the interest degree of the candidate user corresponding to the candidate score in the target category is low.

[0058] In the embodiments of the present application, for the user feature of each candidate user in the at least one candidate user, the user feature can further include a historical feature, which is used to indicate at least one object operated by the candidate user in a historical time period.

[0059] Here, for the historical time period, the embodiments of the present application do not make any limitation. For example, the historical time period is the past seven days, or the historical time period is the past one month.

[0060] S202, the information processing device determines at least one target user from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user.

[0061] Here, for the determination condition of determining at least one target user from the at least one candidate user, the embodiments of the present application do not make any limitation. For example, the candidate user whose candidate score in the at least one candidate score is greater than a set score threshold is determined as the at least one target user.

[0062] In an example, the at least one candidate user includes candidate user 1, candidate user 2 and candidate user 3, wherein the candidate score of candidate user 1 is 0.3, the candidate score of candidate user 2 is 0.6, the candidate score of candidate user 3 is 0.7, the set score threshold is 0.5, and the information processing device will determine candidate user 2 and candidate user 3 whose candidate score is greater than the set score threshold 0.5 from candidate user 1, candidate user 2 and candidate user 3 according to the candidate score 0.3, the candidate score 0.6 and the candidate score 0.7.

[0063] S203, the information processing device outputs the object corresponding to the target category to the at least one target user.

[0064] Here, for the object corresponding to the target category, the object can include objects such as commodities, advertisements and news displayed to the user.

[0065] After determining the at least one target user, the information processing device outputs the object corresponding to the target category to the at least one target user.

[0066] The embodiment of the present application provides an information processing method, an information processing device inputs first features into a screening model to obtain candidate scores of each candidate user in at least one candidate user, wherein the first features at least include: description information of a target category, user features of each candidate user in the at least one candidate user, and the candidate score of the candidate user is used to represent the degree of interest of the candidate user corresponding to the candidate score to the target category; at least one target user is determined from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user; and an object corresponding to the target category is output to the at least one target user. In this way, when screening the target user, the screening model screens according to the first features, so that the accuracy and efficiency of the screened target user can be improved.

[0067] In the embodiment of the present application, taking one holiday in a plurality of holidays as an example, a circle person model can be used to circle out users corresponding to each holiday in at least one holiday. The circle person model can be implemented by the screening model.

[0068] The features input into the circle person model include at least one of commonalities between different holidays, characteristics of different holidays, user behavior features and user portrait features. The commonalities between different holidays include the same holiday attributes between different holidays, the user behavior features are used to represent the behavior of the user to the commodity, and the user portrait features include user features, the user features include user age, user occupation, user gender and the like.

[0069] For the circle person model, the deep learning network model can be improved to obtain the circle person model, so that the circle person model can circle out different users corresponding to different holiday scenes according to different holiday scenes, thereby guiding the business party to accurately put the holiday commodities to the users circled out or accurately put the holiday coupons to the users circled out. The information processing method provided by the embodiment of the present application can determine the user group with potential shopping demand for different holidays, such as more shopping demand of lovers on Valentine's Day and more shopping demand of students on Teacher's Day.

[0070] Here, for the commonalities between different holidays and the characteristics of different holidays, the commonalities between different holidays and the characteristics of different holidays are analyzed in detail to obtain analysis conclusions of the commonalities between different holidays and the characteristics of different holidays, and the analysis conclusions of the commonalities and characteristics are input into the circle person model as features, so that the circle person model circles people according to the analysis conclusions of the commonalities and characteristics.

[0071] In some embodiments, the screening model at least includes: an interest determination subnetwork, S201 inputs the first feature into the screening model to obtain a candidate score of each candidate user in the at least one candidate user, including S2011 and S2012:

[0072] S2011, for each candidate user in the at least one candidate user, the information processing device inputs the user feature of the candidate user into the interest determination subnetwork to obtain the interest information of the candidate user.

[0073] Here, the interest information represents at least two interests of the candidate user. Wherein, for the content of the at least two interests, the embodiments of the present application do not make any limitation. For example: the at least two interests include fresh food and clothing goods.

[0074] In an example, the at least one candidate user includes: candidate user 1, candidate user 2 and candidate user 3, and the user feature of the candidate user includes: historical features, wherein the historical feature corresponding to the candidate user 1 is feature 11, the historical feature corresponding to the candidate user 2 is feature 21, and the historical feature corresponding to the candidate user 3 is feature 31. The information processing device inputs the historical feature 11 of the candidate user 1, the historical feature 21 of the candidate user 2 and the historical feature 31 of the candidate user 3 into the interest determination subnetwork to obtain that the at least two interests of the candidate user 1 include fresh food and clothing goods, the at least two interests of the candidate user 2 include stationery and furniture, and the at least two interests of the candidate user 3 include fresh food and stationery.

[0075] S2012, the information processing device determines the candidate user in the at least one candidate user whose at least two interests include the target category, and determines the candidate score of the candidate user whose at least two interests include the target category.

[0076] Here, after determining the at least two interests corresponding to each candidate user in the at least one candidate user, for the candidate user whose at least two interests include the target category, the information processing device determines the candidate score of the candidate user.

[0077] In an example, the at least one candidate user includes: candidate user 1, candidate user 2 and candidate user 3, wherein the at least two interests of the candidate user 1 include fresh food and clothing goods, the at least two interests of the candidate user 2 include stationery and furniture, and the at least two interests of the candidate user 3 include fresh food and stationery. The target category is fresh food, and the information processing device determines the candidate scores of the candidate user 1 and the candidate user 3 whose at least two interests include fresh food.

[0078] The deep learning network model is improved in the application to obtain the circle person model in the application. The improvement of the deep learning network model includes adding a multi-interest network and a label-aware attention mechanism in the deep learning network model to form the circle person model in the application.

[0079] Here, after adding the multi-interest network, the multiple interests of the user can be obtained through the multi-interest network, so that the information processing device can recommend the goods matched with the multiple interests of the user according to the obtained multiple interests of the user; after adding the label-aware attention mechanism, the multiple interests of the user can be better learned.

[0080] In some embodiments, the information processing method provided by the embodiments of the application further includes steps S204 to S205:

[0081] S204, the information processing device obtains at least two pieces of first training data, each of the at least two pieces of first training data includes: a first training feature and a first label corresponding to the first training feature, the first training feature includes: description information of a training interest, training features of a training user, the first label is used to represent the interested degree of the training user to the training interest, and the first label is determined based on the historical behavior of the training user.

[0082] Here, the description information of the training interest is used to represent the training interest. For example, the training interest represented by the description information of the training interest is fresh food.

[0083] The user features of the training user include features representing the user's own attributes such as age, occupation and gender, and historical features.

[0084] S205, the information processing device trains the initial screening model according to the at least two pieces of first training data to obtain the screening model.

[0085] In the embodiments of the application, the order between S204 to S205 and steps S201 to S203 is not limited.

[0086] In some embodiments, the above S205 includes:

[0087] S205a, the information processing device inputs the first training feature in the at least two pieces of first training data to the initial screening model to obtain a first predicted score;

[0088] Here, the first prediction score is used to represent the predicted interest level of each of the at least one training user in the training interest. The application embodiments do not make any limitation on the value range of the first prediction score. For example, the value range of the first prediction score is 0 to 1.

[0089] In an example, the at least one training user includes training user 1 and training user 2, wherein the first prediction score of the training user 1 is 0.9, indicating that the predicted interest level of the training user 1 in the training interest is 90%, and the first prediction score of the candidate user 2 is 0.5, indicating that the predicted interest level of the candidate user 2 in the training interest is 50%.

[0090] S205b, the information processing device determines a first loss value of the initial screening model according to the first prediction score corresponding to each of the at least two pieces of first training data and the first label of the first training data, in a case where the initial screening model does not converge.

[0091] Here, the information processing device can determine the first loss value according to the first prediction score and the first label after obtaining the first prediction score, and the first loss value can be the cross-entropy between the first prediction score and the first label.

[0092] S205c, the information processing device updates the parameters of the initial screening model based on the first loss value to obtain the screening model, and continues to input the first training features in the at least two pieces of first training data to the initial screening model until the initial screening model converges to obtain the screening model.

[0093] Here, the information processing device continues to input the first training features in the at least two pieces of first training data to the initial screening model until the initial screening model converges to obtain the screening model, after updating the parameters of the initial screening model based on the first loss value.

[0094] In some embodiments, determining the first loss value of the initial screening model according to the first prediction score corresponding to each of the at least two pieces of first training data and the first label of the first training data includes: the information processing device determines the first loss value of the initial screening model according to the first prediction score corresponding to each of the at least two pieces of training data, the first label of the training data, and the weight corresponding to the first label.

[0095] In some embodiments, the method further includes: determining the user characteristics of the training user, user interaction objects, and interest objects corresponding to the training interest among the user interaction objects, based on the first label used to characterize the degree of interest of the training user in the training interest, and determining the weight corresponding to the first label.

[0096] Here, the user features used to train users are obtained through P. u This indicates that the user interaction object is accessed through I u This indicates that the objects of interest corresponding to the training interests in the user interaction objects are represented by Fi, and the information processing device will use I... u and P u Determine the training user's training interest V u Specifically, as shown in the following formula (1):

[0097] V u =F i (I u ,P u ) formula (1);

[0098] According to F i Determine the training object e to be output to the training user. i Specifically, as shown in formula (2) below:

[0099] e i =G item (F i ) formula (2);

[0100] According to e i and V u Determine the weight 'a' corresponding to the first label. u Specifically, as shown in formula (3) below:

[0101] a u =Attention(e i V u V u ) formula (3).

[0102] Taking the target category as the target holiday as an example, such as Figure 3 As shown, the steps for selecting users corresponding to the target holiday using the user segmentation model include:

[0103] S301. Design a user base pool corresponding to holiday scenarios.

[0104] According to business analysis, the user base pool for each holiday scenario is different.

[0105] Here, in the case that the festival scenario is Valentine's Day, the festival attribute of the Valentine's Day is determined as love, the festival attribute of the love is input into the circle person model, the circle person model determines the goods corresponding to the love festival, including chocolate, roses and other goods representing love, and then it is determined whether the user's historical behavior data has clicked, browsed or added to purchase chocolate, roses and other goods representing love, and the user who has clicked, browsed or added to purchase chocolate, roses and other goods representing love is determined as the user corresponding to the love festival.

[0106] S302, construct features input into the circle person model and labels of users.

[0107] Here, the features input into the circle person model include user behavior features, user portrait features, commonalities between different festivals, and characteristics of different festivals.

[0108] For the label of the user, if the user has ordered goods in the historical behavior data, the label is 1, and if the user does not have ordered goods in the historical behavior data, the label is 0.

[0109] S303, improve the deep learning network model to obtain the circle person model.

[0110] Here, the multi-interest network layer and the label-aware-attention mechanism can be added to the deep learning network model.

[0111] S304, train the circle person model to generate a circle person model file.

[0112] S305, determine new data of online users.

[0113] Here, the new data of the user includes at least one of the commonalities between different festivals, the characteristics of different festivals, the user behavior features and the user portrait features.

[0114] S306, input the new data of the user into the circle person model to predict the user corresponding to the target festival.

[0115] In some embodiments, the information processing method provided by the embodiments of the present application further includes: the information processing device inputs the second feature into the selected product model to obtain a candidate score of each candidate object in the at least one candidate object; determines at least one target object from the at least one candidate object according to the candidate score of each candidate object in the at least one candidate object; and correspondingly, the outputting the object corresponding to the target category to the at least one target user includes outputting the at least one target object to the at least one target user.

[0116] The second feature includes at least: description information of the target category, object features of each candidate object in the at least one candidate object, and a candidate score of the candidate object, which is used to represent a matching degree of the candidate object corresponding to the candidate score and the target category.

[0117] In an example, the at least one candidate object includes: candidate object 1, candidate object 2, and candidate object 3, wherein the user feature of candidate object 1 is feature A, the user feature of candidate object 2 is feature B, the user feature of candidate object 3 is feature C, and the description information of the target category is fresh food, and the information processing device inputs feature A, feature B, feature C, and fresh food into the screening model to obtain a candidate score of 0.3 for candidate object 1, a candidate score of 0.8 for candidate object 2, and a candidate score of 0.9 for candidate object 3.

[0118] For the object features of each candidate object in the at least one candidate object, the object features of each candidate object include: object attribute features and object features, wherein the object attribute features include features of object color, object model, and other self attributes; and the object features include object price and object sales and other features.

[0119] Here, after obtaining the at least one target user, for each target user in the at least one target user, the information processing device outputs at least one target object corresponding to the target user to the target user.

[0120] In an example, the at least one target user includes: target user 1 and target user 2, and the at least one target object includes: target object 11, target object 12, and target object 13, and the information processing device outputs target object 11, target object 12, and target object 13 to target user 1 and target user 2.

[0121] In some embodiments, the information processing method provided by the embodiments of the present application further includes:

[0122] The information processing device obtains at least two second training data, each second training data in the at least two second training data includes: second training features and a second label corresponding to the second training features, the second training features include: description information of a training interest and object features of a training object, the second label is used to represent a matching degree of the training object and the training interest, and the second label is determined based on the object features of the training object; and the information processing device trains the initial product selection model according to the at least two second training data to obtain the product selection model.

[0123] Here, the object features of the training object include: object attribute features and object features. The object attribute features include features of object colors, object models, and other self attributes. The object features include object prices, object sales, and other features.

[0124] In some embodiments, the information processing device trains the initial product selection model according to the at least two second training data to obtain the product selection model, including:

[0125] The information processing device inputs the second training features in the at least two second training data into the initial product selection model to obtain second prediction scores. In the case that the initial product selection model does not converge, the information processing device determines a second loss value of the initial product selection model according to the second prediction scores corresponding to each second training data and the second labels of the at least two second training data. The information processing device updates the parameters of the initial product selection model based on the second loss value to obtain the product selection model, and continues to input the second training features in the at least two second training data into the initial product selection model until the initial product selection model converges to obtain the product selection model.

[0126] Here, the second prediction score is used to represent the matching degree of each candidate object in the at least one candidate object with the training interest.

[0127] In an example, the second prediction score is 0.9, which represents that the matching degree of the training object with the training interest is 90%.

[0128] In some embodiments, the product selection model at least includes an interaction feedback module, and the second features further include historical behavior features of each candidate user in the at least one candidate user. The historical behavior features are used to indicate at least one object operated in a historical time period. The method further includes: for each candidate user in the at least one candidate user, the information processing device inputs the historical behavior features of the candidate user into the interaction feedback module; for each candidate object in the at least one candidate object, the information processing device inputs the object portrait features of the candidate object into the interaction feedback module; based on the historical behavior features and the object portrait features of the candidate object, the information processing device determines the similarity between each object operated in the historical time period by the candidate user and each candidate object in the at least one candidate object.

[0129] Here, for the historical behavior features, the historical behavior features can be as shown in the following formula (4):

[0130]

[0131] wherein, Bc represents a historical behavior feature, t represents a historical time period, C1, C2, …, C n1 represents at least one object operating in the historical time period, wherein C1 represents a first object operating in the historical time period, C n1 represents an nth1 object operating in the historical time period.

[0132] After obtaining the historical behavior feature B c , the information processing device will perform tensor operation (embedding) on the historical behavior feature B c , and calculate Q (Query), K (key), and V (value) according to the embedded historical behavior feature B c , as shown in the following formulas (5) to (7):

[0133] Q = W Q *B c (Formula 5);

[0134] K = W K *B c (Formula 6);

[0135] V = W V *B c (Formula 7);

[0136] wherein W Q , W K , and W V are parameter matrices of the item selection model,

[0137] After calculating Q, K, and V, the formula for calculating self-attention Attention is as shown in the following formula (8):

[0138]

[0139] wherein n h represents the dimension of the embedding vector of Q, K, and V.

[0140] The calculation of the ith head is as shown in the following formula (9):

[0141] head i = Attention(W i Q Q, W i K K, W i V V) (Formula 9);

[0142] wherein W iQ , W i K and W i V are parameters of a multi-head-attention model.

[0143] The output vector F of the self-attention mechanism c is shown in the following formula (10):

[0144] F c = concat (head1, head2, …, head h ) * W O (formula 10) ;

[0145] wherein, concat is used to represent the connection of head1 to head n F c is a parameter of the multi-head-attention model.

[0146] After obtaining F c , the similarity f c can be calculated by Average-pooling, as shown in the following formula (11):

[0147] f c = Average-pooling (F c ) (formula 11) ;

[0148] wherein, Average-pooling can be understood as an average pooling calculation.

[0149] In some embodiments, the product selection model further comprises a feature optimization module, and the method further comprises: inputting the first feature and the second feature into the feature optimization module by the information processing device to obtain at least one optimized feature type.

[0150] Here, the at least one optimized feature type includes a first-order feature type, a second-order feature type, and a high-order feature type.

[0151] For the first-order feature type The first-order feature type is linear, and the calculation formula is shown in the following formula (12):

[0152]

[0153] wherein, is a parameter of the product selection model, x i is a feature in the first feature and the second feature, b i is a bias term.

[0154] For the second order feature type y FM , the calculation formula is shown in the following formula (13):

[0155]

[0156] Wherein, f i and f j are two different features.

[0157] For the high order feature f (i+1) , the calculation formula is shown in the following formula (14):

[0158] f (i+1) =ReLU(W( i) *f (i) +b (i) ) Formula (14);

[0159] Wherein, W (i ) represents a parameter, f (i) represents an operator, and b (i) represents a bias term.

[0160] With the development of big data and artificial intelligence, more and more users like online shopping. For e-commerce platforms, how to provide different gifts corresponding to different festivals to users according to different festivals to improve the efficiency of festival matching between people and goods is worth studying. Based on the above scenario, the embodiment of the application provides an information processing method, which can improve the efficiency of matching between people and goods.

[0161] In the embodiment of the application, after the users corresponding to each festival are selected, the selected users can be recommended interested goods by the product selection model.

[0162] For the product selection model, the embodiment of the application improves the deep learning network model to obtain the product selection model. The improvement of the deep learning network model can include: adding a deep interaction feedback network and a feature multi-level interaction network in the deep learning network model.

[0163] Here, based on the deep interaction feedback network, the product selection model can recommend the goods more matched to the user to the user according to the positive feedback or negative feedback of the user to a certain good, for example: the user's evaluation of the good A is full score evaluation, and the full score evaluation is positive feedback, the product selection model will recommend similar goods to the good A to the user according to the positive feedback, so that the product selection model recommends the goods more matched to the user to the user; after adding the feature multi-level interaction network, the product selection model can recommend the goods more matched to the user to the user according to the interaction between the features and the features.

[0164] The features input into the product selection model include: user behavior habits in different festivals and commodity portrait features. The user behavior habits are used to represent the commodities that the user habitually purchases in different festivals. The commodity portrait features include: commodity attribute features and commodity features. The commodity attribute features are used to represent the attributes of the commodities, such as: color, model, and other attributes of the commodities. The commodity features include: price, sales volume, and other features of the commodities.

[0165] Here, the behavior habits of different users in different festivals can be analyzed to obtain an analysis conclusion of the behavior habits of different users in different festivals. After obtaining the analysis conclusion of the behavior habits, the analysis conclusion of the behavior habits is input into the product selection model as a feature, so that the product selection model recommends commodities to the user according to the analysis conclusion of the behavior habits.

[0166] Taking a target category and a target festival as an example, as shown in Figure 4 The steps of selecting commodities for the user corresponding to the target festival by the product selection model include:

[0167] S401, design a commodity base pool corresponding to a festival scene.

[0168] According to business analysis, the commodities in the commodity base pool corresponding to each festival scene in different festival scenes are different.

[0169] Here, in the case of a festival scene being Valentine's Day, the commodities in the commodity base pool corresponding to the Valentine's Day include: chocolates, roses, and other love-expressing commodities.

[0170] S402, construct features input into the product selection model and labels of commodities.

[0171] Here, the features input into the product selection model include: behavior habits of a target user in different festivals and commodity portrait features.

[0172] For the labels of the commodities, the sales volume of the commodities is divided into at least one sales volume level, and each sales volume level in the at least one sales volume level corresponds to a label.

[0173] S403, improve the deep learning network model to obtain a product selection model.

[0174] S404, train the product selection model to generate the product selection model.

[0175] S405, determine online commodity new data.

[0176] Here, the commodity new data includes: behavior habits of a target user in different festivals and commodity portrait features.

[0177] S406, input the commodity new data into the product selection model to screen out commodities.

[0178] In the embodiments of the present application, since the circle user process and the product selection process have commonalities, the present application can recommend goods to the circled users after the users are circled, so as to link the circle user process and the product selection process, and a machine learning deep learning model system is designed, which includes a circle user model and a product selection model, to predict users and goods corresponding to different festival scenarios.

[0179] Next, taking the target category as a festival in a festival scenario as an example, the information processing method provided by the present application will be described.

[0180] For each festival scenario in the multiple festival scenarios, after the users in the festival scenario are selected, the users will be recommended the goods of interest to the users, i.e. potential goods. Among them, the goods with more search times and less purchases by the users in the festival scenario can be determined as potential goods.

[0181] The process of determining the potential goods of each festival category in the festival scenario is as shown in Figure 5

[0182] S500, determining potential goods in the festival category.

[0183] Here, for the festival category as a search category, when determining the potential goods in the search category, if the supply of goods in the search category is insufficient or the richness is poor, it is considered that the goods in the search category are potential goods.

[0184] For the search category with the number of returned goods greater than threshold 1, if there is no user behavior in the search category, it is considered that the supply of goods in the search category is insufficient. For the search category with the number of returned goods greater than threshold 1, if there is user behavior in the search category, if the search term has a brand term, it is determined whether the search term has conversion, if the search term has low conversion, it is considered that the price of the goods in the search category is too high or the goods are not rich; if the search term has no conversion, it is determined whether there is an order in the same secondary category, if there is an order in the secondary category, and the relevance between the search term and the search result is low, it is determined that the supply of goods in the search category is insufficient, if there is an order in the secondary category, and the brand in the click does not contain the search brand, it is considered that the supply of goods in the search category is insufficient.

[0185] For the search category with the number of goods less than threshold 2, if there is no user behavior in the search category, it is considered that the supply of goods in the search category is insufficient, if there is user behavior in the category, it is considered that the richness of the goods in the search category is poor.

[0186] It can be understood that Figure 5 The threshold 2 in the above formula is less than the threshold 1.

[0187] As​Figure 6 As shown, after obtaining the order table 601, the information processing device can determine the order user data 602 and order line data 603 based on the order table, and determine potential or abnormal products based on the order user data and order line data. The order user data includes: month-on-month data 604, year-on-year data 605, and absolute value (count) 606; the order line data includes: month-on-month data, year-on-year data, and count.

[0188] like Figure 7 As shown, after obtaining the order table 601, the information processing device can determine the order data 701, order line data 702, and average order line data per person 703 based on the order table. Then, based on this order data, order line data, and average order line data per person, it can train the automated customer segmentation model or the automated product selection model. The automated customer segmentation model is the screening model described in the above embodiments, and the automated product selection model is the product selection model described in the above embodiments.

[0189] like Figure 8 As shown, after obtaining the order table 601, the information processing device can determine user data 801, product data 802, and holiday search term data 803 based on the order table. It then analyzes the user data, product data, and holiday search term data to obtain user characteristics, product characteristics, and holiday search term characteristics. After obtaining these characteristics, it can train the automated user segmentation model or automated product selection model. Specifically, the user data includes: user gender data 8011, user age data 8012, and user geographic data 8013; the product data includes: primary category data 8021 and secondary and tertiary category data 8022; and the holiday search term data includes: brand term data, attribute term data, and category term data 8031.

[0190] In this embodiment of the application, the process for determining the commonalities and characteristics among different festivals is as follows: Figure 9 As shown:

[0191] For the holiday category 901, this category can include: Love category 9011, General category 9012, and Other category 9013. In one example, Love category 9011 can include: Valentine's Day, May 20th (520 Day), Qixi Festival, Christmas, etc. General category 9012 can include: New Year's Day, Spring Festival, Dragon Boat Festival, Mid-Autumn Festival, etc., and Other category 9013 can be any holiday other than Love category 9011 and General category 9012.

[0192] For the commonness and difference mining of the same festival 902, the mining can be performed from three angles of user festival portrait 9021, commodity portrait 9022 and search term analysis 9023. The user festival portrait 9021 includes information: user stickiness classification, user attribute analysis, user festival purchasing power prediction, wherein the user stickiness classification is started from the festival purchasing power frequency, and can include: festival-high-frequency purchasing user, festival-medium-frequency purchasing user, festival-low-frequency purchasing user. The user attribute analysis can include user label and user behavior, and the user label can include: gender, age range, region, micro-group type, etc. The user festival purchasing power prediction can divide the users into: core customer group, general customer group, potential customer group, etc. The commodity portrait 9022 can include: commodity distribution in search, commodity distribution in click and commodity distribution in order, wherein the commodity information of the commodity distribution in order can include: category, brand, shop, single product, etc. The search term analysis 9023 can include: search term nature analysis and TOP search term, wherein the TOP search term can be divided into the following categories: high search and high conversion of high conversion rate, high search and low conversion of low conversion rate, and high search and no conversion of no conversion rate.

[0193] For the commonness and difference mining of different festivals 903, the mining can be performed from two angles of data insight 9031 and difference solution 9032. The data insight 9031 can be analyzed from three aspects of people, goods and field. For people, such as: commonness and difference of user festival portrait under different festivals, for goods, such as: commonness and difference of commodity portrait under different festival categories. For field, such as: commonness and difference of search and push field performance under different festival categories. The difference solution 9032 can propose different solutions for different festival categories, such as: solution for love category, solution for general category and other solutions, etc.

[0194] As shown in Figure 10 The present application provides an information processing method, which includes: festival business core background module 1001, festival operation scheme module 1002, festival search term field analysis module 1003, festival automatic circle person improvement module 1004, festival automatic product selection improvement module 1005 and festival mining landing module 1006, which will be described in detail below.

[0195] For the festival business core background module 1001, the functions realized by the festival business core background module include: festival promotion differentiation 10011, improvement of people and goods matching efficiency 10012, festival search traffic attribution 10013 and user fine operation 10014.

[0196] Here, for the holiday promotion differentiation, for different holidays, such as: Qixi, Valentine's Day, Chongyang Festival and Christmas, etc. Due to the increment of user behavior more in the demand for gift giving, it will lead to the user having different purchasing power performance in the non-holiday scene, in the holiday scene, the average single of the related category has a greater improvement than the average single in the non-holiday scene.

[0197] For improving the efficiency of matching people and goods, the holiday business core background module will analyze the demand of users in different holiday scenarios, understand the supply and demand matching status, and thus realize the improvement of the efficiency of matching people and goods.

[0198] For holiday search traffic attribution, in the case that the search volume of a certain commodity is high in the holiday scene, but the conversion rate is low, the holiday business core background module will perform attribution analysis on the case of high search volume and low conversion rate, so as to determine whether there is a problem.

[0199] For user fine operation, the holiday business core background module can analyze the user portrait data to determine which type of user can bring more growth.

[0200] As shown in Figure 10 For the holiday operation scheme module 1002, the functions implemented by the holiday operation scheme model include: circle potential users 10021, solve user gift giving appeal 10022, holiday similar and different pain point analysis 10023 and holiday mining scheme overview 10024.

[0201] Here, for the circle potential users, the information processing device can classify the holidays according to the gift giving scene, respectively mine the user portrait features of the same type of holiday and different type of holiday, and circle out the key people with promotion potential. Among them, the key people with promotion potential circled out refers to the candidate crowd of the current holiday according to the user holiday attribute characteristics. For example: when circling people in Valentine's Day, the user's shopping dynamic line in love related holidays will be analyzed from the user's historical behavior data, the user's love holiday preference will be calculated, and further combined with the user's gender, age, occupation and purchasing power and other characteristics to circle people.

[0202] For solving user gift giving appeal, the information processing device can divide the search words into exclusive search words and general search words, wherein the exclusive search words are aimed at the scene that the user knows the gift to buy, and the general search words are aimed at the scene that the user does not know the gift to buy. Here, the information processing device will perform supply and demand analysis and holiday field dynamic line analysis to optimize the product selection.

[0203] For the festival similarities and differences pain point disassembly, the information processing device will be based on the commonness and difference between festivals, and the user portrait will be analyzed in depth for different festivals to determine the user analysis report under different festivals, and generate user tags. For the users with high, medium and low frequency of purchase in the festival scene, the information processing device will have targeted fine operation, among which the high-frequency users will increase the average unit price, the medium-frequency users will increase the purchase frequency, and the low-frequency users will increase the repurchase rate.

[0204] As shown in Figure 11 , the commonness and characteristics between different festivals can be determined from three dimensions of festival classification 1101, user characteristics 1102 and user purchasing power characteristics 1103.

[0205] As shown in Figure 10 , for the festival search word field analysis module 1003, the functions implemented by the festival search word field analysis module include: general search word analysis 10031, exclusive search word analysis 10032, search word classification design 10033 and festival field dynamic line analysis 10034.

[0206] For general search word analysis, in the case that the user does not know the gift to be selected, the user will search by general search words, which include: gift, gift to girlfriend and gift to teacher, etc. without specific object search words.

[0207] For exclusive search word analysis, in the case that the user knows the gift to be bought, the user will search by exclusive search words, which include: lipstick, perfume, etc. with specific object search words.

[0208] For search word classification design, the information processing device will increase some classification rules based on the existing classification according to the user's search.

[0209] For festival field dynamic line analysis, in the case that the user does not know the gift to be bought, the information processing device will perform dynamic line analysis in the search and push field. In the case that the user knows the gift to be bought, the information processing device will perform supply and demand analysis to provide more goods meeting the user's demand for the user.

[0210] In an example, for dynamic line analysis in the search and push field, in the case that the general search word is "gift to girlfriend", the information processing device will determine the cost-effective goods under "gift to girlfriend", and recommend the determined cost-effective goods to the user.

[0211] As shown in Figure 10As shown, for the holiday automation circle person improvement module 1004, the functions implemented by the holiday automation circle person improvement module include: determining Dense features and Identity Document (ID) features 10041, determining user portraits 10042, holiday multi-interest networks 10043, and holiday interest reinforcement 10044.

[0212] For the holiday automation circle person model, the features of the training data include: Dense features and ID features. Among them, the Dense features include: user portrait features, commodity portrait features, user and commodity interaction statistical features, and context features; the ID features include: user historical time window clicks, user added shopping, attention paid to commodities, categories, brands, and stores, etc.

[0213] For the label, if the user has ordered commodities in the historical holiday behavior data, the label is 1, and if the user does not have ordered commodities in the historical holiday behavior data, the label is 0.

[0214] The holiday automation circle person model is as shown in Figure 12

[0215] S1201, constructing a candidate pool of user portraits and user purchasing power.

[0216] S1202, obtaining Dense features and ID features.

[0217] S1203, embedding the Dense features and the ID features.

[0218] S1204, multi-interest extraction through a capsule network.

[0219] S1205, determining at least one target user.

[0220] For the user portrait, the user portrait includes: user stickiness classification, user attribute analysis, and user holiday purchasing power prediction.

[0221] For the holiday multi-interest network, the multi-interest network is used to extract the multi-interests of the user, and the multi-interest network is composed of three layers of Deep Neural Networks (DNN).

[0222] For the holiday interest reinforcement, in order to better learn the multi-interests of the user in the holiday, a Label-aware-attention mechanism is designed, which is represented as a u = Attention(e i ,V u ,V u ).​

[0223] For the holiday automation product selection improvement module 1005, the functions implemented by the holiday automation product selection improvement module include: determining holiday attribute goods 10051, determining holiday user purchase behavior 10052, holiday goods user deep interaction feedback 10053, holiday goods user feature interaction optimization 10054.

[0224] For the holiday automation product selection model, the features of the training data include: product image features, user holiday purchase behavior features, and holiday attribute features.

[0225] When screening holiday goods, different labels are designed for different holiday use scenarios. For example, when screening goods on Valentine's Day, goods with sales in historical love-related holidays are classified, and a label is designed for each level after classification.

[0226] The structure of the holiday automation product selection model is as shown in Figure 13

[0227] S1301, obtain holiday attribute features, user purchasing power features, and product features.

[0228] S1302, embedding the holiday attribute features, user purchasing power features, and product features.

[0229] S1303, product-user feature interaction optimization layer.

[0230] S1304, holiday product-user deep interaction feedback network layer.

[0231] For holiday attribute goods, the holiday attributes include love attributes and general family attributes. The love attribute holidays include Valentine's Day, Qixi Festival, etc.; the general family attribute holidays include Spring Festival, Mid-Autumn Festival, and Dragon Boat Festival, etc. Love attribute goods include flowers, chocolates, and necklace rings, etc. that can express love; general family attribute goods include couplets, mooncakes, and zongzi, etc.

[0232] For holiday user purchase behavior, during the holiday, users will be divided into high-frequency purchase users, medium-frequency purchase users, and low-frequency purchase users according to the frequency of purchasing goods. Different goods will be pushed to different types of users on the holiday.

[0233] For the holiday product-user deep interaction feedback module, as shown in the above formula (1), taking the user historical click behavior sequence as an example, the goods clicked by the user in the historical time period are obtained, and the goods clicked by the user in the historical time period are spliced to obtain Bc.

[0234] ​The target-side products are spliced ​​together to obtain the embedding of the target-side products.

[0235] like Figure 10 As shown, the holiday mining and implementation module 1006 has the following functions: holiday user stickiness operation 10061, holiday product supply and demand upgrade 10062, search and promotion holiday penetration enhancement 10063, and solving users' gift-giving needs 10064.

[0236] For holiday user engagement management, after categorizing users into high-frequency, medium-frequency, and low-frequency users, high-frequency users will be identified as core users, medium-frequency users as general users, and low-frequency users as potential users. Specifically, for core users, the goal is to increase average order value; for general users, the goal is to increase purchase frequency; and for potential users, the goal is to boost activity levels.

[0237] In response to the evolving supply and demand of holiday goods, information processing equipment will conduct supply and demand analysis to provide users with more products that meet their needs.

[0238] To enhance search engine penetration during holidays, information processing devices will start with search terms and analyze users' subsequent actions to provide users with more products that meet their needs.

[0239] To address users' gift-giving needs, information processing equipment will determine users' actual requirements, understand the current supply and demand of holiday goods, and improve the efficiency of matching people and goods, thereby helping users better resolve their gift-giving needs.

[0240] This application provides a storage medium, namely a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the information processing method provided in the above embodiments.

[0241] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned information processing method. The electronic device is an information processing apparatus.

[0242] Figure 14 An information processing apparatus provided in the embodiments of this application, such as Figure 14 As shown, the information processing device 1400 includes:

[0243] The first determining unit 1401 is used for...

[0244] input the first feature into the screening model to obtain a candidate score of each candidate user in the at least one candidate user, the first feature at least comprising: description information of the target category, user feature of each candidate user in the at least one candidate user, the candidate score of the candidate user being used to represent the degree of interest of the candidate user corresponding to the candidate score in the target category;

[0245] The second determination unit 1402 is configured to determine at least one target user from the at least one candidate user according to the candidate score of each candidate user in the at least one candidate user.

[0246] The output unit 1403 is configured to output the object corresponding to the target category to the at least one target user.

[0247] In some embodiments, the screening model at least comprises an interest determination subnetwork, and the second determination unit 1402 is configured to:

[0248] For each candidate user in the at least one candidate user, input the user feature of the candidate user into the interest determination subnetwork to obtain interest information of the candidate user, the interest information representing at least two interests of the candidate user.

[0249] Determine the candidate score of the candidate user whose at least two interests include the target category in the at least one candidate user.

[0250] In some embodiments, the information processing apparatus 1400 further comprises a training unit configured to:

[0251] Obtain at least two pieces of first training data, each piece of the at least two pieces of first training data comprising: first training feature and first label corresponding to the first training feature, the first training feature comprising: description information of training interest, user feature of training user, the first label being used to represent the degree of interest of the training user in the training interest, the first label being determined based on historical behavior of the training user;

[0252] Train the initial screening model according to the at least two pieces of first training data to obtain the screening model.

[0253] In some embodiments, the training unit is further configured to:

[0254] Input the first training feature in the at least two pieces of first training data into the initial screening model to obtain a first prediction score;

[0255] In a case where the initial screening model does not converge, a first loss value of the initial screening model is determined according to a first prediction score corresponding to each of the at least two pieces of first training data and the first label of the first training data;

[0256] The parameters of the initial screening model are updated based on the first loss value to obtain the screening model, and the first training feature in the at least two pieces of first training data is continuously input into the initial screening model until the initial screening model converges, so as to obtain the screening model.

[0257] In some embodiments, the training unit is further configured to:

[0258] A first loss value of the initial screening model is determined according to a first prediction score corresponding to each of the at least two pieces of training data, the first label of the training data, and a weight corresponding to the first label.

[0259] In some embodiments, the training unit is further configured to:

[0260] The first label used to represent the degree of interest of the training user in the training interest is determined, the user feature of the training user, the user interaction object, and the interested object in the user interaction object corresponding to the training interest are determined, and the weight corresponding to the first label is determined.

[0261] In some embodiments, the training unit is further configured to:

[0262] The second feature is input into the product selection model to obtain a candidate score of each candidate object in the at least one candidate object, the second feature at least including: description information of the target category, object feature of each candidate object in the at least one candidate object, and the candidate score of the candidate object, the candidate score being used to represent the matching degree between the candidate object corresponding to the candidate score and the target category;

[0263] At least one target object is determined from the at least one candidate object according to the candidate score of each candidate object in the at least one candidate object.

[0264] Correspondingly, the object corresponding to the target category is output to the at least one target user, including:

[0265] The at least one target object is output to the at least one target user.

[0266] In some embodiments, the training unit is further configured to:

[0267] obtaining at least two pieces of second training data, each of the at least two pieces of second training data comprising: a second training feature and a second label corresponding to the second training feature, the second training feature comprising: description information of a training interest, and object features of a training object, the second label being used to represent a matching degree between the training object and the training interest, and the second label being determined based on the object features of the training object;

[0268] training the initial product selection model according to the at least two pieces of second training data, to obtain the product selection model.

[0269] In some embodiments, the training unit is further configured to:

[0270] inputting the second training feature in each of the at least two pieces of second training data into the initial product selection model, to obtain a second predicted score;

[0271] in a case where the initial product selection model does not converge, determining a second loss value of the initial product selection model according to the second predicted score corresponding to each of the at least two pieces of second training data and the second label of the second training data;

[0272] updating parameters of the initial product selection model based on the second loss value, to obtain the product selection model, and continuing to input the second training feature in each of the at least two pieces of second training data into the initial product selection model, until the initial product selection model converges, to obtain the product selection model.

[0273] It should be noted that the data processing system provided in the embodiments of the present application includes various logical units included therein, and can be implemented by a processor in an electronic device; of course, it can also be implemented by a specific logical circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0274] The above description of the system embodiments is similar to the description of the method embodiments described above, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the system embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0275] It should be noted that, in the embodiments of the present application, if the page display method described above is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product in essence or the part that contributes to the related art, and the computer software product is stored in a storage medium, including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the methods described in the embodiments of the present application. The storage medium mentioned above includes: a U disk, a mobile hard disk, a read-only memory (Read Only Memory, ROM), a magnetic disk or an optical disk, and various media that can store program codes. Thus, the embodiments of the present application are not limited to any specific hardware and software combination.

[0276] It should be noted that: the description of the above storage medium embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0277] It should be noted that, Figure 15 A hardware entity diagram of an electronic device according to an embodiment of the present application is shown in FIG. 15. As shown in FIG. 15, the electronic device 1500 includes a processor 1501, at least one communication bus 1502, at least one external communication interface 1504, and a memory 1505. The communication bus 1502 is configured to realize the connection and communication between the components. In an example, the electronic device 1500 further includes a user interface 1503, wherein the user interface 1503 can include a display screen, and the external communication interface 1504 can include a standard wired interface and a wireless interface. Figure 15

[0278] The memory 1505 is configured to store instructions and applications executable by the processor 1501, and can also cache data to be processed by the processor 1501 and data to be processed or having been processed by each module in the electronic device (for example, image data, audio data, voice communication data, and video communication data), which can be realized by a flash (FLASH) or a random access memory (Random Access Memory, RAM).

[0279] ​It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0280] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0281] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0282] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0283] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0284] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, the foregoing program can be stored in a computer readable storage medium, and the program executes the steps of the above-mentioned method embodiments when executed; and the foregoing storage medium includes a mobile storage device, a read only memory (ROM), a magnetic disc or an optical disc and various storage medium capable of storing program codes.

[0285] Alternatively, the integrated units of the present application can be stored in a computer readable storage medium if the integrated units are realized in the form of software function modules and sold or used as independent products. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of software product, and the computer software product is stored in a storage medium, includes a plurality of instructions to make a computer device (which can be a personal computer, a server or a network device) execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes a mobile storage device, a ROM, a magnetic disc or an optical disc and various storage medium capable of storing program codes.

[0286] The above is only a preferred embodiment of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0287] The above is only a preferred embodiment of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An information processing method, characterized in that, The method includes: The first feature is input into the screening model to obtain the candidate scores of each candidate user among at least one candidate user. The first feature includes at least: descriptive information of the target category and user features of each candidate user among the at least one candidate user. The candidate scores of the candidate users are used to characterize the degree of interest of the candidate user corresponding to the candidate score in the target category. Based on the candidate scores of each candidate user among the at least one candidate user, at least one target user is determined from the at least one candidate user. Output the object corresponding to the target category to the at least one target user; The screening model includes at least an interest determination subnetwork, wherein the first feature includes historical features of candidate users, and the step of inputting the first feature into the screening model to obtain candidate scores for each candidate user among at least one candidate user includes: For each of the at least one candidate users, the historical features of the candidate user are input into the interest determination subnetwork to obtain the interest information of the candidate user, and the interest information represents at least two interests of the candidate user. Among the at least one candidate user, at least two candidate users whose interests include the target category are determined, and candidate ratings of the at least two candidate users whose interests include the target category are determined.

2. The method according to claim 1, characterized in that, The method further includes: At least two sets of first training data are obtained. Each set of first training data includes: a first training feature and a first label corresponding to the first training feature. The first training feature includes: descriptive information of training interest and user characteristics of training user. The first label is used to characterize the degree of interest of the training user in the training interest. The first label is determined based on the historical behavior of the training user. The initial screening model is trained based on the at least two first training data points to obtain the screening model.

3. The method according to claim 2, characterized in that, The step of training the initial screening model based on the at least two first training data points to obtain the screening model includes: The first training feature from the at least two first training data points is input into the initial screening model to obtain the first prediction score; If the initial screening model fails to converge, the first loss value of the initial screening model is determined based on the first prediction score corresponding to each of the at least two first training data and the first label of the first training data. The parameters of the initial screening model are updated based on the first loss value to obtain the screening model. The first training features from the at least two first training data are then input into the initial screening model until the initial screening model converges to obtain the screening model.

4. The method according to claim 3, characterized in that, The step of determining the first loss value of the initial screening model based on the first predicted score corresponding to each of the at least two first training data points and the first label of the first training data includes: The first loss value of the initial screening model is determined based on the first predicted score corresponding to each of the at least two training data sets, the first label of the training data set, and the weight corresponding to the first label.

5. The method according to claim 4, characterized in that, The method further includes: For the first label used to characterize the degree of interest of the training user in the training interest, determine the user characteristics of the training user, the user interaction object, and the interest object corresponding to the training interest in the user interaction object, and determine the weight corresponding to the first label.

6. The method according to claim 1, characterized in that, The method further includes: The second feature is input into the product selection model to obtain the candidate scores of each candidate object in at least one candidate object. The second feature includes at least: the description information of the target category and the object features of each candidate object in at least one candidate object. The candidate scores of the candidate objects are used to characterize the degree of matching between the candidate object corresponding to the candidate score and the target category. Based on the candidate scores of each candidate object in the at least one candidate object, at least one target object is determined from the at least one candidate object; Correspondingly, outputting the object corresponding to the target category to the at least one target user includes: Output the at least one target object to the at least one target user.

7. The method according to claim 6, characterized in that, The method further includes: At least two sets of second training data are obtained. Each set of at least two sets of second training data includes: a second training feature and a second label corresponding to the second training feature. The second training feature includes: descriptive information of the training interest and object features of the training object. The second label is used to characterize the degree of matching between the training object and the training interest. The second label is determined based on the object features of the training object. The initial product selection model is trained based on the at least two second training data to obtain the product selection model.

8. The method according to claim 7, characterized in that, The initial product selection model is trained based on the at least two sets of second training data to obtain the product selection model, including: The second training feature from the at least two second training data points is input into the initial product selection model to obtain the second prediction score; If the initial product selection model fails to converge, a second loss value for the initial product selection model is determined based on the second prediction score corresponding to each of the at least two second training data points and the second label of the second training data. The parameters of the initial product selection model are updated based on the second loss value to obtain the product selection model. The second training features from the at least two second training data are then input into the initial product selection model until the initial product selection model converges to obtain the product selection model.

9. An information processing device, characterized in that, The device includes: The first determining unit is used to input a first feature into a screening model to obtain candidate scores for each candidate user among at least one candidate user. The first feature includes at least: descriptive information of the target category and user features of each candidate user among the at least one candidate user. The candidate scores of the candidate users are used to characterize the degree of interest of the candidate user corresponding to the candidate score in the target category. The second determining unit is used to determine at least one target user from the at least one candidate user based on the candidate scores of each candidate user among the at least one candidate user; An output unit is used to output the object corresponding to the target category to the at least one target user; The screening model includes at least: an interest determination subnetwork, the first feature including historical features of candidate users, and the first determination unit, which is further configured to, for each of the at least one candidate users, input the historical features of the candidate user into the interest determination subnetwork to obtain the interest information of the candidate user, the interest information representing at least two interests of the candidate user; determine that at least two of the at least one candidate user's interests include the target category, and determine the candidate ratings of the candidate users whose interests include the target category.

10. A storage medium storing a computer program, which, when executed by a processor, performs the information processing method according to any one of claims 1 to 8.

Images