Transaction information recommendation method and device, storage medium, product and electronic equipment

Abstract

The application discloses a transaction information recommendation method and device, a storage medium, a product and an electronic device. The method comprises the following steps: acquiring a time period preference model, the time period preference model being used for calculating a preference score of a user for transaction information in different time periods; obtaining a time effectiveness preference vector of a target user based on a user feature of the target user and a time effectiveness feature of exposed transaction information, the exposed transaction information being transaction information exposed to the target user; performing parameter modulation processing on the time period preference model by using the time effectiveness preference vector to obtain a target preference model; performing preference prediction processing on the target user by using the target preference model based on the user feature and a content feature of target transaction information to obtain a preference score of the target user for the target transaction information; and determining a recommendation strategy of the target transaction information for the target user based on the preference score.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, storage medium, product, and electronic device for recommending transaction information. Background Technology

[0002] Existing technologies can use historical data generated from users' browsing behavior of advertisements, news, articles and other information on the Internet, and predict users' preferences based on their personal characteristics, thereby pushing information that users may like to users. However, existing technologies can only profile users based on their personal characteristics, but users' preferences for information often vary at different times. Existing technologies cannot predict user preferences over time, so a method that can predict user preferences at different times is needed. Summary of the Invention

[0003] This application provides a method, apparatus, storage medium, product, and electronic device for recommending transaction information. By combining user preferences for transaction information at different times with the timeliness of the transaction information, a user preference score is calculated, improving the accuracy and comprehensiveness of the preference score, thereby enhancing the rationality of the transaction information recommendation. The technical solution is as follows:

[0004] In a first aspect, embodiments of this application provide a method for recommending transaction information, the method comprising:

[0005] Obtain a time period preference model, which is used to calculate the user's preference score for transaction information in different time periods;

[0006] Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference vector of the target user is obtained, wherein the exposed transaction information is the transaction information exposed to the target user;

[0007] The time-sensitivity preference vector is used to modulate the parameters of the time-period preference model to obtain the target preference model;

[0008] Based on the user characteristics and the content characteristics of the target transaction information, the target preference model is used to perform preference prediction processing on the target user to obtain the target user's preference score for the target transaction information.

[0009] Based on the preference score, a recommendation strategy for the target transaction information of the target user is determined.

[0010] Secondly, embodiments of this application provide a method for training a time-period preference model, the method comprising:

[0011] An initial preference model is created, and the historical access dataset of the sample users is obtained. The historical access dataset includes the user characteristics of the sample users, the sample exposure transaction information corresponding to the sample users, and the click tags of the user in relation to the sample exposure transaction information. The sample exposure transaction information is the transaction information exposed to the sample users.

[0012] Based on the historical access dataset, the initial preference model is trained for at least one round to obtain the sample user's sample preference score for the sample exposure transaction information;

[0013] The parameters of the initial preference model are adjusted based on the sample preference scores and the click tags until the initial preference model completes model training, thus obtaining the time-period preference model.

[0014] Thirdly, embodiments of this application provide a method for training a timeliness preference model, the method comprising:

[0015] Create an initial timeliness model to obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users;

[0016] Based on the user characteristics and timeliness characteristics of the sample users, the initial timeliness model is trained for at least one round to obtain the sample timeliness preference vector corresponding to the sample users;

[0017] Based on the sample timeliness preference vector and the preset timeliness preference, the parameters of the initial timeliness model are adjusted until the initial timeliness model completes model training, and a timeliness preference model is obtained.

[0018] Fourthly, embodiments of this application provide a method for training a parameter modulation model, the method comprising:

[0019] Create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector;

[0020] Based on the sample timeliness preference vector and the sample cluster center vector, the initial modulation model is trained for at least one round to obtain the sample modulation input parameters corresponding to the sample user.

[0021] Based on the sample modulation input parameters, the time period preference model is subjected to parameter modulation processing to obtain the sample preference model corresponding to the sample user. Based on the sample preference model, the sample user is subjected to preference prediction processing to obtain the sample user's sample timeliness preference score for sample exposure transaction information.

[0022] Based on the sample timeliness preference score and the click tags of the sample users for the sample exposure transaction information, the parameters of the initial modulation model are adjusted until the initial modulation model completes model training and a parameter modulation model is obtained.

[0023] Fifthly, embodiments of this application provide a transaction information recommendation device, the device comprising:

[0024] The model acquisition module is used to acquire the time period preference model, which is used to calculate the user's preference score for transaction information in different time periods;

[0025] The timeliness preference calculation module is used to obtain the timeliness preference vector of the target user based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, wherein the exposed transaction information is the transaction information exposed to the target user;

[0026] The parameter modulation module is used to perform parameter modulation processing on the time period preference model using the timeliness preference vector to obtain the target preference model;

[0027] The preference prediction module is used to perform preference prediction processing on the target user based on the user characteristics and the content characteristics of the target transaction information, using the target preference model to obtain the target user's preference score for the target transaction information;

[0028] The recommendation strategy determination module is used to determine a recommendation strategy for the target transaction information of the target user based on the preference score.

[0029] Sixthly, embodiments of this application provide a time-period preference model training apparatus, the apparatus comprising:

[0030] The first model creation module is used to create an initial preference model and obtain the historical access dataset of the sample users. The historical access dataset includes the user characteristics of the sample users, the sample exposure transaction information corresponding to the sample users, and the click tags of the user in relation to the sample exposure transaction information. The sample exposure transaction information is the transaction information exposed to the sample users.

[0031] The first model training module is used to perform at least one round of model training on the initial preference model based on the historical access dataset to obtain the sample user's sample preference score for the sample exposure transaction information.

[0032] The first model acquisition module is used to adjust the parameters of the initial preference model based on the sample preference score and the click tag until the initial preference model completes model training and obtains the time period preference model.

[0033] Seventhly, embodiments of this application provide a timeliness preference model training apparatus, the apparatus comprising:

[0034] The second model creation module is used to create an initial timeliness model, obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users;

[0035] The second model training module is used to perform at least one round of model training on the initial timeliness model based on the user characteristics and timeliness characteristics of the sample users, so as to obtain the sample timeliness preference vector corresponding to the sample users.

[0036] The second model acquisition module is used to adjust the parameters of the initial timeliness model based on the sample timeliness preference vector and the preset timeliness preference until the initial timeliness model completes model training and obtains the timeliness preference model.

[0037] Eighthly, embodiments of this application provide a parameter modulation model training apparatus, the apparatus comprising:

[0038] The third model creation module is used to create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector;

[0039] The third model training module is used to perform at least one round of model training on the initial modulation model based on the sample timeliness preference vector and the sample cluster center vector to obtain the sample modulation input parameters corresponding to the sample user.

[0040] The model modulation module is used to perform parameter modulation processing on the time period preference model based on the sample modulation input parameters to obtain the sample preference model corresponding to the sample user, and to perform preference prediction processing on the sample user based on the sample preference model to obtain the sample user's sample timeliness preference score for sample exposure transaction information.

[0041] The third model acquisition module is used to adjust the parameters of the initial modulation model based on the sample timeliness preference score and the click tags of the sample users for the sample exposure transaction information, until the initial modulation model completes model training and obtains the parameter modulation model.

[0042] Ninthly, embodiments of this application provide a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the above-described method steps.

[0043] In a tenth aspect, embodiments of this application provide a computer program product that stores a plurality of instructions adapted for loading by a processor and executing the above-described method steps.

[0044] Eleventhly, embodiments of this application provide an electronic device, which may include: a processor and a memory; wherein the memory stores a computer program, the computer program being adapted to be loaded by the processor and to execute the above-described method steps.

[0045] In one or more embodiments of this application, a time-segment preference model is obtained. This model is used to calculate a user's preference score for transaction information at different time periods. Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference vector for the target user is obtained. The exposed transaction information refers to the transaction information exposed to the target user. The timeliness preference vector is used to perform parameter modulation processing on the time-segment preference model to obtain a target preference model. Based on the user characteristics and the content characteristics of the target transaction information, the target preference model is used to perform preference prediction processing on the target user to obtain a preference score for the target transaction information. Based on the preference score, a recommendation strategy for the target transaction information for the target user is determined. By combining the user's preferences for transaction information at different time periods and the timeliness information of the transaction information to calculate the user's preference score for the transaction information, the accuracy and comprehensiveness of the preference score are improved, thereby improving the rationality of the transaction information recommendation. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is an example diagram illustrating the exposure of transaction information provided in an embodiment of this application;

[0048] Figure 2 This is a schematic diagram illustrating an example of preference prediction processing provided in an embodiment of this application;

[0049] Figure 3 This is a flowchart illustrating a transaction information recommendation method provided in an embodiment of this application;

[0050] Figure 4 This is a flowchart illustrating a transaction information recommendation method provided in an embodiment of this application;

[0051] Figure 5This is a flowchart illustrating a time-period preference model training method provided in an embodiment of this application;

[0052] Figure 6 This is a flowchart illustrating a timeliness preference model training method provided in an embodiment of this application;

[0053] Figure 7 This is a flowchart illustrating a parameter modulation model training method provided in an embodiment of this application;

[0054] Figure 8 This is a schematic diagram of the structure of a transaction information recommendation device provided in an embodiment of this application;

[0055] Figure 9 This is a schematic diagram of the structure of a time-period preference model device provided in an embodiment of this application;

[0056] Figure 10 This is a schematic diagram of the structure of a timeliness preference model device provided in an embodiment of this application;

[0057] Figure 11 This is a schematic diagram of the structure of a parameter modulation model device provided in an embodiment of this application;

[0058] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] Users can browse information about events online. This information can include articles, news, advertisements, etc., and is displayed on the user's browsing interface. Clicking a link corresponding to an event will take the user to the main content page, which displays the event's content. For example, when a user uses a mobile device to browse event information through an application, they may see the event "News A" on the application interface. Clicking the "News A" link will take them to the corresponding main content page, which displays the specific content of "News A." It's understandable that not all event information exposed to users will be clicked and viewed. Different users have different preferences for event information. Therefore, the event information exposed to users—those displayed on their browsing interface—is not necessarily the information that users click and view. Please also refer to [link / reference needed]. Figure 1 This embodiment of the application provides an example diagram illustrating the exposure of transaction information. Users can browse the transaction information using an application on a terminal device, such as... Figure 1 The image shown represents a browsing interface a user is currently viewing. The interface displays "News A," "News B," and "News C." All three news items ("News A," "News B," and "News C") are currently being viewed by the user, representing information exposed to the user. It's understandable that as the user scrolls and browses the interface, more information will be exposed, generating even more news items. However, the user may not necessarily click on all of these exposed news items to view. Figure 1 As shown, if a user is interested in "News A", they can click the link corresponding to "News A". The terminal device's interface can then jump to the main text of "News A" to display the relevant information.

[0061] Understandably, to improve the user's browsing experience of transaction information, it is necessary to recommend transaction information that users are more interested in and prefer. Therefore, a transaction information recommendation device can be used to perform preference prediction processing on the target user, thereby obtaining the target user's preference score for the target transaction information. The target user can be any user, and the target transaction information can be any transaction information. The preference score represents the target user's degree of interest and preference for the target transaction information. A higher preference score indicates a higher degree of preference for the target transaction information, and if the target transaction information is exposed to the target user, the probability of the target user clicking on it is greater. Conversely, a lower preference score indicates a lower degree of preference for the target transaction information, and if the target transaction information is exposed to the target user, the probability of the target user clicking on it is smaller. The preference score can range from 0 to 1. A preference score of 1 indicates that the target user will definitely click on the target transaction information, and a preference score of 0 indicates that the target user will definitely not click on the target transaction information. The transaction information recommendation method provided in this application embodiment can be implemented using a computer program and can run on a transaction information recommendation device based on the von Neumann architecture. This computer program can be integrated into applications or run as a standalone utility application.

[0062] Please see also Figure 2This embodiment of the present application provides an example of preference prediction processing. The time-period preference model training device can train an initial preference model to obtain a time-period preference model based on the user characteristics of the sample user, the sample exposure transaction information corresponding to the sample user, and the click tags of the sample exposure transaction information of the sample user. The time-period preference model can calculate the user's preference score for transaction information in different time periods. For example, the user may have different preference scores for the same transaction information in different time periods. For example, the transaction information may be a "breakfast recommendation" article. The user's preference score for the "breakfast recommendation" article in the morning will be higher than that in the evening. Among them, the user characteristics of the sample users are the feature information that can reflect the attributes of the sample users, such as the age, gender, place of origin or residence of the sample users, etc. The sample exposure transaction information corresponding to the sample users is the transaction information exposed to the sample users. The click tag of the sample user on the sample exposure transaction information is used to indicate whether the sample user clicked on the sample exposure transaction information, and also to indicate the user's preference score for the sample exposure transaction information. If the user clicks on the sample exposure transaction information, it means that the user confirms that the user is interested in the sample exposure transaction information, and the click tag is 1; otherwise, if the user does not click on the sample exposure transaction information, the click tag is 0.

[0063] It is understandable that users not only have different preference scores for the same transaction information at different times, but also for the same content with different timeliness. Therefore, the transaction information recommendation device can also combine the target user's user characteristics and the timeliness characteristics of the exposed transaction information to obtain the target user's timeliness preference vector. The timeliness characteristics are the feature information reflecting the timeliness of the exposed transaction information, such as the publication time, current time, and duration of publication. The timeliness preference vector is the feature vector reflecting the target user's preference for different timeliness exposed transaction vectors. Based on the target user's timeliness preference vector, the transaction information recommendation device can use a parameter modulation model to obtain the modulation input parameters of each layer of the time period preference model. Then, it can use the modulation input parameters to perform parameter modulation processing on the time period preference model to obtain the target preference model. The target preference model is then used to perform preference prediction processing on the target user to obtain the target user's preference score for the target transaction information. This preference score combines the user's preferences at different times and the user's preferences for different timeliness of transaction information, improving the accuracy and comprehensiveness of the preference score. The parameter modulation model is obtained by training the initial modulation model using a parameter modulation model training device. The timeliness preference vector of the target user is input into the parameter modulation model, and the parameter modulation model can output the modulation input parameters for each layer of the time period preference model. The modulation input parameters are used to adjust the parameters of each layer of the time period preference model.

[0064] Among them, the parameter modulation model, time period preference model, timeliness preference model, and target preference model are all deep neural network (DNN) models. The time period preference model training device is used to train the initial preference model to obtain the time period preference model; the timeliness preference model training device is used to train the initial timeliness model to obtain the timeliness preference model; and the parameter modulation model training device is used to train the initial modulation model to obtain the parameter modulation model. The time period preference model training device, the timeliness preference model training device, and the parameter modulation model training device can be modules in the transaction information recommendation device used to implement the model training method, or they can be the same device as the transaction information recommendation device.

[0065] The transaction information recommendation method provided in this application will be described in detail below with reference to specific embodiments.

[0066] Please see Figure 3 This is a flowchart illustrating a transaction information recommendation method provided in an embodiment of this application. Figure 3 As shown, the method described in this application embodiment may include the following steps S101-S110.

[0067] S102, Obtain the time period preference model.

[0068] Specifically, the time-period preference model training device can train the initial preference model based on the historical access dataset of sample users to obtain the time-period preference model. The historical access dataset can include the user characteristics of the sample users, the sample exposure transaction information corresponding to the sample users, and the click tags of the sample exposure transaction information of the sample users. Since the time-period preference model training device can divide the historical access dataset according to time intervals to obtain at least one time interval dataset, and then use the time interval datasets in sequence to train the initial preference model, the time-period preference model can be used to calculate the user's preference score for transaction information in different time periods. That is, the preference score obtained by the time-period preference model combines the user's preference degree for transaction information in different time periods.

[0069] S104. Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, obtain the timeliness preference vector of the target user.

[0070] Specifically, the transaction information recommendation device can obtain the target user's timeliness preference vector based on the target user's user characteristics and the timeliness characteristics of the exposed transaction information. Here, the exposed transaction information is the transaction information exposed to the target user, the timeliness characteristics are the feature information reflecting the timeliness of the transaction information, such as the publication time and duration of publication of the exposed transaction information, etc., and the timeliness preference vector is a feature vector used to reflect the target user's preference for different timeliness exposure transaction vectors. For example, the timeliness preference vector can reflect whether the target user prefers exposed transaction information with a longer publication duration or exposed transaction information with a shorter publication duration.

[0071] For example, a transaction information recommendation device can input user characteristics and timeliness characteristics into a timeliness preference model, which can output a timeliness preference vector for the target user. The timeliness preference model is used to calculate the timeliness preference vector based on user characteristics and timeliness characteristics.

[0072] S106, the time-sensitive preference vector is used to modulate the parameters of the time-period preference model to obtain the target preference model.

[0073] Specifically, the transaction information recommendation device can use a time-sensitivity preference vector to modulate the parameters of the time-segment preference model to obtain the target preference model. Understandably, the time-segment preference model calculates preference parameters by combining the user's preference for transaction information at different times. For example, the time-segment preference model can distinguish between different user preferences for the same transaction information in the morning and evening. However, it cannot distinguish between user preferences for different time-sensitivity transactions. By using a time-sensitivity preference vector to modulate the parameters of the time-segment preference model, the transaction information recommendation device enables the target preference model to also learn to distinguish user preferences for different time-sensitivity transactions. Therefore, the preference score calculated by the target preference model can simultaneously consider the user's preference for transaction information at different times, as well as the preference for different time-sensitivity transactions.

[0074] S108. Based on user characteristics and the content characteristics of target transaction information, a target preference model is used to perform preference prediction processing on target users to obtain the target user's preference score for target transaction information.

[0075] Specifically, the transaction information recommendation device can perform preference prediction processing based on user characteristics and the content characteristics of the target transaction information using a target preference model. This involves inputting user characteristics and content characteristics into the target preference model for preference prediction, and the target preference model can output a preference score for the target user regarding the target transaction information. Here, the content characteristics are feature vectors that reflect the specific content of the transaction information.

[0076] S110, Based on preference scores, determine the recommendation strategy for target transaction information for the target user.

[0077] Specifically, the preference score reflects the target user's degree of preference for the target information. A higher preference score indicates a higher degree of preference for the target information, and if the target information is exposed to the target user, the probability of the target user clicking on it is greater. Conversely, a lower preference score indicates a lower degree of preference for the target information, and if the target information is exposed to the target user, the probability of the target user clicking on it is smaller. Therefore, the information recommendation device can determine the recommendation strategy for the target information based on the preference score.

[0078] The recommendation strategy can be to decide whether to recommend target transaction information to the target user. For example, if the preference score is less than the recommendation score threshold, the recommendation strategy is not to recommend the target transaction information to the target user. If the preference score is greater than or equal to the recommendation score threshold, the recommendation strategy is to recommend the target transaction information to the target user, thus exposing the target transaction information to the target user. The recommendation score threshold is used to determine whether to recommend the target transaction information to the target user. It can be the initial setting of the transaction information recommendation device or set by relevant staff or users. For example, it can be 0.5.

[0079] Understandably, recommendation strategies can also include methods of recommending target transaction information to target users. These methods can be pop-up windows, where a pop-up window corresponding to the target transaction information appears on the user's browsing interface, and the user can jump to the main text of the target transaction information by clicking the pop-up window. Alternatively, the recommendation method can be to make the target transaction information appear as the next transaction information that the target user will see when they swipe through the browsing interface. The recommendation method can also be related to preference scores.

[0080] In this embodiment, a time-segment preference model is obtained. This model calculates a user's preference score for transaction information at different time periods. Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference vector for the target user is obtained. The exposed transaction information refers to the transaction information exposed to the target user. The timeliness preference vector is used to modulate the parameters of the time-segment preference model to obtain a target preference model. Based on the user characteristics and the content characteristics of the target transaction information, the target preference model is used to predict the target user's preferences, obtaining a preference score for the target transaction information. Based on the preference score, a recommendation strategy for the target transaction information for the target user is determined. By combining the user's preferences for transaction information at different time periods with the timeliness information of the transaction information, the user's preference score for the transaction information is calculated, improving the accuracy and comprehensiveness of the preference score, thereby improving the rationality of the transaction information recommendation.

[0081] Please see Figure 4 This is a flowchart illustrating a transaction information recommendation method provided in an embodiment of this application. Figure 4 As shown, the method described in this application embodiment may include the following steps S202-S216.

[0082] S202, Obtain the time period preference model.

[0083] Specifically, the time-period preference model training device can train the initial preference model based on the historical access dataset of sample users to obtain the time-period preference model. The historical access dataset can include the user characteristics of the sample users, the sample exposure transaction information corresponding to the sample users, and the click tags of the sample exposure transaction information of the sample users. Since the time-period preference model training device can divide the historical access dataset according to time intervals to obtain at least one time interval dataset, and then use the time interval datasets in sequence to train the initial preference model, the time-period preference model can be used to calculate the user's preference score for transaction information in different time periods. That is, the preference score obtained by the time-period preference model combines the user's preference degree for transaction information in different time periods.

[0084] Optional, T u，t The dataset for user u within time interval t can be generated using the following formula:

[0085]

[0086] Among them, user u has a total of N within the time interval t. t Information on exposed transactions, i k This refers to the exposure of transaction information, where k ranges from 1 to N. t Integers within the range For user u within time interval t, the information i of the exposed transaction is... k Click the tag. It can be a time-period preference model, which can calculate the preference score of a target user for target transaction information within a time interval t.

[0087] S204, Obtain the user characteristics of the target user and the timeliness characteristics of the exposed transaction information.

[0088] Specifically, the transaction information recommendation device can obtain the target user's timeliness preference vector based on the target user's user characteristics and the timeliness characteristics of the exposed transaction information. Here, the exposed transaction information is the transaction information exposed to the target user, and the timeliness characteristics are the feature information that reflects the timeliness of the transaction information, such as the publication time and duration of the exposed transaction information.

[0089] S206. Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference model is used to extract the preference features of the target user and obtain the timeliness preference vector of the target user.

[0090] Specifically, the transaction information recommendation device can extract the target user's preference features based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, and use a timeliness preference model to obtain the target user's timeliness preference vector. That is, the transaction information recommendation device can input the target user's user characteristics and the timeliness characteristics of the exposed transaction information into the timeliness preference model, and then the timeliness preference model can output the target user's timeliness preference vector. The timeliness preference vector is a feature vector used to reflect the target user's preference for different timeliness exposed transaction vectors.

[0091] The timeliness preference model is used to calculate a timeliness preference vector based on user characteristics and timeliness features. The timeliness preference model training device can train an initial timeliness model based on the user characteristics of sample users and the sample timeliness features of the exposed transaction information corresponding to those users. Sample exposed transaction information refers to the transaction information exposed to the sample users, and sample timeliness features are the timeliness characteristics of the exposed transaction information, reflecting the timeliness of the transaction information, such as the publication time and duration of publication.

[0092] Optionally, the timeliness preference vector of the target user can be u shift The formula is as follows:

[0093] in, For the timeliness preference model, X u User characteristics of the target users To expose the timeliness of transaction information, and, u shift ∈R d The dimension of the timeliness preference vector is d.

[0094] S208. Based on the timeliness feature, obtain the cluster center vector corresponding to the timeliness preference vector.

[0095] Specifically, the timeliness feature is calculated based on the target user's user characteristics and their past browsing history. If the target user has only browsed transaction information for a short period, such as if the target user is a newly registered user, then the amount of transaction information browsed by the target user is limited, and the exposure to transaction information by the target user is also limited. Understandably, because the sample user's corresponding information is too limited, and the exposure to transaction information is limited, the sample user's timeliness preference vector is easily affected by noise and may not be accurate. Therefore, the transaction information recommendation device can use clustering to assist cold-start target users in calculating their preference scores. The transaction information recommendation device can obtain the cluster center vector corresponding to the timeliness preference vector based on the target user's timeliness feature, and then combine the cluster center vector and the timeliness preference vector to calculate the target user's preference score.

[0096] Optionally, the transaction information recommendation device can obtain central timeliness features that match the timeliness features from the central timeliness feature set, and then determine the cluster center vector corresponding to the central timeliness features. The central timeliness feature set includes at least one cluster center vector and the central timeliness features corresponding to each cluster center vector in the at least one cluster center vector.

[0097] Optionally, the central timeliness feature set can have K cluster center vectors, where c is the number of cluster center vectors. k c k The corresponding central timeliness characteristics are k is an integer in the range 1 to K, s l,k Indicates whether user l is related to cluster center vector c. k The matching formula is as follows:

[0098]

[0099] If s l,k =1 indicates that user l It matches the cluster center vector ck, and conversely, if s l,k =0 indicates that user l With cluster center vector c k Mismatch Let be the timeliness preference vector for user l.

[0100] Wherein, the cluster center vector c k The update rules are as follows:

[0101]

[0102] Where ∝ represents the cluster center update rate, ranging from 0 to 1, and n represents the number of all timeliness preference vectors, i.e., the number of all users. The cluster center vector c obtained by this update rule... kIt can represent the timeliness characteristics of all users in the center. The preference vector is used to assist in calculating the preference vector of the target user during cold start.

[0103] S210, based on the cluster center vector and the timeliness preference vector, uses a parameter modulation model to obtain the modulation input parameters for each layer of the time period preference model.

[0104] Specifically, the transaction information push device can use a parametric modulation model based on the cluster center vector and the timeliness preference vector to obtain the modulation input parameters for each layer of the time-period preference model. That is, the transaction information push device can input the cluster center vector and the timeliness preference vector into the parametric modulation model, and the parametric modulation model can output the modulation input parameters for each layer of the time-period preference model. The parametric modulation model is the training template for the parametric modulation model. It is obtained by training the initial modulation model based on the sample timeliness preference vector of the sample users and the sample cluster center vector corresponding to the sample timeliness preference vector. The parametric modulation model can output the modulation input parameters according to the cluster center vector and the timeliness preference vector.

[0105] Optionally, the formula for calculating the modulation input parameters is as follows:

[0106]

[0107] Among them, the time period preference model obtained by training i on the time period dataset of users in time period t is: For time period preference model Modulation input parameters of the m-th layer This is a parameter modulation model, a fully connected layer, activated by the sigmoid function σ.

[0108] S212, Based on the modulation input parameters, the time period preference model is subjected to parameter modulation processing to obtain the target preference model.

[0109] Specifically, the time-period preference model calculates preference parameters by combining the user's preference for transaction information at different times. For example, the time-period preference model can identify that the user's preference for the same transaction information differs between the morning and evening. However, the time-period preference model cannot identify the user's preference for transaction information with different timeliness. The transaction information recommendation device can modulate the parameters of the time-period preference model with several modulated input parameters to obtain the target preference model. This allows the target preference model to also learn to identify the user's preference for transaction information with different timeliness. Thus, the preference score calculated by the target preference model can simultaneously take into account the user's preference for transaction information at different times and the preference for transaction information with different timeliness.

[0110] Optionally, the transaction information recommendation device can calculate the element-wise product of the modulation input parameters and the parameters of the time-period preference model, as shown in the following formula:

[0111]

[0112] in, The time period preference model obtained for user u within the time interval t The parameters of the m-th layer, To obtain the element-wise product, which represents the parameters of the target preference model, the transaction information recommendation device modulates the parameters of the time-period preference model based on the element-wise product, thus obtaining the target preference model. For example, the transaction information recommendation device can use... Replacement time period preference model The target preference model can then be obtained, as shown in the following formula:

[0113]

[0114] in, This is the target preference model, which can be used to calculate the preference score of user u for transaction information i within a time interval t.

[0115] S214. Based on user characteristics and the content characteristics of target transaction information, a target preference model is used to perform preference prediction processing on target users to obtain the target user's preference score for target transaction information.

[0116] Specifically, the transaction information recommendation device can perform preference prediction processing based on user characteristics and the content characteristics of the target transaction information using a target preference model. This involves inputting user characteristics and content characteristics into the target preference model for preference prediction, and the target preference model can output a preference score for the target user regarding the target transaction information. Here, the content characteristics are feature vectors that reflect the specific content of the transaction information.

[0117] S216, Based on preference scores, determine the recommendation strategy for target transaction information for the target user.

[0118] Specifically, the preference score reflects the target user's degree of preference for the target information. A higher preference score indicates a higher degree of preference for the target information, and if the target information is exposed to the target user, the probability of the target user clicking on it is greater. Conversely, a lower preference score indicates a lower degree of preference for the target information, and if the target information is exposed to the target user, the probability of the target user clicking on it is smaller. Therefore, the information recommendation device can determine the recommendation strategy for the target information based on the preference score.

[0119] The recommendation strategy can be to decide whether to recommend target transaction information to the target user. For example, if the preference score is less than the recommendation score threshold, the recommendation strategy is not to recommend the target transaction information to the target user. If the preference score is greater than or equal to the recommendation score threshold, the recommendation strategy is to recommend the target transaction information to the target user, thus exposing the target transaction information to the target user. The recommendation score threshold is used to determine whether to recommend the target transaction information to the target user. It can be the initial setting of the transaction information recommendation device or set by relevant staff or users. For example, it can be 0.5.

[0120] Understandably, recommendation strategies can also include methods of recommending target transaction information to target users. These methods can be pop-up windows, where a pop-up window corresponding to the target transaction information appears on the user's browsing interface, and the user can jump to the main text of the target transaction information by clicking the pop-up window. Alternatively, the recommendation method can be to make the target transaction information appear as the next transaction information that the target user will see when they swipe through the browsing interface. The recommendation method can also be related to preference scores.

[0121] In this embodiment, a time-segment preference model is obtained. This model is used to calculate a user's preference score for transaction information in different time periods. User characteristics of the target user are obtained, as are the timeliness characteristics of the exposed transaction information. Based on the user characteristics and the timeliness characteristics of the exposed transaction information, the timeliness preference model is used to extract preference features from the target user, obtaining a timeliness preference vector. Based on the timeliness characteristics, the cluster center vector corresponding to the timeliness preference vector is obtained. Based on the cluster center vector and the timeliness preference vector, a parameter modulation model is used to obtain modulation input parameters for each layer of the time-segment preference model. The cluster center vector obtained based on the timeliness preference vectors of all users is used to calculate the preference score for the target user. This avoids the problems of low accuracy and high susceptibility to noise in the timeliness preference vector caused by the lack of historical data for cold-start users. The time-segment preference model is then subjected to parameter modulation processing based on the modulation input parameters to obtain a target preference model. Based on the user characteristics and the content features of the target transaction information, the target preference model is used to predict the target user's preference score for the target transaction information. By combining users' preferences for transaction information at different times with the timeliness of the transaction information, a preference score for users on transaction information is calculated, which improves the accuracy and comprehensiveness of the preference score, thereby improving the rationality of transaction information recommendations.

[0122] Please see Figure 5 The diagram below illustrates a method for training a time-period preference model, as provided in this application. Figure 5 As shown, the method described in this application embodiment may include the following steps S302-S308.

[0123] S302, Create an initial preference model and obtain the historical access dataset of sample users.

[0124] Specifically, the time-period preference model training device can create an initial preference model and then acquire a historical access dataset of sample users. The sample users can be any user browsing transaction information using the application. The historical access dataset contains browsing traces left by the sample users while browsing transaction information. Specifically, the historical access dataset includes the user characteristics of the sample users, the content characteristics of the sample exposed transaction information corresponding to the sample users, and the click tags of the sample users for the sample exposed transaction information. The sample exposed transaction information refers to the transaction information exposed to the sample users. The click tags indicate whether the sample user clicked on the sample exposed transaction information and also represent the user's preference score for the sample exposed transaction information. If the user clicked on the sample exposed transaction information, it means the user confirmed their interest in the sample exposed transaction information, and the click tag is 1; otherwise, if the user did not click on the sample exposed transaction information, the click tag is 0.

[0125] S304. Based on the historical access dataset, train the initial preference model for at least one round to obtain the sample user's sample preference score for sample exposure transaction information.

[0126] Specifically, the initial preference model is trained at least once based on the historical access dataset. In each round of model training, the sample user's preference score for the sample exposure transaction information can be obtained. That is, the user characteristics of the sample user and the content characteristics of the sample exposure transaction information corresponding to the sample user are input into the initial preference model, and the preference model can output the sample user's preference score for the sample exposure transaction information.

[0127] Optionally, in order for the time-period preference model to learn the degree of user preference for transaction information in different time intervals, the time-period preference model training device can divide the historical access dataset into at least one time interval dataset according to the time interval, and then perform at least one round of model training on the initial preference model based on at least one time interval dataset to obtain the sample user's sample preference score for sample exposed transaction information in each time interval.

[0128] S306, adjust the parameters of the initial preference model based on the sample preference score and click tag until the initial preference model completes model training and the time period preference model is obtained.

[0129] Specifically, the time-period preference model training device can calculate the time-period preference loss of the initial preference model based on sample preference scores and click labels. Based on this time-period preference loss, the parameters of the initial preference model are adjusted during backpropagation training until the initial preference model completes training, resulting in the time-period preference model. The time-period preference loss can be the difference between the sample preference score and the click label.

[0130] Optionally, the time-period preference model training device can set a training task based on the training set for each time interval to obtain the time-period preference model corresponding to each time interval, such as the time-period preference model. That is, based on the time interval preference model trained on the time interval training set within the time interval t, the user's preference score for transaction information within the time interval t can be calculated.

[0131] This application proposes a time-period preference model training method. An initial preference model is created by obtaining a historical access dataset of sample users. This dataset is divided into at least one time-interval dataset. The initial preference model is then trained at least once based on this dataset to obtain sample preference scores for the exposed transaction information of sample users within each time interval. The parameters of the initial preference model are adjusted based on these preference scores and click tags until the initial preference model completes training, resulting in a time-period preference model. This model is then used to calculate the target user's preference for transaction information in different time intervals, further improving the accuracy of the final preference score.

[0132] Please see Figure 6 This is a flowchart illustrating a timeliness preference model training method provided in this application embodiment. Figure 6 As shown, the method described in this application embodiment may include the following steps S402-S406.

[0133] S402, Create an initial timeliness model, obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users.

[0134] Specifically, the timeliness preference model training device can create an initial timeliness model, obtain the user characteristics of sample users, the sample timeliness characteristics of sample exposure transaction information corresponding to sample users, and the preset timeliness preferences of sample users. The preset timeliness preferences can be the timeliness preference information set by the sample users themselves and sent to the timeliness preference model training device, including the sample users' description of their timeliness preferences for transaction information, such as "preferring transaction information released within seven days". The preference model training device can extract the characteristics of the timeliness preference information to obtain the preset timeliness preferences.

[0135] S404, based on the user characteristics and timeliness characteristics of the sample users, perform at least one round of model training on the initial timeliness model to obtain the sample timeliness preference vector corresponding to the sample users.

[0136] Specifically, the timeliness preference model training device can train the initial model at least once based on the user characteristics of the sample users and the timeliness characteristics of the samples. In each round of model training, the sample timeliness preference vector corresponding to the sample users can be obtained. That is, the user characteristics and sample timeliness characteristics are input into the initial timeliness model, and the initial timeliness model can output the sample timeliness preference vector.

[0137] S406, based on the sample timeliness preference vector and the preset timeliness preference, adjust the parameters of the initial timeliness model until the initial timeliness model completes model training and obtains the timeliness preference model.

[0138] Specifically, the timeliness preference model training device actively calculates the timeliness preference loss of the initial timeliness model using the sample timeliness preference vector and the preset timeliness preference. Based on the timeliness preference loss, the parameters of the initial timeliness model are modulated during the backpropagation training process until the initial timeliness model completes model training, thus obtaining the timeliness preference model. The timeliness preference loss can be the Euclidean distance between the sample timeliness preference vector and the preset timeliness preference.

[0139] This application proposes a training method for a timeliness preference model. An initial timeliness model is created by acquiring user characteristics of sample users, sample timeliness characteristics of sample exposure transaction information corresponding to the sample users, and preset timeliness preferences of the sample users. Based on the user characteristics and sample timeliness characteristics of the sample users, the initial timeliness model is trained for at least one round to obtain sample timeliness preference vectors corresponding to the sample users. Based on the sample timeliness preference vectors and preset timeliness preferences, the parameters of the initial timeliness model are adjusted until the initial timeliness model completes training, resulting in the timeliness preference model. The timeliness preference model can be deployed in a transaction information recommendation device to calculate the timeliness preference vector of target users, assisting in calculating the target users' preference scores for transaction information, improving the accuracy and comprehensiveness of the preference scores, and thus improving the rationality of transaction information recommendations.

[0140] Please see Figure 7 This is a flowchart illustrating a parameter modulation model training method provided in this application. Figure 7 As shown, the method described in this application embodiment may include the following steps S502-S506.

[0141] S502, create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector.

[0142] Specifically, the parameter modulation model training device can create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector. Based on the user characteristics of the sample users and the sample timeliness characteristics of the sample exposure transaction information of the sample users, the parameter modulation model training device can use a timeliness preference model to obtain the sample timeliness preference vector of the sample users.

[0143] S504, based on the sample timeliness preference vector and the sample cluster center vector, performs at least one round of model training on the initial modulation model to obtain the sample modulation input parameters corresponding to the sample users.

[0144] Specifically, the initial modulation model is trained at least once based on the sample timeliness preference vector and the sample cluster center vector. In each round of model training, the sample modulation input parameters corresponding to the sample users can be obtained. That is, the sample timeliness preference vector and the sample cluster center vector are input into the initial modulation model, and the initial modulation model can output the sample modulation input parameters.

[0145] S506, based on the sample modulation input parameters, the time period preference model is processed by parameter modulation to obtain the sample preference model corresponding to the sample user, and based on the sample preference model, the sample user's preference prediction is processed to obtain the sample user's sample timeliness preference score for sample exposure transaction information.

[0146] Specifically, the parameter modulation model training device can perform parameter modulation processing on the time period preference model based on the sample modulation input parameters to obtain the sample preference model corresponding to the sample user. Then, based on the sample preference model, it can perform preference prediction processing on the sample user to obtain the sample timeliness preference score of the sample user for the sample exposure transaction information.

[0147] Optionally, the parameter modulation model training device can acquire the content features of sample exposure transaction information, input the user features of sample users and the content features of sample exposure transaction information into the sample preference model, and the sample preference model can output the sample user's sample preference score for sample exposure transaction information.

[0148] S508, based on the sample timeliness preference score and the click tags of sample users for sample exposure transaction information, adjusts the parameters of the initial modulation model until the initial modulation model completes model training and obtains the parameter modulation model.

[0149] Specifically, the training transpose of the parameter modulation model can be calculated based on the sample timeliness preference score and the click tags of sample users for sample exposure transaction information. Based on the log loss, the parameters of the initial preference model are adjusted during the backpropagation training process until the initial modulation model completes model training, resulting in the parameter modulation model. The formula for calculating the log loss L(φ) is as follows:

[0150]

[0151] in, This refers to the click tags of sample users on sample exposure transaction information, indicating that user u clicked on the exposure transaction information i within the time interval t. k Click the tag, The timeliness preference score is the sample timeliness score.

[0152] This application proposes a training method for a parameter modulation model. An initial modulation model is created, and the sample timeliness preference vector of a sample user and the corresponding sample cluster center vector are obtained. Based on the sample timeliness preference vector and the sample cluster center vector, the initial modulation model is trained for at least one round to obtain the sample modulation input parameters corresponding to the sample user. Based on the sample modulation input parameters, the time-period preference model is subjected to parameter modulation processing to obtain the sample preference model corresponding to the sample user. Based on the sample preference model, preference prediction processing is performed on the sample user to obtain the sample timeliness preference score of the sample user for sample exposure transaction information. Based on the sample timeliness preference score and the click tags of the sample user for sample exposure transaction information, the parameters of the initial modulation model are adjusted until the initial modulation model completes training, resulting in the parameter modulation model. The parameter modulation model can be deployed in a transaction information recommendation device to calculate the input modulation parameters, assisting in calculating the target user's preference score for transaction information, improving the accuracy and comprehensiveness of the preference score, thereby improving the rationality of transaction information recommendation.

[0153] It is understandable that the parameter modulation model, time period preference model, and timeliness preference model are all DNN models. The data used in the training process of these models can all come from the historical access dataset of sample users. The historical access dataset can include the user characteristics of the sample users, the sample exposure transaction information corresponding to the sample users, the click tags of the sample users for the sample exposure transaction information, the sample timeliness characteristics of the sample exposure transaction information, and the sample users' preset timeliness preferences, etc. The historical access dataset can be divided into a support set S. u，t and query set Q u，t The time-period preference model training device, the time-sensitivity preference model training device, and the parameter modulation model training device can be based on S u，t and Qu，t Model-Agnostic Meta-Learning (MAML) was used to train the parameter modulation model, time period preference model, and timeliness preference model.

[0154] Optionally, since the time-period preference model training device can divide the historical access dataset into at least one time-period dataset according to time intervals, and set a training task based on each time-period training set, the time-period preference training model corresponding to each time interval can be obtained. The time-period preference model training device can divide each time-period training set into a support set Su, t and a query set Q. u,t Based on S u，t and Q u,t The MAML method was used to train the time period preference model for each time interval.

[0155] The following will be combined with the appendix Figure 8 This application provides a detailed description of the transaction information recommendation device provided in its embodiments. It should be noted that the appendix... Figure 8 The transaction information recommendation device in the application is used to perform the transaction information recommendation function in this application. Figure 3 and Figure 4 The methods shown in the embodiments are for illustrative purposes only, illustrating the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figure 3 and Figure 4 The example shown.

[0156] Please see Figure 8 This illustration shows a schematic diagram of a transaction information recommendation device provided in an exemplary embodiment of this application. The transaction information recommendation device can be implemented as all or part of a device through software, hardware, or a combination of both. The device 1 includes a model acquisition module 11, a timeliness preference calculation module 12, a parameter modulation module 13, a preference prediction module 14, and a recommendation strategy determination module 15.

[0157] The model acquisition module 11 is used to acquire a time period preference model, which is used to calculate the user's preference score for transaction information in different time periods;

[0158] The timeliness preference calculation module 12 is used to obtain the timeliness preference vector of the target user based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, wherein the exposed transaction information is the transaction information exposed to the target user;

[0159] Optionally, the timeliness preference calculation module 12 is specifically used to obtain the user characteristics of the target user and the timeliness characteristics of the exposed transaction information;

[0160] Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference model is used to extract the preference features of the target user to obtain the timeliness preference vector of the target user.

[0161] Parameter modulation module 13 is used to perform parameter modulation processing on the time period preference model using the timeliness preference vector to obtain the target preference model;

[0162] Optionally, the parameter modulation module 13 is specifically used to obtain modulation input parameters for each layer of the time period preference model based on the timeliness preference vector;

[0163] Based on the modulation input parameters, the time period preference model is subjected to parameter modulation processing to obtain the target preference model.

[0164] Optionally, the parameter modulation module 13 is specifically used to obtain the cluster center vector corresponding to the timeliness preference vector based on the timeliness feature;

[0165] Based on the cluster center vector and the timeliness preference vector, a parameter modulation model is used to obtain the modulation input parameters for each layer of the time period preference model.

[0166] Optionally, the parameter modulation module 13 is specifically used to obtain a central timeliness feature that matches the timeliness feature from the central timeliness feature set;

[0167] Determine the cluster center vector corresponding to the central timeliness feature, wherein the central timeliness feature set includes at least one cluster center vector and the central timeliness feature corresponding to each cluster center vector in the at least one cluster center vector.

[0168] Optionally, the parameter modulation module 13 is specifically used to calculate the element-wise product of the modulation input parameters and the parameters of the time period preference model;

[0169] The target preference model is obtained by performing parameter modulation processing on the time period preference model based on the element-wise product.

[0170] Preference prediction module 14 is used to perform preference prediction processing on the target user based on the user characteristics and the content characteristics of the target transaction information, using the target preference model to obtain the target user's preference score for the target transaction information;

[0171] The recommendation strategy determination module 15 is used to determine a recommendation strategy for the target transaction information of the target user based on the preference score.

[0172] In this embodiment, a time-segment preference model is obtained. This model is used to calculate a user's preference score for transaction information in different time periods. User characteristics of the target user and the timeliness characteristics of the exposed transaction information are obtained. Based on the user characteristics and the timeliness characteristics of the exposed transaction information, the timeliness preference model is used to extract preference features from the target user, obtaining a timeliness preference vector for the target user. Based on the timeliness characteristics, the cluster center vector corresponding to the timeliness preference vector is obtained. Based on the cluster center vector and the timeliness preference vector, a parameter modulation model is used to obtain modulation input parameters for each layer of the time-segment preference model. The cluster center vector obtained based on the timeliness preference vectors of all users is used to calculate the preference score for the target user. This avoids the problem of low accuracy and high susceptibility to noise in the timeliness preference vector due to the lack of historical data for cold-start users. The time-segment preference model is then subjected to parameter modulation processing based on the modulation input parameters to obtain a target preference model. Based on the user characteristics and the content characteristics of the target transaction information, the target preference model is used to predict the target user's preference, obtaining the target user's preference score for the target transaction information. By combining users' preferences for transaction information at different times with the timeliness of the transaction information, a user preference score for the transaction information is calculated, which improves the accuracy and comprehensiveness of the preference score, thereby improving the rationality of the transaction information recommendation.

[0173] Please see Figure 9 This illustration shows a schematic diagram of the structure of a time-period preference model training device provided in an exemplary embodiment of this application. This time-period preference model training device can be implemented as all or part of a device through software, hardware, or a combination of both. The device 2 includes a first model creation module 21, a first model training module 22, and a first model acquisition module 23.

[0174] The first model creation module 21 obtains the historical access dataset of the sample user. The historical access dataset includes the user characteristics of the sample user, the content characteristics of the sample exposure transaction information corresponding to the sample user, and the click tags of the sample user for the sample exposure transaction information. The sample exposure transaction information is the transaction information exposed to the sample user.

[0175] The first model training module 22 is used to perform at least one round of model training on the initial preference model based on the historical access dataset to obtain the sample user's sample preference score for the sample exposure transaction information.

[0176] Optionally, the first model training module 22 is specifically used to divide the historical access dataset into at least one time interval dataset according to the time interval;

[0177] The initial preference model is trained for at least one round based on the at least one time interval dataset to obtain the sample user's sample preference score for the sample exposure transaction information in each time interval.

[0178] The first model acquisition module 23 is used to adjust the parameters of the initial preference model based on the sample preference score and the click tag until the initial preference model completes model training and obtains the time period preference model.

[0179] In this embodiment, a time-period preference model training method is proposed. An initial preference model is created, and a historical access dataset of sample users is obtained. This dataset is divided into at least one time-period dataset. Then, the initial preference model is trained at least once based on this dataset to obtain sample preference scores for the exposed transaction information of sample users in each time interval. The parameters of the initial preference model are adjusted based on these preference scores and click tags until the initial preference model completes training, resulting in the time-period preference model. This model is used to calculate the target user's preference for transaction information in different time intervals, further improving the accuracy of the final preference score.

[0180] Please see Figure 10 This illustration shows a schematic diagram of the structure of a timeliness preference model training device provided in an exemplary embodiment of this application. This timeliness preference model training device can be implemented as all or part of a device through software, hardware, or a combination of both. The device 3 includes a second model creation module 31, a second model training module 32, and a second model acquisition module 33.

[0181] The second model creation module 31 is used to create an initial timeliness model, obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users.

[0182] The second model training module 32 is used to perform at least one round of model training on the initial timeliness model based on the user characteristics and timeliness characteristics of the sample users, so as to obtain the sample timeliness preference vector corresponding to the sample users.

[0183] The second model acquisition module 33 is used to adjust the parameters of the initial timeliness model based on the sample timeliness preference vector and the preset timeliness preference until the initial timeliness model completes model training and obtains the timeliness preference model.

[0184] This embodiment proposes a training method for a timeliness preference model. An initial timeliness model is created by acquiring the user characteristics of sample users, the sample timeliness characteristics of sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users. Based on the user characteristics and sample timeliness characteristics of the sample users, the initial timeliness model is trained for at least one round to obtain the sample timeliness preference vector corresponding to the sample users. Based on the sample timeliness preference vector and the preset timeliness preferences, the parameters of the initial timeliness model are adjusted until the initial timeliness model completes training, resulting in the timeliness preference model. The timeliness preference model can be deployed in a transaction information recommendation device to calculate the timeliness preference vector of target users, assisting in calculating the target users' preference scores for transaction information, improving the accuracy and comprehensiveness of the preference scores, and thus improving the rationality of transaction information recommendations.

[0185] Please see Figure 11 This illustration shows a schematic diagram of the structure of a parameter modulation model training apparatus provided in an exemplary embodiment of this application. The parameter modulation model training apparatus can be implemented as all or part of the apparatus through software, hardware, or a combination of both. The apparatus 4 includes a third model creation module 41, a third model training module 42, a model modulation module 43, and a third model acquisition module 44.

[0186] The third model creation module 41 is used to create an initial modulation model, obtain the sample timeliness preference vector of the sample user, and the sample cluster center vector corresponding to the sample timeliness preference vector;

[0187] The third model training module 42 is used to perform at least one round of model training on the initial modulation model based on the sample timeliness preference vector and the sample cluster center vector to obtain the sample modulation input parameters corresponding to the sample user.

[0188] Model modulation module 43 is used to perform parameter modulation processing on the time period preference model based on the sample modulation input parameters to obtain the sample preference model corresponding to the sample user, and perform preference prediction processing on the sample user based on the sample preference model to obtain the sample user's sample timeliness preference score for sample exposure transaction information;

[0189] The third model acquisition module 44 is used to adjust the parameters of the initial modulation model based on the sample timeliness preference score and the click tags of the sample users for the sample exposure transaction information, until the initial modulation model completes model training and obtains the parameter modulation model.

[0190] This application proposes a training method for a parameter modulation model. An initial modulation model is created, and the sample timeliness preference vector of a sample user and the corresponding sample cluster center vector are obtained. Based on the sample timeliness preference vector and the sample cluster center vector, the initial modulation model is trained for at least one round to obtain the sample modulation input parameters corresponding to the sample user. Based on the sample modulation input parameters, the time-period preference model is subjected to parameter modulation processing to obtain the sample preference model corresponding to the sample user. Based on the sample preference model, preference prediction processing is performed on the sample user to obtain the sample timeliness preference score of the sample user for sample exposure transaction information. Based on the sample timeliness preference score and the click tags of the sample user for sample exposure transaction information, the parameters of the initial modulation model are adjusted until the initial modulation model completes training, resulting in the parameter modulation model. The parameter modulation model can be deployed in a transaction information recommendation device to calculate the input modulation parameters, assisting in calculating the target user's preference score for transaction information, improving the accuracy and comprehensiveness of the preference score, thereby improving the rationality of transaction information recommendation.

[0191] This application also provides a computer storage medium that can store multiple instructions, which are adapted to be loaded and executed by a processor as described above. Figures 1-7 The specific execution process of the transaction information recommendation method, time period preference model training method, timeliness preference model training method, and parameter modulation model training method described in the illustrated embodiment can be found in [reference needed]. Figures 1-7 The specific details of the illustrated embodiments will not be elaborated here.

[0192] This application also provides a computer program product storing at least one instruction, which is loaded and executed by the processor as described above. Figure 1-7 The transaction information recommendation method, time period preference model training method, timeliness preference model training method, and parameter modulation model training method described in the illustrated embodiment can be found in the following documentation for their specific execution process. Figures 1-7 The specific details of the illustrated embodiments will not be elaborated here.

[0193] Please refer to Figure 12 This diagram illustrates a structural block diagram of an electronic device provided in an exemplary embodiment of this application. The electronic device in this application may include one or more components such as a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, memory 120, input device 130, and output device 140 may be connected via the bus 150.

[0194] Processor 110 may include one or more processing cores. Processor 110 connects to various parts of the electronic device using various interfaces and lines, and executes various functions of terminal 100 and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 120, and by calling data stored in memory 120. Optionally, processor 110 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 110 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user page, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 110 and may be implemented separately using a communication chip.

[0195] The memory 120 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 120 may include non-transitory computer-readable storage medium. The memory 120 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described above, etc. The operating system may be the Android system, including systems deeply developed based on the Android system, the iOS system developed by Apple Inc., including systems deeply developed based on the iOS system, or other systems.

[0196] The memory 120 can be divided into operating system space and user space. The operating system runs in the operating system space, while native and third-party applications run in user space. To ensure that different third-party applications can achieve good running performance, the operating system allocates corresponding system resources for each application. However, different application scenarios within the same third-party application have different requirements for system resources. For example, in local resource loading scenarios, third-party applications have high requirements for disk read speed; in animation rendering scenarios, third-party applications have high requirements for GPU performance. Since the operating system and third-party applications are independent of each other, the operating system often cannot promptly perceive the current application scenario of a third-party application, resulting in the operating system's inability to adapt system resources accordingly.

[0197] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to establish data communication between the third-party applications and the operating system. This would allow the operating system to obtain the current scenario information of the third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0198] The input device 130 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 140 is used to output instructions or data, and includes, but is not limited to, a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 can be a touch display screen.

[0199] The touch display screen can be designed as a full-screen, curved screen, or irregularly shaped screen. It can also be designed as a combination of a full-screen and a curved screen, or a combination of an irregularly shaped screen and a curved screen; however, this application does not limit the specific design in this regard.

[0200] In addition, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, Wireless Fidelity (WiFi) modules, power supplies, Bluetooth modules, etc., which will not be described in detail here.

[0201] exist Figure 12In the electronic device shown, the processor 110 can be used to call the transaction information recommendation application stored in the memory 120 and execute it to implement the transaction information recommendation method, time period preference model training method, timeliness preference model training method and parameter modulation model training method as described in the various method embodiments of this specification.

[0202] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0203] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

[0204] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, user characteristics and exposure information involved in this specification were obtained with full authorization.

Claims

1. A method for recommending transaction information, the method comprising: Obtain a time period preference model, which is used to calculate the user's preference score for transaction information in different time periods; Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference vector of the target user is obtained, wherein the exposed transaction information is the transaction information exposed to the target user; The time-sensitivity preference vector is used to modulate the parameters of the time-period preference model to obtain the target preference model; Based on the user characteristics and the content characteristics of the target transaction information, the target preference model is used to perform preference prediction processing on the target user to obtain the target user's preference score for the target transaction information. Based on the preference score, a recommendation strategy for the target transaction information of the target user is determined; The step of using the timeliness preference vector to perform parameter modulation processing on the time period preference model to obtain the target preference model includes: Based on the timeliness preference vector, modulation input parameters for each layer of the time period preference model are obtained; The element-wise product is calculated based on the modulation input parameters and the parameters of the time period preference model. The time period preference model is then subjected to parameter modulation processing based on the element-wise product to obtain the target preference model.

2. The method according to claim 1, wherein obtaining the timeliness preference vector of the target user based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information includes: Obtain the user characteristics of the target users and the timeliness characteristics of the exposed transaction information; Based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, a timeliness preference model is used to extract the preference features of the target user to obtain the timeliness preference vector of the target user.

3. The method according to claim 1, wherein obtaining the modulation input parameters for each layer of the time-period preference model based on the timeliness preference vector includes: Based on the timeliness feature, obtain the cluster center vector corresponding to the timeliness preference vector; Based on the cluster center vector and the timeliness preference vector, a parameter modulation model is used to obtain the modulation input parameters for each layer of the time period preference model.

4. The method according to claim 3, wherein obtaining the cluster center vector corresponding to the timeliness preference vector based on the timeliness feature includes: Obtain the central timeliness feature that matches the timeliness feature from the central timeliness feature set; Determine the cluster center vector corresponding to the central timeliness feature, wherein the central timeliness feature set includes at least one cluster center vector and the central timeliness feature corresponding to each cluster center vector in the at least one cluster center vector.

5. The method according to claim 1, wherein the method comprises: An initial preference model is created, and a historical access dataset of sample users is obtained. The historical access dataset includes the user characteristics of the sample users, the content characteristics of the sample exposure transaction information corresponding to the sample users, and the click tags of the sample users for the sample exposure transaction information. The sample exposure transaction information is the transaction information exposed to the sample users. Based on the historical access dataset, the initial preference model is trained for at least one round to obtain the sample user's sample preference score for the sample exposure transaction information; The parameters of the initial preference model are adjusted based on the sample preference scores and the click tags until the initial preference model completes model training, resulting in a time-period preference model.

6. The method according to claim 5, wherein training the initial preference model on the historical access dataset for at least one round to obtain the sample user's sample preference score for the sample exposure transaction information includes: The historical access dataset is divided into at least one time interval dataset according to the time interval; The initial preference model is trained for at least one round based on the at least one time interval dataset to obtain the sample user's sample preference score for the sample exposure transaction information in each time interval.

7. The method according to claim 1, wherein the method comprises: Create an initial timeliness model to obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users; Based on the user characteristics and timeliness characteristics of the sample users, the initial timeliness model is trained for at least one round to obtain the sample timeliness preference vector corresponding to the sample users; Based on the sample timeliness preference vector and the preset timeliness preference, the parameters of the initial timeliness model are adjusted until the initial timeliness model completes model training, and a timeliness preference model is obtained.

8. The method according to claim 1, wherein the method comprises: Create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector; Based on the sample timeliness preference vector and the sample cluster center vector, the initial modulation model is trained for at least one round to obtain the sample modulation input parameters corresponding to the sample user. Based on the sample modulation input parameters, the time period preference model is subjected to parameter modulation processing to obtain the sample preference model corresponding to the sample user. Based on the sample preference model, the sample user is subjected to preference prediction processing to obtain the sample user's sample timeliness preference score for sample exposure transaction information. Based on the sample timeliness preference score and the click tags of the sample users for the sample exposure transaction information, the parameters of the initial modulation model are adjusted until the initial modulation model completes model training and a parameter modulation model is obtained.

9. A transaction information recommendation device, the device comprising: The model acquisition module is used to acquire the time period preference model, which is used to calculate the user's preference score for transaction information in different time periods; The timeliness preference calculation module is used to obtain the timeliness preference vector of the target user based on the user characteristics of the target user and the timeliness characteristics of the exposed transaction information, wherein the exposed transaction information is the transaction information exposed to the target user; The parameter modulation module is used to perform parameter modulation processing on the time period preference model using the timeliness preference vector to obtain the target preference model; The preference prediction module is used to perform preference prediction processing on the target user based on the user characteristics and the content characteristics of the target transaction information, using the target preference model to obtain the target user's preference score for the target transaction information; The recommendation strategy determination module is used to determine a recommendation strategy for the target transaction information of the target user based on the preference score. The parameter modulation module is specifically used to obtain modulation input parameters for each layer of the time period preference model based on the timeliness preference vector. The element-wise product is calculated based on the modulation input parameters and the parameters of the time period preference model. The time period preference model is then subjected to parameter modulation processing based on the element-wise product to obtain the target preference model.

10. The apparatus according to claim 9, further comprising: The first model creation module is used to create an initial preference model and obtain the historical access dataset of the sample users. The historical access dataset includes the user characteristics of the sample users, the content characteristics of the sample exposure transaction information corresponding to the sample users, and the click tags of the sample users for the sample exposure transaction information. The sample exposure transaction information is the transaction information exposed to the sample users. The first model training module is used to perform at least one round of model training on the initial preference model based on the historical access dataset to obtain the sample user's sample preference score for the sample exposure transaction information. The first model acquisition module is used to adjust the parameters of the initial preference model based on the sample preference score and the click tag until the initial preference model completes model training and obtains the time period preference model.

11. The apparatus according to claim 9, further comprising: The second model creation module is used to create an initial timeliness model, obtain the user characteristics of the sample users, the sample timeliness characteristics of the sample exposure transaction information corresponding to the sample users, and the preset timeliness preferences of the sample users; The second model training module is used to perform at least one round of model training on the initial timeliness model based on the user characteristics and timeliness characteristics of the sample users, so as to obtain the sample timeliness preference vector corresponding to the sample users. The second model acquisition module is used to adjust the parameters of the initial timeliness model based on the sample timeliness preference vector and the preset timeliness preference until the initial timeliness model completes model training and obtains the timeliness preference model.

12. The apparatus according to claim 9, further comprising: The third model creation module is used to create an initial modulation model, obtain the sample timeliness preference vector of the sample users, and the sample cluster center vector corresponding to the sample timeliness preference vector; The third model training module is used to perform at least one round of model training on the initial modulation model based on the sample timeliness preference vector and the sample cluster center vector to obtain the sample modulation input parameters corresponding to the sample user. The model modulation module is used to perform parameter modulation processing on the time period preference model based on the sample modulation input parameters to obtain the sample preference model corresponding to the sample user, and to perform preference prediction processing on the sample user based on the sample preference model to obtain the sample user's sample timeliness preference score for sample exposure transaction information. The third model acquisition module is used to adjust the parameters of the initial modulation model based on the sample timeliness preference score and the click tags of the sample users for the sample exposure transaction information, until the initial modulation model completes model training and obtains the parameter modulation model.

13. A computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps of any one of claims 1 to 8.

14. A computer program product storing a plurality of instructions adapted for loading by a processor and executing the method steps of any one of claims 1 to 8.

15. An electronic device comprising: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed the method steps as claimed in any one of claims 1 to 8.

Images