Resource recommendation and estimation model acquisition method and apparatus, electronic device, and medium

Abstract

The present disclosure provides resource recommendation and estimation model obtaining method and device, and relates to the fields of artificial intelligence such as deep learning, big data processing and information flow recommendation. The resource recommendation method can include: determining an estimation value of each candidate resource by using an estimation model according to user information of a user to be recommended and resource information of each candidate resource, the estimation value being used to represent the satisfaction degree of the user to be recommended to the candidate resource, the estimation model being obtained by training a training sample in advance, the training sample including user information of a sample user, resource information of any recommended resource recommended to the sample user and a label, the label being used to represent whether the sample user is satisfied with the recommended resource and being determined according to operation behavior of the sample user to the recommended resource; and determining a candidate resource recommended to the user to be recommended from each candidate resource according to the estimation value. By applying the scheme of the present disclosure, the information cocoon problem during resource recommendation can be effectively alleviated.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to methods, apparatuses, electronic devices and media for acquiring resource recommendation and prediction models in the fields of deep learning, big data processing and information flow recommendation. Background Technology

[0002] In today's society, there is a wide variety of information. Among them, information push notifications through applications, i.e., resource recommendations, have become a common way for users to obtain information. Summary of the Invention

[0003] This disclosure provides methods, apparatus, electronic devices, and media for obtaining resource recommendation and prediction models.

[0004] A resource recommendation method, comprising:

[0005] Based on the user information of the user to be recommended and the resource information of each candidate resource, a prediction value for each candidate resource is determined using a prediction model. The prediction value is used to represent the user's satisfaction with the candidate resource. The prediction model is trained using a pre-constructed training sample, which includes: user information of the sample user, resource information of any recommended resource recommended to the sample user, and a label. The label is used to indicate whether the sample user is satisfied with the recommended resource, and the label is determined based on the sample user's operational behavior towards the recommended resource.

[0006] Based on the estimated value, candidate resources are determined from each candidate resource to be recommended to the user.

[0007] A method for obtaining a prediction model, comprising:

[0008] Construct training samples, which include: user information of sample users, resource information of any recommended resource recommended to the sample users, and tags. The tags are used to indicate whether the sample users are satisfied with the recommended resources, and the tags are determined based on the sample users' operation behavior on the recommended resources.

[0009] The prediction model is trained based on the training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from among the candidate resources.

[0010] A resource recommendation device includes: an information acquisition module and a resource recommendation module;

[0011] The information acquisition module is used to determine the estimated value of each candidate resource based on the user information of the user to be recommended and the resource information of each candidate resource using a prediction model. The estimated value is used to represent the satisfaction of the user to be recommended with the candidate resource. The prediction model is trained using a pre-constructed training sample. The training sample includes: user information of the sample user, resource information of any recommended resource recommended to the sample user, and a label. The label is used to indicate whether the sample user is satisfied with the recommended resource. The label is determined based on the sample user's operation behavior on the recommended resource.

[0012] The resource recommendation module is used to determine candidate resources to recommend to the user based on the estimated value from each candidate resource.

[0013] A predictive model acquisition device includes: a sample construction module and a model training module;

[0014] The sample construction module is used to construct training samples, which include: user information of the sample user, resource information of any recommended resource recommended to the sample user, and tags. The tags are used to indicate whether the sample user is satisfied with the recommended resource, and the tags are determined based on the sample user's operation behavior on the recommended resource.

[0015] The model training module is used to train the prediction model based on the training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from among the candidate resources.

[0016] An electronic device, comprising:

[0017] At least one processor; and

[0018] A memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described above.

[0020] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above.

[0021] A computer program product includes a computer program / instructions that, when executed by a processor, implement the method described above.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0023] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0024] Figure 1 This is a flowchart of an embodiment of the resource recommendation method described in this disclosure;

[0025] Figure 2 This is a flowchart of an embodiment of the prediction model acquisition method described in this disclosure;

[0026] Figure 3 This is a schematic diagram of the composition structure of Embodiment 300 of the resource recommendation device described in this disclosure;

[0027] Figure 4 This is a schematic diagram of the composition structure of the first embodiment 400 of the predictive model acquisition device described in this disclosure;

[0028] Figure 5 This is a schematic diagram of the composition structure of the second embodiment 500 of the predictive model acquisition device described in this disclosure;

[0029] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0030] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0031] Furthermore, it should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0032] Figure 1This is a flowchart illustrating an embodiment of the resource recommendation method described in this disclosure. Figure 1 As shown, the specific implementation methods are as follows.

[0033] In step 101, based on the user information of the user to be recommended and the resource information of each candidate resource, the estimated value of each candidate resource is determined using the prediction model. The estimated value is used to represent the satisfaction of the user to be recommended with the candidate resource. The prediction model is trained using a pre-constructed training sample. The training sample includes: user information of the sample user, resource information of any recommended resource recommended to the sample user, and a label. The label is used to indicate whether the sample user is satisfied with the recommended resource and is determined based on the sample user's operation behavior on the recommended resource.

[0034] In step 102, candidate resources to be recommended to the user are determined from each candidate resource based on the estimated value.

[0035] In traditional methods, for each user to be recommended to, a relevance score is typically obtained between each candidate resource and the user. The candidate resources are then ranked from highest to lowest relevance score, and the top-ranked resources are recommended to the user. However, this approach can easily lead to the concentration of resources related to the user's interests, resulting in information silos.

[0036] The solution described in the above method embodiment can recommend resources to the user based on the user's satisfaction with each candidate resource, thereby effectively alleviating the information cocoon problem during resource recommendation and making the recommended resources more in line with the user's needs, thus improving the accuracy of the recommendation results.

[0037] To differentiate between different users, the users to be recommended resources are called users to be recommended, and the users in the training samples are called sample users.

[0038] Assuming there are 1000 candidate resources (the number is for illustrative purposes only and will not be repeated below), referred to as candidate resource 1 to candidate resource 1000, for each candidate resource, the estimated value of the candidate resource can be determined by using the prediction model based on the resource information of the candidate resource and the user information of the user to be recommended. The resource information and user information can be used as inputs to the prediction model to obtain the output estimated value.

[0039] The specific content included in user information and resource information can be determined based on actual needs. For example, user information may include user attributes such as age, gender, and occupation, as well as user type, such as whether they are a light or cold-start user, or a moderate to heavy user. Similarly, resource information may include the resource itself and its attributes.

[0040] The estimated value can reflect the satisfaction level of the users to be recommended with each candidate resource. Accordingly, the candidate resources to be recommended to the users to be recommended can be determined from each candidate resource based on the estimated value.

[0041] Preferably, the candidate resources can be sorted in descending order of their estimated values, and the top M candidate resources after sorting can be recommended to the user to be recommended, where M is a positive integer and less than the number of candidate resources. Alternatively, the comprehensive score of each candidate resource can be determined by combining the estimated value and other predetermined parameters, and then the candidate resources can be sorted in descending order of their comprehensive scores, and the top M candidate resources after sorting can be recommended to the user to be recommended.

[0042] The specific value of M can be determined according to actual needs. For example, taking 1000 candidate resources as an example, after sorting the 1000 candidate resources in descending order of estimated value, the top 10 candidate resources can be recommended to the user to be recommended.

[0043] Furthermore, the specific parameters for other predetermined parameters are not restricted and can be determined according to actual needs. For example, they could include the estimated click duration of candidate resources identified using a deep neural network model. Accordingly, for each candidate resource, its comprehensive score can be calculated using a predetermined fusion formula based on its corresponding estimated value and other predetermined parameters.

[0044] In other words, the estimated value can be used to directly rank each candidate resource, or it can be combined with other predetermined parameters to rank each candidate resource, which is very flexible and convenient.

[0045] In addition, it can be seen that the implementation of the resource recommendation method described in this disclosure depends on the pre-trained prediction model. The following explains how to obtain the prediction model.

[0046] Figure 2 This is a flowchart illustrating an embodiment of the prediction model acquisition method described in this disclosure. Figure 2 As shown, the specific implementation methods are as follows.

[0047] In step 201, training samples are constructed, which include: user information of the sample users, resource information of any recommended resource recommended to the sample users, and tags. The tags are used to indicate whether the sample users are satisfied with the recommended resources, and the tags are determined based on the sample users' operation behavior on the recommended resources.

[0048] In step 202, a prediction model is trained based on the training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from among the candidate resources.

[0049] By employing the scheme described in the above-described method embodiments, it is possible to determine whether sample users are satisfied with the recommended resources based on the landing page feedback of sample users. Training samples can be constructed accordingly, and a prediction model can be trained. Consequently, when recommending resources, the prediction model can be used to determine the estimated value of each candidate resource, i.e., the user's satisfaction level. Resources can then be recommended to the user based on this satisfaction level, effectively alleviating the information cocoon problem during resource recommendation and making the recommended resources more aligned with the user's needs, thus improving the accuracy of the recommendation results.

[0050] For example, if a resource is recommended to a user, it can be called a recommended resource to distinguish it from other resources. Then, the user can be used as a sample user, and a training sample can be constructed accordingly. The training sample includes the user's information, the resource information of the recommended resource, and the tag.

[0051] Preferably, the label can be determined as follows: in response to determining that a sample user has performed an interactive action on the recommended resource, the label is determined to be satisfactory; in response to determining that a sample user has not performed an interactive action on the recommended resource, the label is determined to be satisfactory or unsatisfactory based on the sample user's click duration and / or click completion rate on the recommended resource.

[0052] The interactive behavior can refer to actions such as liking, commenting, or sharing.

[0053] If it is determined that the sample users have performed interactive behavior on the recommended resources, then the sample users can be considered to be satisfied with the recommended resources, and the label can be determined accordingly as satisfied. Otherwise, the label can be further determined as satisfied or dissatisfied based on the sample users' click duration and / or click completion rate on the recommended resources.

[0054] Recommended resources can be of any type, such as videos or text / images. For videos, click duration refers to the time sample users spent watching the video, and click completion rate refers to the percentage of viewing time relative to the total video length. For text / images, click duration refers to the time sample users spent reading the text / images, or browsing time, and click completion rate refers to the percentage of read content relative to the total text / image length.

[0055] As can be seen, the above processing method can determine whether the required label is satisfactory or unsatisfactory based on whether the sample users have performed interactive behavior on the recommended resources. It is simple and convenient, and ensures the accuracy of the determination results.

[0056] Preferably, when determining whether a label is "satisfied" or "unsatisfied" based on the click duration and / or click completion rate of sample users for recommended resources, the target threshold corresponding to the sample users can be obtained first. Then, the label can be determined as "satisfied" or "unsatisfied" by comparing the click duration and / or click completion rate of sample users for recommended resources with the target threshold.

[0057] In the solution described in this disclosure, a "user satisfaction" evaluation system can be established, which defines high click duration, high click completion rate, and interactive behavior exhibited by users on the landing page as user satisfaction with the recommended resources. Satisfaction can be determined if any one of the three conditions is met.

[0058] Unlike interactive behaviors, which have clear signals, high click duration and high click completion rate do not have explicit tracking information. Therefore, a method can be used to obtain a target threshold and compare the click duration and / or click completion rate of sample users for recommended resources with the target threshold to determine whether the label is satisfactory or unsatisfactory.

[0059] Preferably, when obtaining the target threshold corresponding to the sample user, in response to determining that there is a personalized threshold corresponding to the sample user, the personalized threshold includes: personalized click duration threshold and personalized click completion rate threshold corresponding to different resource types respectively, and the personalized click duration threshold and personalized click completion rate threshold corresponding to the resource type of the recommended resource in the personalized threshold can be used as the target threshold.

[0060] Assuming there are two resource types, video and text / image, the personalized thresholds for the sample users can include: personalized click duration thresholds and personalized click completion rate thresholds for the video resource type, and personalized click duration thresholds and personalized click completion rate thresholds for the text / image resource type. If the recommended resource type is video, then the personalized click duration threshold and personalized click completion rate threshold for the video resource type can be used as the desired target thresholds from the personalized thresholds for the sample users.

[0061] Based on the obtained target threshold, the label can be determined as either "satisfied" or "unsatisfied." Preferably, the label is determined to be "satisfied" in response to any of the following conditions being met: the sample user's click duration on the recommended resource is greater than the personalized click duration threshold in the target threshold; the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold in the target threshold; or the sample user's click duration on the recommended resource is greater than the personalized click duration threshold and the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold. The label is determined to be "unsatisfied" in response to any of the conditions not being met.

[0062] If a personalized threshold exists for a sample user, the target threshold can be determined based on the personalized threshold, making the obtained target threshold more targeted and thus improving the accuracy of the determined tags.

[0063] In addition, preferably, in response to determining that there is no personalized threshold for sample users, the target threshold can be determined based on the pre-obtained global threshold modeling results.

[0064] Preferably, the global threshold modeling result includes: a first polynomial relationship and a second polynomial relationship corresponding to different resource types. The first polynomial relationship represents the relationship between the bucket number and the global click duration threshold, and the second polynomial relationship represents the relationship between the bucket number and the global click completion rate threshold. Accordingly, the method for determining the target threshold based on the pre-obtained global threshold modeling result may include: determining the bucket to which the recommended resource belongs based on the resource type and resource length of the recommended resource, and determining the global click duration threshold and the global click completion rate threshold corresponding to the bucket number based on the bucket number and the first and second polynomial relationships corresponding to the resource type of the recommended resource. The determined global click duration threshold and global click completion rate threshold are used as the required target threshold. In this case, each resource type corresponds to more than one bucket, each bucket corresponds to its own resource length range, and the resource length ranges corresponding to any two buckets do not overlap, and each bucket has a unique serial number.

[0065] Resource length refers to the length of a video or the length of images and text, etc.

[0066] Assuming there are two resource types, video and text / image, the global threshold modeling result can include the first and second polynomial relationships corresponding to the video resource type, as well as the first and second polynomial relationships corresponding to the text / image resource type. Assuming the recommended resource type is video, the bucket to which it belongs can be determined based on the video length. Then, based on the bucket number and the first and second polynomial relationships corresponding to the video resource type, the global click duration threshold and global click completion rate threshold corresponding to that bucket number can be determined. These determined global click duration thresholds and global click completion rate thresholds can then be used as the desired target thresholds.

[0067] Taking the first polynomial relationship as an example, it can be understood as a relationship model between x and y, where x is the bin number and y is the global click duration threshold. Accordingly, using the bin number as x, the corresponding y can be determined according to the first polynomial relationship, that is, the global click duration threshold corresponding to the bin number can be determined. Similarly, with the help of the second polynomial relationship, the global click completion rate threshold corresponding to the bin number can be determined.

[0068] Preferably, the label can be determined as satisfactory in response to any of the following conditions being met: the sample user's click duration on the recommended resource is greater than the global click duration threshold in the target threshold; the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold in the target threshold; or the sample user's click duration on the recommended resource is greater than the global click duration threshold and the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold. In response to any of the conditions not being met, the label is determined as unsatisfactory.

[0069] It can be seen that if there is no personalized threshold for the sample user, the target threshold can be determined based on the global threshold modeling results. This provides a good supplement for the case where there is no personalized threshold, ensuring that the required target threshold can still be obtained, thus ensuring the smooth progress of subsequent processing.

[0070] Both the personalized threshold and the global threshold modeling results can be obtained offline in advance and can be updated periodically, for example, once every 24 hours.

[0071] Preferably, target users that meet predetermined requirements can be identified. Target users that meet predetermined requirements include: target users whose posterior information quantity is greater than a predetermined threshold, i.e., moderate to heavy users. A first user set is formed using target users that meet predetermined requirements, and a second user set is formed using the remaining target users. Target users are users who have made resource recommendations within the most recent predetermined time period. For each target user in the first user set, a corresponding personalized threshold is determined. For each target user in the second user set, global threshold modeling is performed to obtain the global threshold modeling result common to all target users in the second user set.

[0072] That is, different processing methods can be adopted for different types of users, such as medium and heavy users in the first user set and light or cold start users in the second user set, so that the solution described in this disclosure can be applied to different types of users.

[0073] The specific value of the most recent predetermined duration can be determined according to actual needs, such as the most recent 14 days. Similarly, the specific value of the predetermined threshold can also be determined according to actual needs.

[0074] To distinguish them from other users, users who have received resource recommendations within the most recent scheduled time are referred to as target users.

[0075] The following sections explain the specific methods for obtaining the modeling results of personalized thresholds and global thresholds.

[0076] 1) Personalized threshold

[0077] Preferably, for any target user in the first user set, for any resource type, the following processing can be performed respectively: obtain the resources of the resource type that have been recommended to the target user and clicked within the most recent predetermined time period, as the target resources; obtain the click duration of the target user for each target resource; take the click duration corresponding to the predetermined percentile of each obtained click duration as the personalized click duration threshold of the target user; obtain the click completion rate of the target user for each target resource; take the click completion rate corresponding to the predetermined percentile of each obtained click completion rate as the personalized click completion rate threshold of the target user.

[0078] Assuming the first user set includes 1000 target users, and assuming the resource types include video and text / image, then for each of these 1000 target users, the following information can be obtained: the personalized click duration threshold and personalized click completion rate threshold for the video resource type, and the personalized click duration threshold and personalized click completion rate threshold for the text / image resource type.

[0079] The specific value of the predetermined quantile can be determined according to actual needs, such as the 80th quantile.

[0080] Taking target user a and video as the resource type as an example, the corresponding personalized click duration threshold can be obtained in the following way: Suppose that 50 videos are recommended to target user a and clicked within the recently scheduled time. These 50 videos can be used as target videos, and the click duration of target user a for each of these 50 target videos can be obtained. Then, the click duration corresponding to the 80th percentile of the 50 click durations can be obtained as the personalized click duration threshold for the video resource type corresponding to target user a.

[0081] 2) Global threshold modeling results

[0082] Preferably, for any resource type, the following processing can be performed: Resources of the resource type that have been recommended to target users in the second user set and clicked within the most recent predetermined time period are obtained as target resources; the resource type is assigned to a bucket according to its length, with the resource type corresponding to more than one bucket; for each bucket, the following processing is performed: for each target resource belonging to that bucket, the click duration of the corresponding target user on each target resource is obtained, and the click duration corresponding to the predetermined percentile of each obtained click duration is used as the global click duration threshold for that bucket; for each target resource belonging to that bucket, the click completion rate of the corresponding target user on each target resource is obtained, and the click completion rate corresponding to the predetermined percentile of each obtained click completion rate is used as the global click completion rate threshold for that bucket; based on the obtained bucket number and the global click duration threshold for each bucket, a first polynomial relationship is obtained by fitting using the least squares method; based on the obtained bucket number and the global click completion rate threshold for each bucket, a second polynomial relationship is obtained by fitting using the least squares method.

[0083] Considering that the distribution of click duration and click completion rate is greatly affected by resource length, the scheme described in this disclosure adopts a polynomial modeling approach to fit the relationship between resource length and global click duration threshold and global click completion rate threshold.

[0084] Assuming the second user set includes 500 target users, and assuming the resource types include video and text / image, then for these 500 target users, the following information can be obtained: the first and second polynomial relations corresponding to the video resource type, and the first and second polynomial relations corresponding to the text / image resource type.

[0085] Taking video as an example, the first polynomial relationship can be obtained as follows: Assume that 1500 videos were recommended to target users in the second user set within the recently scheduled time and were clicked. These 1500 videos can be taken as target videos, and each target video can be bucketed according to its length. That is, the bucket to which each target video belongs can be determined. Assume there are 100 buckets in total, with serial numbers from 1 to 100. Among them, the video length range corresponding to bucket 1 can be less than 2 minutes, the video length range corresponding to bucket 2 can be 2 to 4 minutes, and the video length range corresponding to bucket 3 can be 4 minutes or more. For each bucket, a global click duration threshold can be determined. Specifically, taking bucket 1 as an example, the click duration of each target video belonging to bucket 1 can be obtained, and the click duration corresponding to the 80th percentile of each obtained click duration can be determined as the global click duration threshold for bucket 1. Similarly, the global click duration thresholds corresponding to 100 buckets can be obtained. Furthermore, based on the obtained sequence numbers of the 100 buckets and the global click duration thresholds corresponding to the 100 buckets, the required first polynomial relationship can be obtained by fitting the least squares method.

[0086] As can be seen, in the scheme described in this disclosure, by statistically analyzing and modeling user history, for moderate to heavy users with rich posterior information, statistical analysis methods can be used to determine the personalized thresholds for each user. For light and cold-start users with sparse posterior information, multinomial modeling can be used to define thresholds according to resource type. Correspondingly, when constructing training samples, the satisfaction of sample users with the recommended resources can be determined according to the user type of the sample users. Then, the evaluation index of user satisfaction can be used as the learning target for model training, providing a comprehensive and reliable basis for model learning.

[0087] In practical applications, the labels in the training samples can be represented by 1 and 0 to indicate user satisfaction and dissatisfaction, respectively.

[0088] When a sufficient number of training samples are constructed, a prediction model can be trained using these samples. This prediction model can be used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value represents the user's satisfaction with each candidate resource, and this satisfaction level is used to determine which candidate resources to recommend to the user.

[0089] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this disclosure. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this disclosure. Furthermore, for parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0090] In summary, the solution described in the embodiments of this disclosure can effectively alleviate the information cocoon problem in resource recommendation and improve the accuracy of recommendation results. Moreover, it is applicable to different user types and different resource types, thus having broad applicability.

[0091] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0092] Figure 3 This is a schematic diagram of the structural composition of embodiment 300 of the resource recommendation device described in this disclosure. Figure 3 As shown, it includes: an information acquisition module 301 and a resource recommendation module 302.

[0093] The information acquisition module 301 is used to determine the estimated value of each candidate resource based on the user information of the user to be recommended and the resource information of each candidate resource using a prediction model. The estimated value is used to represent the satisfaction of the user to be recommended with the candidate resource. The prediction model is trained using a pre-constructed training sample. The training sample includes: user information of the sample user, resource information of any recommended resource recommended to the sample user, and a label. The label is used to indicate whether the sample user is satisfied with the recommended resource and is determined based on the sample user's operation behavior on the recommended resource.

[0094] The resource recommendation module 302 is used to determine the candidate resources to be recommended to the user from among the candidate resources based on the estimated value.

[0095] By adopting the solution described in the above-described device embodiment, resources can be recommended to the user based on the user's satisfaction with each candidate resource, thereby effectively alleviating the information cocoon problem during resource recommendation and making the recommended resources more in line with the user's needs, thus improving the accuracy of the recommendation results.

[0096] For each candidate resource, the information acquisition module 301 can determine the estimated value of the candidate resource based on the resource information of the candidate resource and the user information of the user to be recommended, using the prediction model. The resource information and user information can be used as inputs to the prediction model to obtain the output estimated value.

[0097] The estimated value can reflect the satisfaction of the user to be recommended with each candidate resource. Accordingly, the resource recommendation module 302 can determine the candidate resources to be recommended to the user from each candidate resource based on the estimated value.

[0098] Preferably, the resource recommendation module 302 can sort the candidate resources in descending order of the estimated value, and recommend the top M candidate resources to the user to be recommended, where M is a positive integer and less than the number of candidate resources. Alternatively, the resource recommendation module 302 can combine the estimated value and other predetermined parameters to determine the comprehensive score of each candidate resource, and then sort the candidate resources in descending order of the comprehensive score, and recommend the top M candidate resources to the user to be recommended.

[0099] Figure 4 This is a schematic diagram of the structural composition of the first embodiment 400 of the predictive model acquisition device described in this disclosure. Figure 4 As shown, it includes: a sample construction module 401 and a model training module 402.

[0100] The sample construction module 401 is used to construct training samples, which include: user information of the sample user, resource information of any recommended resource recommended to the sample user, and tags. The tags are used to indicate whether the sample user is satisfied with the recommended resource, and the tags are determined based on the sample user's operation behavior on the recommended resource.

[0101] The model training module 402 is used to train a prediction model based on training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from among the candidate resources.

[0102] By employing the scheme described in the above-mentioned device embodiment, it is possible to determine whether the sample user is satisfied with the recommended resources based on the feedback from the sample user's landing page. Accordingly, training samples can be constructed and a prediction model can be trained. Consequently, when recommending resources, the prediction model can be used to determine the estimated value of each candidate resource, i.e., the user's satisfaction level. Then, resources can be recommended to the user to be recommended based on the satisfaction level, thereby effectively alleviating the information cocoon problem during resource recommendation and making the recommended resources more in line with the user's needs, i.e., improving the accuracy of the recommendation results.

[0103] Preferably, the sample construction module 401 determines the tag in the following manner: in response to determining that the sample user has performed an interactive behavior on the recommended resource, the tag is determined to be "satisfied"; in response to determining that the sample user has not performed an interactive behavior on the recommended resource, the tag is determined to be "satisfied" or "unsatisfied" based on the sample user's click duration and / or click completion rate on the recommended resource. The interactive behavior may refer to liking, commenting, or sharing, etc.

[0104] If it is determined that the sample users have performed interactive behavior on the recommended resources, then the sample users can be considered to be satisfied with the recommended resources, and the label can be determined accordingly as satisfied. Otherwise, the label can be further determined as satisfied or dissatisfied based on the sample users' click duration and / or click completion rate on the recommended resources.

[0105] Preferably, when determining whether a label is "satisfied" or "unsatisfied" based on the click duration and / or click completion rate of the sample users on the recommended resources, the sample construction module 401 can first obtain the target threshold corresponding to the sample users, and then determine whether the label is "satisfied" or "unsatisfied" by comparing the click duration and / or click completion rate of the sample users on the recommended resources with the target threshold.

[0106] Preferably, when the sample construction module 401 obtains the target threshold corresponding to the sample user, it responds to determining that there is a personalized threshold corresponding to the sample user. The personalized threshold includes: personalized click duration threshold and personalized click completion rate threshold corresponding to different resource types. The personalized click duration threshold and personalized click completion rate threshold corresponding to the resource type of the recommended resource in the personalized threshold can be used as the target threshold.

[0107] Based on the obtained target threshold, the label can be determined as either "satisfied" or "unsatisfied." Preferably, the sample construction module 401 determines the label as "satisfied" in response to determining that any of the following conditions are met: the sample user's click duration on the recommended resource is greater than the personalized click duration threshold in the target threshold; the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold in the target threshold; or the sample user's click duration on the recommended resource is greater than the personalized click duration threshold and the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold. In response to determining that none of the above conditions are met, the label is determined as "unsatisfied."

[0108] In addition, preferably, in response to determining that there is no personalized threshold corresponding to a user with no sample, the sample construction module 401 can determine the target threshold based on the pre-obtained global threshold modeling results.

[0109] Preferably, the global threshold modeling result includes: a first polynomial relationship and a second polynomial relationship corresponding to different resource types. The first polynomial relationship is used to represent the relationship between the bucket number and the global click duration threshold, and the second polynomial relationship is used to represent the relationship between the bucket number and the global click completion rate threshold. Accordingly, the sample construction module 401 determines the target threshold based on the pre-obtained global threshold modeling result in the following ways: determining the bucket to which the recommended resource belongs based on the resource type and resource length of the recommended resource, and determining the global click duration threshold and the global click completion rate threshold corresponding to the bucket number based on the bucket number and the first and second polynomial relationships corresponding to the resource type of the recommended resource. The determined global click duration threshold and global click completion rate threshold are used as the required target threshold. In this case, any resource type corresponds to more than one bucket, each bucket corresponds to its own resource length range, and the resource length ranges corresponding to any two buckets do not overlap, and each bucket has a unique number.

[0110] Preferably, the sample construction module 401 determines the label as satisfactory in response to determining that any of the following conditions are met: the sample user's click duration on the recommended resource is greater than the global click duration threshold in the target threshold; the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold in the target threshold; or the sample user's click duration on the recommended resource is greater than the global click duration threshold and the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold. In response to determining that none of the above conditions are met, the label is determined as unsatisfactory.

[0111] Figure 5 This is a schematic diagram of the composition structure of the second embodiment 500 of the predictive model acquisition device described in this disclosure. Figure 5As shown, it includes: a sample construction module 401, a model training module 402, and a preprocessing module 403.

[0112] Among them, the sample construction module 401 and the model training module 402 and Figure 4 The same applies to the embodiments shown, and will not be repeated here.

[0113] The preprocessing module 403 is used to identify target users who meet predetermined requirements. Target users who meet predetermined requirements include those whose posterior information quantity is greater than a predetermined threshold, i.e., moderate to heavy users. A first user set is formed using target users who meet predetermined requirements, and a second user set is formed using the remaining target users. Target users are users who have made resource recommendations within the most recent predetermined time period. For each target user in the first user set, a corresponding personalized threshold is determined. For each target user in the second user set, global threshold modeling is performed to obtain the global threshold modeling result that is common to all target users in the second user set.

[0114] Preferably, for any target user in the first user set and for any resource type, the preprocessing module 403 may perform the following processing: obtain the resources of the resource type that have been recommended to the target user and clicked within the most recent predetermined time period as target resources; obtain the click duration of the target user for each target resource; use the click duration corresponding to the predetermined percentile of each obtained click duration as the personalized click duration threshold of the target user; obtain the click completion rate of the target user for each target resource; use the click completion rate corresponding to the predetermined percentile of each obtained click completion rate as the personalized click completion rate threshold of the target user.

[0115] Preferably, the preprocessing module 403 may perform the following processing for any resource type: obtain resources of the resource type that have been recommended to target users in the second user set and clicked within the most recent predetermined time period, and use these as target resources; determine the bucket to which each target resource belongs according to its resource length, wherein the resource type corresponds to more than one bucket; for each bucket, perform the following processing: for each target resource belonging to that bucket, obtain the click duration of the corresponding target user on each target resource, and use the click duration corresponding to the predetermined percentile in each obtained click duration as... For each target resource belonging to a given bucket, the global click duration threshold is determined. The click completion rate of each target user for that resource is then calculated. The click completion rate at a predetermined quantile within each completed click completion rate is used as the global click completion rate threshold for that bucket. Based on the bucket number and the global click duration threshold, a first polynomial relationship is obtained through least squares fitting. Similarly, a second polynomial relationship is obtained based on the bucket number and the global click completion rate threshold.

[0116] Figures 3-5 The specific workflow of the device embodiment shown can be found in the relevant descriptions in the foregoing method embodiments, and will not be repeated here.

[0117] In summary, the solution described in the embodiments of this disclosure can effectively alleviate the information cocoon problem during resource recommendation and improve the accuracy of recommendation results. Moreover, it is applicable to different user types and different resource types, thus having broad applicability.

[0118] The solutions described in this disclosure can be applied to the field of artificial intelligence, particularly deep learning, big data processing, and information flow recommendation. Artificial intelligence is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It involves both hardware and software technologies. Artificial intelligence hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. Artificial intelligence software technologies mainly include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0119] The resources and operational behaviors described in the embodiments of this disclosure are not targeted at any specific user and do not reflect the personal information of any specific user. The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solutions of this disclosure all comply with relevant laws and regulations and do not violate public order and good morals.

[0120] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0121] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0122] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0123] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0124] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as those described in this disclosure. For example, in some embodiments, the methods described in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the methods described in this disclosure can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform the methods described in this disclosure by any other suitable means (e.g., by means of firmware).

[0125] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0126] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0127] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0128] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0129] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0130] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0131] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0132] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A resource recommendation method, comprising: Based on the user information of the user to be recommended and the resource information of each candidate resource, a prediction value for each candidate resource is determined using a prediction model. The prediction value represents the user's satisfaction with the candidate resource. The prediction model is trained using pre-constructed training samples, which include: user information of the sample user, resource information of any recommended resource to the sample user, and tags. The tags indicate whether the sample user is satisfied with the recommended resource. The tags are determined by comparing the sample user's click duration and / or click completion rate with the target threshold after determining that the sample user has not interacted with the recommended resource and obtaining the target threshold corresponding to the sample user. The target threshold is determined based on the personalized threshold or global threshold modeling result corresponding to the sample user. The personalized threshold is the corresponding personalized threshold determined for each target user in the first user set, and the global threshold modeling result is the common global threshold modeling result obtained after performing global threshold modeling for each target user in the second user set. The first user set is composed of target users who meet predetermined requirements, and the second user set is composed of the remaining target users. The target users who meet the predetermined requirements include those with a posterior information quantity greater than a predetermined threshold, and those who have recommended resources within the most recent predetermined time period. The global threshold modeling result includes a first polynomial relationship and a second polynomial relationship corresponding to different resource types. The first polynomial relationship represents the relationship model between the bucket number and the global click duration threshold, and the second polynomial relationship represents the relationship model between the bucket number and the global click completion rate threshold. Both the first and second polynomial relationships are obtained by least squares fitting. The target thresholds determined based on the global threshold modeling result include: determining the bucket to which the recommended resource belongs based on its resource type and length, and determining the global click duration threshold and global click completion rate threshold corresponding to the bucket number based on the bucket number and the first and second polynomial relationships corresponding to the resource type of the recommended resource. Based on the estimated value, candidate resources are determined from each candidate resource to be recommended to the user.

2. The method according to claim 1, wherein, The step of determining the candidate resources to recommend to the user based on the estimated value includes: The candidate resources are sorted in descending order of the estimated value, and the top M candidate resources are recommended to the user to be recommended, where M is a positive integer and less than the number of candidate resources. Alternatively, by combining the estimated value and other predetermined parameters, a comprehensive score for each candidate resource can be determined, and the candidate resources can be sorted in descending order of the comprehensive score. The candidate resources ranked in the top M positions after sorting can be recommended to the user to be recommended.

3. A method for obtaining a prediction model, comprising: Construct training samples, which include: user information of the sample user, resource information of any recommended resource recommended to the sample user, and tags, wherein the tags are used to indicate whether the sample user is satisfied with the recommended resource; The prediction model is trained based on the training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from each candidate resource. The method for determining the tag includes: in response to determining that the sample user has not performed an interactive behavior on the recommended resource, obtaining a target threshold corresponding to the sample user, and determining the tag as satisfied or dissatisfied by comparing the click duration and / or click completion rate of the sample user on the recommended resource with the target threshold; the target threshold is determined based on the personalized threshold or global threshold modeling results corresponding to the sample user; The method for obtaining the personalized threshold and the global threshold modeling result includes: identifying target users who meet predetermined requirements, including target users whose posterior information quantity is greater than a predetermined threshold; forming a first user set using the target users who meet the predetermined requirements; forming a second user set using the remaining target users; the target users are users who have recommended resources within the most recent predetermined time period; determining the corresponding personalized threshold for each target user in the first user set; performing global threshold modeling for each target user in the second user set to obtain the global threshold modeling result common to all target users; the global threshold modeling result includes a first polynomial relationship and a second polynomial relationship corresponding to different resource types, the first polynomial relationship representing the relationship model between the bucket number and the global click duration threshold, and the second polynomial relationship representing the relationship model between the bucket number and the global click completion rate threshold; both the first polynomial relationship and the second polynomial relationship are obtained by least squares fitting. Determining the target threshold based on the global threshold modeling results includes: determining the bucket to which the recommended resource belongs based on the resource type and resource length of the recommended resource; and determining the global click duration threshold and global click completion rate threshold corresponding to the bucket number and the first polynomial relationship and the second polynomial relationship corresponding to the resource type of the recommended resource, based on the bucket number and the first polynomial relationship and the second polynomial relationship corresponding to the resource type of the recommended resource, as the target threshold.

4. The method according to claim 3, wherein, The methods for determining the label also include: In response to determining that the sample user has performed an interactive action on the recommended resource, the tag is determined to be satisfactory.

5. The method according to claim 3, wherein, The target threshold for obtaining the sample user includes: In response to determining the existence of a personalized threshold corresponding to the sample user, the personalized threshold includes: a personalized click duration threshold and a personalized click completion rate threshold corresponding to different resource types, and the personalized click duration threshold and the personalized click completion rate threshold corresponding to the resource type of the recommended resource in the personalized threshold are used as the target threshold.

6. The method according to claim 5, wherein, The step of determining whether the tag is satisfactory or unsatisfactory by comparing the click duration and / or click completion rate of the sample users on the recommended resources with the target threshold includes: The tag is determined to be satisfactory in response to any of the following conditions: the sample user's click duration on the recommended resource is greater than the personalized click duration threshold, the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold, or the sample user's click duration on the recommended resource is greater than the personalized click duration threshold and the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold. In response to determining that any of the conditions are not met, the label is determined to be unsatisfactory.

7. The method according to claim 3, wherein, The step of obtaining the target threshold corresponding to the sample user includes: in response to determining that there is no personalized threshold corresponding to the sample user, determining the target threshold based on the global threshold modeling result.

8. The method according to claim 7, wherein, Each resource type corresponds to more than one bucket, each bucket corresponds to its own resource length range, and the resource length ranges corresponding to any two buckets do not overlap, and each bucket has a unique sequence number.

9. The method according to claim 8, wherein, The step of determining whether the tag is satisfactory or unsatisfactory by comparing the click duration and / or click completion rate of the sample users on the recommended resources with the target threshold includes: The tag is determined to be satisfactory in response to any of the following conditions: the sample user's click duration on the recommended resource is greater than the global click duration threshold, the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold, or the sample user's click duration on the recommended resource is greater than the global click duration threshold and the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold. In response to determining that any of the conditions are not met, the label is determined to be unsatisfactory.

10. The method according to claim 5, wherein, The step of determining the corresponding personalized threshold for each target user in the first user set includes: For any target user in the first user set, and for any resource type, the following processing is performed respectively: Get the resources of the resource type that were recommended to the target user and clicked within the most recent scheduled time period, and use them as the target resources; The click duration of the target user for each target resource is obtained respectively, and the click duration corresponding to the predetermined percentile in each obtained click duration is used as the personalized click duration threshold of the target user; The click completion rate of the target user for each target resource is obtained respectively, and the click completion rate corresponding to the predetermined percentile of each obtained click completion rate is used as the personalized click completion rate threshold of the target user.

11. The method according to claim 8, wherein, The step of performing global threshold modeling for each target user in the second user set to obtain the global threshold modeling result common to all target users includes: For each resource type, the following processing is performed: Get the resource of the resource type that was recommended to the target user in the second user set and clicked within the most recent scheduled time period, and use it as the target resource; Each target resource is assigned a bucket based on its length, and the resource type corresponds to more than one bucket. For each bucket, the following processing is performed: For each target resource belonging to the bucket, the click duration of the corresponding target user for each target resource is obtained, and the click duration corresponding to the predetermined percentile in each obtained click duration is used as the global click duration threshold for the bucket; For each target resource belonging to the bucket, the click completion rate of the corresponding target user for each target resource is obtained, and the click completion rate corresponding to the predetermined percentile in each obtained click completion rate is used as the global click completion rate threshold for the bucket. Based on the obtained sequence number of each bucket and the global click duration threshold corresponding to each bucket, the first polynomial relationship is obtained by fitting using the least squares method. Based on the obtained sequence number of each bucket and the global click completion rate threshold corresponding to each bucket, the second polynomial relationship is obtained by fitting using the least squares method.

12. A resource recommendation device, comprising: Information acquisition module and resource recommendation module; The information acquisition module is used to determine the estimated value of each candidate resource based on the user information of the user to be recommended and the resource information of each candidate resource using a prediction model. The estimated value represents the satisfaction of the user to be recommended with the candidate resource. The prediction model is trained using a pre-constructed training sample, which includes: user information of the sample user, resource information of any recommended resource recommended to the sample user, and tags. The tags represent whether the sample user is satisfied with the recommended resource. The tags are determined by comparing the click duration and / or click completion rate of the sample user with the target threshold after determining that the sample user has not performed an interaction with the recommended resource and obtaining the target threshold corresponding to the sample user. The target threshold is determined based on the personalized threshold or global threshold modeling result corresponding to the sample user. The personalized threshold is the corresponding personalized threshold determined for each target user in the first user set, and the global threshold modeling result is the common corresponding global threshold model obtained after performing global threshold modeling for each target user in the second user set. The modeling results are as follows: the first user set consists of target users who meet predetermined requirements, and the second user set consists of the remaining target users. The target users who meet the predetermined requirements include those with a posterior information quantity greater than a predetermined threshold, and those who have recommended resources within the most recent predetermined time period. The global threshold modeling results include: a first polynomial relationship and a second polynomial relationship corresponding to different resource types. The first polynomial relationship represents the relationship model between the bucket number and the global click duration threshold, and the second polynomial relationship represents the relationship model between the bucket number and the global click completion rate threshold. Both the first and second polynomial relationships are obtained through least squares fitting. The target thresholds determined based on the global threshold modeling results include: determining the bucket to which the recommended resource belongs based on its resource type and length, and determining the global click duration threshold and global click completion rate threshold corresponding to the bucket number based on the bucket number and the first and second polynomial relationships corresponding to the resource type of the recommended resource. The resource recommendation module is used to determine candidate resources to recommend to the user to be recommended from among the candidate resources based on the estimated value.

13. The apparatus according to claim 12, wherein, The resource recommendation module sorts the candidate resources in descending order of the estimated value, and recommends the top M candidate resources to the user to be recommended, where M is a positive integer and less than the number of candidate resources. Alternatively, the resource recommendation module combines the estimated value and other predetermined parameters to determine the comprehensive score of each candidate resource, sorts the candidate resources in descending order of the comprehensive score, and recommends the top M candidate resources to the user to be recommended.

14. A predictive model acquisition device, comprising: The module consists of a preprocessing module, a sample construction module, and a model training module. The sample construction module is used to construct training samples, which include: user information of the sample user, resource information of any recommended resource recommended to the sample user, and tags. The tags are used to indicate whether the sample user is satisfied with the recommended resource. The method for determining the tags includes: in response to determining that the sample user has not performed any interaction with the recommended resource, obtaining a target threshold corresponding to the sample user, and determining the tag as satisfied or dissatisfied by comparing the sample user's click duration and / or click completion rate with the target threshold; the target threshold is determined based on the personalized threshold or global threshold modeling results corresponding to the sample user. The model training module is used to train the prediction model based on the training samples. The prediction model is used to determine the estimated value of each candidate resource. The estimated value is determined based on the user information of the user to be recommended and the resource information of each candidate resource. The estimated value is used to represent the satisfaction of the user to be recommended with each candidate resource. The satisfaction is used to determine the candidate resources to be recommended to the user from among the candidate resources. The preprocessing module is used to identify target users who meet predetermined requirements. These target users include those with a posterior information quantity greater than a predetermined threshold. A first user set is formed using these target users, and a second user set is formed using the remaining target users. The target users are those who have recently participated in resource recommendations within a predetermined timeframe. For each target user in the first user set, a corresponding personalized threshold is determined. For each target user in the second user set, global threshold modeling is performed to obtain a global threshold modeling result common to all target users. The global threshold modeling result includes a first polynomial relationship and a second polynomial relationship corresponding to different resource types. The first polynomial relationship represents the relationship model between the bucket number and the global click duration threshold, and the second polynomial relationship represents the relationship model between the bucket number and the global click completion rate threshold. Both the first and second polynomial relationships are obtained through least squares fitting. The sample construction module determines the bucket to which the recommended resource belongs based on the resource type and resource length of the recommended resource, and determines the global click duration threshold and global click completion rate threshold corresponding to the bucket number based on the first polynomial relation and the second polynomial relation corresponding to the resource type of the recommended resource, as the target threshold.

15. The apparatus according to claim 14, wherein, The sample construction module is further configured to determine the tag as satisfactory in response to determining that the sample user has performed an interactive behavior on the recommended resource.

16. The apparatus according to claim 14, wherein, The sample construction module responds to determining that there is a personalized threshold corresponding to the sample user. The personalized threshold includes: personalized click duration threshold and personalized click completion rate threshold corresponding to different resource types. The personalized click duration threshold and personalized click completion rate threshold corresponding to the resource type of the recommended resource in the personalized threshold are used as the target threshold.

17. The apparatus according to claim 16, wherein, The sample construction module determines the tag as satisfactory in response to determining that any of the following conditions are met: the sample user's click duration on the recommended resource is greater than the personalized click duration threshold, the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold, or the sample user's click duration on the recommended resource is greater than the personalized click duration threshold and the sample user's click completion rate on the recommended resource is greater than the personalized click completion rate threshold; and in response to determining that any of the conditions are not met, the tag is determined as unsatisfactory.

18. The apparatus according to claim 14, wherein, The sample construction module is further configured to determine the target threshold based on the global threshold modeling result in response to determining that there is no personalized threshold corresponding to the sample user.

19. The apparatus according to claim 18, wherein, Each resource type corresponds to more than one bucket, each bucket corresponds to its own resource length range, and the resource length ranges corresponding to any two buckets do not overlap, and each bucket has a unique sequence number.

20. The apparatus according to claim 19, wherein, The sample construction module determines the tag as satisfactory in response to determining that any of the following conditions are met: the sample user's click duration on the recommended resource is greater than the global click duration threshold, the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold, or the sample user's click duration on the recommended resource is greater than the global click duration threshold and the sample user's click completion rate on the recommended resource is greater than the global click completion rate threshold; and in response to determining that any of the conditions are not met, the tag is determined as unsatisfactory.

21. The apparatus according to claim 16, wherein, For any target user in the first user set and for any resource type, the preprocessing module performs the following processing: It obtains resources of the resource type that have been recommended to and clicked by the target user within the most recent predetermined time period, and uses these as target resources; it obtains the click duration of each target resource by the target user, and uses the click duration corresponding to a predetermined percentile in each obtained click duration as the personalized click duration threshold for the target user; it obtains the click completion rate of each target resource by the target user, and uses the click completion rate corresponding to a predetermined percentile in each obtained click completion rate as the personalized click completion rate threshold for the target user.

22. The apparatus according to claim 19, wherein, The preprocessing module performs the following processing for any resource type: It acquires resources of the resource type that have been recommended to and clicked by target users in the second user set within the most recent predetermined time period, and designates them as target resources; it determines the bucket to which each target resource belongs based on its resource length, with each resource type corresponding to more than one bucket; for each bucket, it performs the following processing: for each target resource belonging to the bucket, it acquires the click duration of the corresponding target user for each target resource, and uses the click duration corresponding to the predetermined percentile in each acquired click duration as the global click duration threshold for the bucket; for each target resource belonging to the bucket, it acquires the click completion rate of the corresponding target user for each target resource, and uses the click completion rate corresponding to the predetermined percentile in each acquired click completion rate as the global click completion rate threshold for the bucket; based on the acquired bucket number and the global click duration threshold for each bucket, it fits the first polynomial relationship using the least squares method; based on the acquired bucket number and the global click completion rate threshold for each bucket, it fits the second polynomial relationship using the least squares method.

23. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

24. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-11.

25. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the method of any one of claims 1-11.

Images