Resource Recommendation Method, Apparatus, Electronic Device, and Storage Medium

Abstract

The present disclosure provides a resource recommendation method, apparatus, electronic device, and storage medium, which relate to the field of computer technology, and particularly to technical fields such as artificial intelligence, big data, and information flow. The specific implementation solution is as follows: filtering out the historical accessed resources that meet the preset conditions from the historical accessed resource set of the target object to obtain a reference resource set; determining the degree of interest in each reference resource based on the user operation data of the target object for each reference resource; and screening out a plurality of resources to be recommended from the similar resource sets of each reference resource based on the degree of interest in each reference resource and recommending them to the target object. The present disclosure filters out the historical accessed resources that the user is not interested in from the historical accessed resource set based on the preset conditions to obtain an interested reference resource set. Then, based on the user operation data, the degree of interest in each reference resource is determined. Finally, based on the similarity and the degree of interest between the similar resources and the reference resources, resource screening and recommendation are performed, which can improve the recommendation accuracy.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular, to technologies such as artificial intelligence, big data, information flow, etc. Background Art

[0002] Completing resource recommendation generally includes two parts: a recall stage and a ranking stage. Among them, the recall stage is used to screen out resources to be recommended from a large number of resources, and then in the ranking stage, the resources to be recommended are ranked and recommended to users. In the recall stage, generally, the historical browsing records of users are used as the indexing basis for resources to be recommended to recall similar resources. However, the accuracy of this resource recommendation method still needs to be improved. Summary of the Invention

[0003] The present disclosure provides a resource recommendation method, apparatus, electronic device, and storage medium.

[0004] According to one aspect of the present disclosure, there is provided a resource recommendation method, including:

[0005] Filtering out historical accessed resources that meet a preset condition from the historical accessed resource set of a target object to obtain a reference resource set; the preset condition is used to screen out resources that the target object is not interested in;

[0006] Based on the user operation data of the target object for each reference resource in the reference resource set, determining the degree of interest of the target object in each reference resource;

[0007] Based on the degree of interest of the target object in each reference resource, screening out a plurality of resources to be recommended from the similar resource sets of each reference resource;

[0008] Recommending the plurality of resources to be recommended to the target object.

[0009] According to another aspect of the present disclosure, there is provided a resource recommendation apparatus, including:

[0010] A filtering module, configured to filter out historical accessed resources that meet a preset condition from the historical accessed resource set of a target object to obtain a reference resource set; the preset condition is used to screen out resources that the target object is not interested in;

[0011] A determining module, configured to determine the degree of interest of the target object in each reference resource based on the user operation data of the target object for each reference resource in the reference resource set;

[0012] A screening module, configured to screen out a plurality of resources to be recommended from the similar resource sets of each reference resource based on the degree of interest of the target object in each reference resource;

[0013] A recommending module, configured to recommend the plurality of resources to be recommended to the target object.

[0014] According to another aspect of the present disclosure, there is provided an electronic device, including:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a resource recommendation method.

[0018] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute a resource recommendation method.

[0019] According to another aspect of the present disclosure, there is provided a computer program product including a computer program, and the computer program implements the above-mentioned resource recommendation method when executed by a processor.

[0020] For the solution provided in this embodiment, the present disclosure filters out the historical accessed resources that the user is not interested in from the historical accessed resource set based on preset conditions to obtain an interested reference resource set. Then, based on the user operation data, the interest degree for each reference resource is determined. Finally, based on the similarity and interest degree between the similar resources and the reference resources, resource screening and recommendation are performed, which can improve the recommendation accuracy.

[0021] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0023] Figure 1 is a schematic flowchart of a resource recommendation method according to an embodiment of the present disclosure;

[0024] Figure 2 is a schematic flowchart of a resource recommendation method according to another embodiment of the present disclosure;

[0025] Figure 3 is a schematic flowchart of a resource recommendation method according to another embodiment of the present disclosure;

[0026] Figure 4 is a schematic diagram for determining the interest degree in a resource recommendation method according to another embodiment of the present disclosure;

[0027] Figure 5 is a schematic structural diagram of a resource recommendation apparatus according to an embodiment of the present disclosure;

[0028] Figure 6 is another structural schematic diagram of a resource recommendation device according to another embodiment of the present disclosure;

[0029] Figure 7 is a block diagram of an electronic device for implementing the resource recommendation method of the embodiments of the present disclosure. Detailed implementation manners

[0030] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, descriptions of well-known functions and structures are omitted below for clarity and conciseness.

[0031] In the technical solution of the present disclosure, the acquisition, storage, and application of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0032] In the related art, the user's historical accessed resources are used as an indexing basis to recall and sort resources similar to the browsing records for recommendation. For example, a user-resource graph model is trained so that the feature vectors of resources can be obtained through this model. Then, the cosine similarity is calculated between two resources using the feature vectors. The larger the value of the cosine similarity, the more similar the two resources are; in the recall stage, the n resources with the highest similarity to the historical accessed resources are retrieved for online recall, where n is a positive integer greater than 1.

[0033] However, considering that not every historical accessed resource is truly of interest to the user, and there are also differences in the degree of interest even among multiple historical accessed resources that the user is interested in. In view of this, if multiple filtering factors can be combined to screen out resources that are more in line with the user's characteristics for recommendation, it can objectively improve the accuracy of resource recommendation. Based on this inventive concept, as Figure 1 shown, a flowchart of a resource recommendation method provided by an embodiment of the first aspect of the present disclosure includes:

[0034] S101, filtering out the historical accessed resources that meet the preset conditions from the historical accessed resource set of the target object to obtain a reference resource set; the preset conditions are used to screen out the resources that the target object is not interested in.

[0035] In the embodiments of the present disclosure, the target object is a unique identifier of the user. The target object can be the user's account on the resource platform, or the target object can also be represented by the unique identifier of the terminal device used by the user. In the embodiments of the present disclosure, any information that can uniquely identify a user can be used as the target object.

[0036] The historical accessed resources can be the resources recently accessed by the target object. For example, the historical accessed resources in the recent one week or one month are used to construct a historical accessed resource set.

[0037] If the target object has been a user of the resource platform for a short time, all its historical accessed resources can be obtained to construct a historical accessed resource set.

[0038] The historical accessed resources within a specified access time period can be limited for constructing a historical accessed resource set, or the upper limit of the number of resources in the historical accessed resource set can be limited. Of course, it can also be required to obtain a specified number of historical accessed resources within a specified access time period to construct a historical accessed resource set. Therefore, this disclosure embodiment does not make specific limitations on which historical accessed resources are obtained to construct a historical accessed resource set.

[0039] Among them, during implementation, due to the differences in user behaviors existing in different resource platforms, in order to accurately screen out the historical accessed resources that the user is not interested in, specific preset conditions can be determined according to the business requirements of different resource platforms, thereby filtering out the historical accessed resources that the user is not interested in and obtaining a reference resource set constructed from the historical accessed resources that the user is interested in.

[0040] S102. Based on the user operation data of the target object for each reference resource in the reference resource set, determine the interest degree of the target object for each reference resource.

[0041] S103. Based on the interest degree of the target object for each reference resource, screen out multiple resources to be recommended from the similar resource sets of each reference resource.

[0042] S104. Recommend the multiple resources to be recommended to the target object.

[0043] Thus, in this disclosure embodiment, first obtain the historical accessed resource set of the target object. Then filter out the historical accessed resources that meet the preset conditions from it, thereby filtering out the historical accessed resources that the user is not interested in, completing the screening of the historical accessed resources, and obtaining a reference resource set that the target object is interested in. Then, in the reference resource set, further determine the interest degree of the target object for each reference resource based on the user operation data of the target object. Since the user operation data is the personalized data of the target object, the determined interest degree can reflect the interest degree difference of the target object for different reference resources. Therefore, based on the interest degree and similarity in this disclosure embodiment, the resources to be recommended can be accurately screened out. In short, the resources to be recommended suitable for the user can be obtained in this disclosure embodiment, thereby improving the accuracy of recommendation.

[0044] Since user behavior can directly and clearly reflect user preferences, therefore, in some embodiments, it can be based on, for example Figure 2The method shown filters out historical access resources that the user is not interested in:

[0045] S201. Obtain the user operation data of the target object for the historical access resources; the user operation data includes at least one first operation parameter.

[0046] For example, based on the behavior interface provided by the resource platform for users, the user behavior for resources may include at least one of the following: liking, giving a heart, disliking, downvoting, commenting, rewarding, collecting, rating, forwarding, sharing, downloading, purchasing resources, etc. Among them, user behaviors such as disliking and downvoting are used to express that the user is not interested in the resource. Therefore, the first operation parameter may include user behaviors such as disliking and downvoting.

[0047] In addition, the reading duration of the resource can also reflect the user's preference for the resource to a certain extent. Therefore, the reading duration can also be included in the first operation parameter.

[0048] S202. When any first operation parameter meets the filtering condition corresponding to the first operation parameter, filter out the historical access resource from the set of historical access resources.

[0049] In the embodiments of the present disclosure, the first operation parameter in the user operation behavior can measure whether the user is interested in the resource. Based on the first operation parameter, historical access resources that the user is not interested in are filtered out according to the user characteristics, and each first operation parameter has its corresponding filtering condition, so as to accurately filter out historical access resources that the user is not interested in based on the characteristics of the first operation parameter itself, thereby improving the accuracy of recommendation.

[0050] For the convenience of description, the first operation parameters generated by user behaviors such as disliking and downvoting can be collectively referred to as negative feedback parameters hereinafter. In short, the negative feedback parameter is used to characterize that the target object is not interested in the historical access resource.

[0051] During implementation, historical access resources that meet the preset conditions can be filtered out based on at least one of the reading duration, completion rate, and negative feedback parameter. A possible implementation method is to filter out historical access resources that meet the preset conditions by at least one of the following methods:

[0052] 1). When the first operation parameter includes the reading duration, filter out historical access resources with a reading duration lower than the duration threshold; that is, the filtering condition for the reading duration is: the reading duration is lower than the duration threshold.

[0053] Among them, the duration threshold can be defined according to the specific resource type in the specific resource platform. Generally, the duration threshold is set to be relatively small. For historical access resources with a reading duration lower than this duration threshold, they can be classified as resources that the user is not interested in.

[0054] 2) The reading duration of the user's historical accessed resources can be compared with the total playback duration of the historical accessed resources to obtain the completion rate. The completion rate can also be used to measure the target object's interest in video historical accessed resources. In the embodiments of the present disclosure, the completion rate can also be used to filter out the historical accessed resources that the user is not interested in.

[0055] Specifically, it can be implemented as filtering out the historical accessed resources with a completion rate lower than the completion rate threshold when the first operation parameter includes the completion rate; that is, the filtering condition of the completion rate is: the completion rate is lower than the completion rate threshold.

[0056] 3) When the first operation parameter includes a negative feedback parameter, filter out the historical accessed resources with the negative feedback parameter.

[0057] In addition, it should be noted that the above three methods can be used alternatively or in combination.

[0058] For example, one combination method is that if any historical accessed resource meets any of the above three situations, then the historical accessed resource needs to be filtered out.

[0059] Another implementation manner of filtering out the historical accessed resources that the target object is not interested in based on at least one of the reading duration, completion rate, and negative feedback parameter is: first, filter out the historical accessed resources that the user is not interested in (referred to as the first resource subset) by using the reading duration. Then, since the total durations of video historical accessed resources themselves are different, and some video historical accessed resources have short total durations, in order to avoid filtering out short videos that the user is actually interested in due to the limitation of the duration threshold, in the embodiments of the present disclosure, the historical accessed resources with a completion rate higher than the set threshold can be filtered out from the first resource subset to obtain the second resource subset. Thus, the historical accessed resources included in the second resource subset are those with a low completion rate and a short reading duration, and these historical accessed resources are all the historical accessed resources that the user is not interested in. In addition, the historical accessed resources with negative feedback parameters are filtered out from the historical accessed resource set to obtain the third resource subset of the historical accessed resources that the user is not interested in. Then finally, the union of the second resource subset and the third resource subset is the historical accessed resources that ultimately need to be filtered out.

[0060] Similarly, in addition to the negative feedback parameter, in the embodiments of the present disclosure, user behaviors such as liking, favoriting, rewarding, loving, sharing, forwarding, downloading, and purchasing resources, which can clearly express the user's interest in the resources, are referred to as positive feedback. Correspondingly, the user behavior parameters generated by such user behaviors can be referred to as positive feedback parameters. Another implementation manner of filtering out the historical access resources that the target object is not interested in based on at least one of the reading duration, completion rate, and negative feedback parameter is as follows: First, screen out the historical access resources with positive feedback parameters from the historical access resource set and store them in the reference resource set. Then, screen out the historical access resources with negative feedback parameters from the remaining historical access resources to obtain a fourth resource subset (i.e., a resource set composed of historical access resources that do not have positive feedback parameters and negative feedback parameters); afterwards, screen out the historical access resources with a reading duration less than the duration threshold in the fourth resource subset to obtain a fifth resource subset, and then screen out the historical access resources with a completion rate lower than the completion rate threshold from the fifth resource subset to obtain a sixth resource subset, and store the sixth resource subset in the reference resource set, thus completing the filtering.

[0061] Another implementation manner of filtering out the historical access resources that the target object is not interested in based on at least one of the reading duration, completion rate, and negative feedback parameter is as follows: Screen out the historical access resources with negative feedback parameters, and screen out the historical access resources with a reading duration less than the duration threshold and without positive feedback parameters, and also screen out the historical access resources with a completion rate lower than the completion rate threshold and without positive feedback parameters. That is, for the reading duration and completion rate, the historical access resources with a short reading duration but with positive feedback parameters will be retained in the reference resource set, and the historical access resources with a low completion rate but with positive feedback parameters will also be retained in the reference resource set.

[0062] Among them, the negative feedback parameter includes certain negative operation behaviors of the user for the historical access resources, and this parameter clearly reflects that the user has lost interest in such historical access resources and can directly filter out this historical access resource. The negative feedback parameter is used to reduce the exposure rate of resources of the same type as this historical access resource. That is, when the negative feedback parameter appears in the user's historical access resources, the similar resources of this historical access resource can also be marked, and at the same time, the recommendation of this type of resources on the user side is reduced.

[0063] In the embodiments of the present disclosure, the historical access resources that the user is not interested in can be accurately filtered out based on the reading duration, completion rate, and negative feedback parameter, thereby improving the accuracy of recommendation.

[0064] In the embodiments of the present disclosure, after screening the historical access resource set, a reference resource set of user interest is obtained. In order to further improve the accuracy of recommendation, the embodiments of the present disclosure adopt the following method to measure the difference in user interest in different reference resources, which can be implemented as follows Figure 3 as shown

[0065] S301, obtain the user operation data of the reference resource, where the user operation data includes at least one second operation parameter.

[0066] As elaborated above, similar to the first operation parameter, user behaviors such as rewarding, collecting, sharing, and downloading can clearly express that the user likes the historical access resource, and the user behavior parameters generated by such user behaviors are called positive feedback parameters. Among them, the second operation parameter may include positive feedback parameters, and may also include operation parameters of other user behaviors. For example, for comments, some comment contents clearly express the user's preference for the historical access resource, while some comments do not clearly express the preference for the resource. However, in implementation, in order to improve the accuracy of interest determination and make up for the possible incomplete filtering based on preset conditions, comments can be classified as the second operation parameter.

[0067] Since the obtained second operation parameter describes user behavior, and user behavior parameters cannot be directly recognized by the interest prediction model for relevant calculations, therefore, in the embodiments of the present disclosure, the second operation parameter needs to be quantified and converted into a digital type feature value for calculating the interest in the reference resource. Therefore, in S302, each second operation parameter is quantified into a feature value recognizable by the interest prediction model.

[0068] In the embodiments of the present disclosure, the value range of the second operation parameter may be a continuous value range. For example, for rewarding, the reward value is a value in a continuous value range. Of course, the second operation parameter may not have a continuous range. For example, the operation parameters of two user behaviors of liking and not liking do not have a continuous range. Even some user operation parameters do not have a value range, such as comments. Therefore, considering the differences between the second operation parameters, quantifying the second operation parameter into a feature value recognizable by the interest prediction model can be implemented as follows

[0069] When the value range of the second operation parameter is a continuous value range, normalize the value of the second operation parameter to obtain the feature value of the second operation parameter. For example, for the reading duration, the value of the second operation parameter can be normalized. Generally, it means normalizing the value range to the [0, 1] interval, and at this time, the normalized value of the second operation parameter is used as the feature value of the second operation parameter. In the embodiments of the present disclosure, normalization is beneficial to integrating different second operation parameters into one space and avoiding some second operation parameters from being too large and affecting the prediction result of interest.

[0070] For a second operation parameter without a value range, when the parameter value of the second operation parameter includes characters, semantic analysis is performed on the characters to obtain a semantic analysis result;

[0071] The situation where the parameter value of the second operation parameter includes characters generally refers to the analysis and processing of the comment content generated in the user's comment behavior. At this time, natural language processing tools can be used to perform relevant processing on the characters in the user comment content to obtain a semantic analysis result of the comment content. Here, except for comment content such as "don't want to see" and "refuse to push again", which clearly means that the user does not want to see this historical access resource again, regardless of whether the comment content is positive or negative objectively, as long as the comment content shows that the user has a strong interest in the historical access resource, it is understood as the historical access resource that the user is interested in. For example, if the user makes some negative evaluations similar to criticism or condemnation of the historical access resource, but also shows the user's interest in the historical access resource. Therefore, after obtaining the semantic analysis result, the eigenvalue of the second operation parameter can be determined based on the semantic analysis result. In implementation, when the obtained semantic analysis result expresses interest in the reference resource, the eigenvalue of the second operation parameter can be a first preset value, for example, 1. Similarly, when the obtained semantic analysis result expresses disinterest in the reference resource, the eigenvalue of the second operation parameter can be a second preset value, for example, -1.

[0072] Of course, for second operation parameters such as liking and giving a heart, if there is positive feedback, the eigenvalue is set to 1, and if there is no positive feedback, that is, no liking or giving a heart, the eigenvalue takes the value of -1.

[0073] In the embodiments of the present disclosure, each second operation parameter can be quantified into an eigenvalue recognizable by the interest prediction model. By performing normalization processing on the values of the second operation parameters, the values of the second operation parameters with smaller values can be prevented from being swallowed, and the evaluation criteria of the sample data are unified, avoiding the situation of inconsistent numerical evaluation criteria for the values of the second operation parameters before normalization, and improving the accuracy of the interest degree. By performing content analysis and processing on the second operation parameter containing characters, the analysis and judgment of the second operation data generated by the user behavior can be further enriched, facilitating the accurate determination of the interest degree in the reference resource.

[0074] S303, input the eigenvalues of the second operation parameters into the interest prediction model to obtain the interest degree of the target object in the reference resource.

[0075] In the embodiments of the present disclosure, based on the interest prediction model, the interest degrees of a user in each reference resource in the reference resource set can be obtained, so as to provide more conditions and bases for distinguishing the interest differences of the user in different reference resources based on the interest degrees and making resource recommendations based on the interest degrees, which is conducive to screening out the historical access resources that the user is more likely to be interested in according to the interest degrees.

[0076] In the embodiments of the present disclosure, the initial model is trained in a supervised manner, and the training label of the sample resource is whether the user is interested in the sample resource. The training label can be manually labeled, and for the sake of improving efficiency, the training label can also be automatically labeled. For example, the automatic labeling method can be implemented as follows:

[0077] When the user operation data of the sample resource includes a positive feedback parameter, determining the classification label of the sample resource as a positive sample;

[0078] When the user operation data of the sample resource includes a negative feedback parameter, determining the classification label of the sample resource as a negative sample.

[0079] A positive sample is a resource that the user is interested in; a negative sample is a resource that the user is not interested in. Then, the initial model can be trained by means of binary classification to obtain an interest prediction model. When implementing, the initial model can be trained based on the following method to obtain an interest prediction model:

[0080] First, obtain the eigenvalue of the second operation parameter of the sample resource; wherein, the method for converting the second operation parameter of the sample resource into an eigenvalue is the same as that described in the foregoing step S302, which will not be elaborated here. Of course, for the sake of improving the training accuracy, in the embodiments of the present application, the reference resource set obtained after filtering out the resources that the user is not interested in can also be used as a training sample.

[0081] After that, input the eigenvalue of the second operation parameter of the sample resource into the initial model to obtain the interest degree of the sample resource; wherein, the initial model can be a linear regression model. Of course, in order to better process the quantified discrete features, an xgbt (eXtreme Gradient Boosting) model can also be selected.

[0082] Then, based on the obtained interest degree, continue to perform binary classification on the sample resource to obtain the classification result of the sample resource; wherein, the classification result is a positive sample or a negative sample; after that, based on the classification result of the sample resource and the classification label of the sample resource, determine the loss value; then, based on the loss value, adjust the initial model to obtain an interest prediction model.

[0083] Among them, the convergence condition for training can be to terminate the training when a preset number of iterations is reached to obtain the interest prediction model, or to terminate the training when the loss value remains basically unchanged to obtain the interest prediction model.

[0084] In the embodiments of the present disclosure, training based on a supervised manner can obtain an accurate interest prediction model. Compared with using the interest in a continuous value range as the training label, using the binary classification result as the training label is more accurate and easier to label the training data.

[0085] In addition, in the embodiments of the present disclosure, in addition to using the user behavior of the target object for each reference resource to determine the interest, posterior data can also be combined to improve the accuracy of the interest. For example, for any reference resource, the resource type of the reference resource can be obtained; then, the number of clicks of the target object on the resource type is obtained. For example, if the reference resource is a sports resource, the number of clicks of the target object on the sports resources can be aggregated and counted as posterior data. Then, based on the number of clicks and the second operation parameter of the reference resource, the interest of the target object in the reference resource is determined.

[0086] In summary, the implementation manner of predicting the interest in the reference resource in the embodiments of the present disclosure is as Figure 4 shown. Quantize the second operation parameters including likes, hearts, rewards, shares, comments, completion rate, downloads, forwards, etc. of the reference resource A into corresponding feature values and input them to the interest prediction model. At the same time, obtain the number of clicks of the same type of resources of the reference resource A and also input them to the interest prediction model, so as to obtain the interest of the target object in the reference resource A.

[0087] Thus, in the embodiments of the present disclosure, based on the number of clicks of the target object on the same type of resources, the interest of the target object in the reference resource is predicted based on posterior data, which can improve the accuracy of interest determination and thus improve the accuracy of the recommendation result.

[0088] The above introduces how to screen out the reference resources and determine the interest of the target object in different reference resources. Based on this interest, in the embodiments of the present disclosure, the recall value of the similar resources can be obtained based on the interest of the target object in the reference resource and the similarity between the similar resources and the reference resources; based on the screening principle of preferentially selecting the similar resources with a high recall value, multiple resources to be recommended are screened out.

[0089] Based on the screening principle of preferentially selecting the similar resources with a high recall value, the similar resources with a high recall value in the set of similar resources are preferentially selected, and multiple resources to be recommended are screened out, which together with the reference resource set form a set of resources to be recommended.

[0090] Among them, the recall value refers to the ultimate reference value basis for sorting the resources to be recommended in the set of resources to be recommended. In the embodiments of the present disclosure, since the degree of interest can measure the interest differences of different reference resources, and the reference resources are obtained by screening the historical access resource set, the introduced degree of interest can well reflect the interests and hobbies of the target object. By determining the recall value based on the degree of interest and similarity, compared with the method of simply recalling resources using similarity, the accuracy of recall can be improved, thereby improving the accuracy of recommendation.

[0091] In a possible implementation manner, each resource can determine the set of similar resources of each resource based on the ICF (item-based collaborative filtering) recall method. For example, an index of resources is constructed based on ICF for online recall. An example of the index is as follows:

[0092] nid1[nid2&0.91,nid3&0.82,nid4&0.71,…]

[0093] nid2[nid1&0.91,nid4&0.85,nid5&0.67,…]

[0094] Among them, each line is an index. The first column is the trigger key (for example, nid1 in the first row is the trigger key), and the second column is the information of multiple resources. For example, in the second column of the first row, nid2&0.91 means that the cosine similarity between nid2 and nid1 is 0.91, and the resources with higher cosine similarity are ranked earlier. Generally, only the top 50 resources with the highest cosine similarity are retained, and a threshold of cosine similarity will be set, and the relationships below this threshold will not be placed in the index.

[0095] During implementation, the recall value of the similar resource can be finally determined based on the degree of interest in the reference resource and the similarity between the reference resource and the similar resource.

[0096] One method for determining the recall value is: determine the product of the degree of interest and the similarity to obtain the recall value of the similar resource. In this embodiment, the similarity and the degree of interest are multiplied to obtain a recall value with relatively high credibility, and this method is simple to calculate and consumes less computing resources.

[0097] Another method for determining the recall value is: use one piece of information in the degree of interest and similarity parameter as the exponent of the other piece of information to determine the recall value between the similar resource and the reference resource.

[0098] For example, when the similarity is used as the exponent, the calculation formula is as shown in Equation (1):

[0099] Z = C L (1)

[0100] In formula (1), Z is the recall value, C is the interest degree, and L is the similarity degree.

[0101] It should be noted that in the embodiments of the present disclosure, the recall value has a positive correlation with both the similarity degree and the interest degree. Any method for calculating the recall value designed based on this positive correlation is applicable to the embodiments of the present disclosure. When determining the recall value in the exponential form shown in formula (1), one of the similarity degree and the interest degree can be emphasized. When the similarity degree is used as the exponent, the purpose of resource recommendation with a greater emphasis on the similarity degree index is achieved. When the interest degree is used as the exponent, the purpose of resource recommendation with a greater emphasis on the interest degree index is achieved.

[0102] In some other embodiments, the interest degree can not only be used for resource recall, but also the interest degree determined in the embodiments of the present disclosure can be used in the sorting stage. For example, after screening out multiple resources to be recommended, the interest degree can be used as an input feature of the sorting model, so that the sorting model can perform resource sorting. The sorting model in the embodiments of the present disclosure can be a two-tower model. Thus, when using the interest degree for sorting, the interest and hobbies of the user for the reference resources can be considered, improving the accuracy of sorting and thus the accuracy of recommendation.

[0103] Based on the same technical concept, in the second aspect of the embodiments of the present disclosure, a resource recommendation device is further provided, as Figure 5 shown, including:

[0104] A filtering module 501, configured to filter out historical accessed resources that meet preset conditions from the historical accessed resource set of the target object to obtain a reference resource set; the preset conditions are used to screen out resources that the target object is not interested in;

[0105] A determination module 502, configured to determine the interest degree of the target object for each reference resource in the reference resource set based on the user operation data of the target object for each reference resource in the reference resource set;

[0106] A screening module 503, configured to screen out multiple resources to be recommended from the similar resource sets of each reference resource based on the interest degree of the target object for each reference resource;

[0107] A recommendation module 504, configured to recommend the multiple resources to be recommended to the target object.

[0108] In some embodiments, on the basis of Figure 5 , as Figure 6 shown, wherein, the filtering module 501 includes:

[0109] A first filtering sub-module 505, configured to obtain the user operation data of the target object for the historical accessed resources; the user operation data includes at least one first operation parameter;

[0110] The second filtering sub-module 506 is configured to filter out historical accessed resources from the set of historical accessed resources when any first operation parameter meets the filtering condition corresponding to the first operation parameter.

[0111] In some embodiments, when any first operation parameter meets the filtering condition corresponding to the first operation parameter, the second filtering sub-module 506 is configured to:

[0112] Filter out historical accessed resources that meet the preset conditions based on at least one of the following methods:

[0113] When the first operation parameter includes the reading duration, filter out historical accessed resources with a reading duration lower than the duration threshold;

[0114] When the first operation parameter includes the completion rate, filter out historical accessed resources with a completion rate lower than the completion rate threshold;

[0115] When the first operation parameter includes a negative feedback parameter, filter out historical accessed resources with the negative feedback parameter, where the negative feedback parameter is used to indicate that the target object is not interested in the historical access.

[0116] In some embodiments, the determining module 502 is configured to:

[0117] Perform the following operations for each reference resource respectively to obtain the interest degree of each reference resource:

[0118] Obtain the user operation data of the reference resource, where the user operation data includes at least one second operation parameter;

[0119] Quantize each second operation parameter into a feature value recognizable by the interest degree prediction model;

[0120] Input the feature values of each second operation parameter into the interest degree prediction model to obtain the interest degree of the target object in the reference resource.

[0121] In some embodiments, as Figure 6 shown, the resource recommendation device further includes:

[0122] A training module 507, configured to train an initial model based on the following method to obtain an interest degree prediction model:

[0123] Obtain the feature values of the second operation parameters of the sample resources;

[0124] Input the feature values of the second operation parameters of the sample resources into the initial model to obtain the interest degree of the sample resources;

[0125] Based on the interest degree, perform binary classification on the sample resources to obtain the classification result of the sample resources; wherein, the classification result is a positive sample or a negative sample;

[0126] Determine a loss value based on the classification result of the sample resources and the classification label of the sample resources;

[0127] Adjust the initial model based on the loss value to obtain an interest prediction model.

[0128] In some embodiments, the determining module 502 quantizes each second operation parameter into a feature value recognizable by the interest prediction model based on at least one of the following methods, including:

[0129] When the value range of the second operation parameter is a continuous value range, normalize the value of the second operation parameter to obtain the feature value of the second operation parameter;

[0130] When the parameter value of the second operation parameter includes characters, perform semantic analysis on the characters to obtain a semantic analysis result; based on the semantic analysis result, determine the feature value of the second operation parameter.

[0131] In some embodiments, the screening module 503 is configured to:

[0132] Obtain a recall value of the similar resources based on the interest of the target object in the reference resources and the similarity between the similar resources and the reference resources;

[0133] Screen out multiple resources to be recommended based on the screening principle of preferentially selecting similar resources with a high recall value.

[0134] In some embodiments, the determining module 502 is further configured to:

[0135] For any reference resource, obtain the resource type of the reference resource;

[0136] Obtain the number of clicks of the target object on the resource type;

[0137] Determine the interest of the target object in the reference resource based on the number of clicks and the user operation parameter of the reference resource.

[0138] For the specific functions and examples of each module and sub-module of the device according to the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the above method embodiments, which will not be elaborated herein.

[0139] According to the embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0140] Figure 7FIG. 0 shows a schematic block diagram of an exemplary electronic device 700 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0141] As Figure 7 shown, the device 700 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0142] A plurality of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as, for example, a keyboard, a mouse, etc.; an output unit 707, such as, for example, various types of displays, speakers, etc.; a storage unit 708, such as, for example, a magnetic disk, an optical disk, etc.; and a communication unit 709, such as, for example, a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0143] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated 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 701 executes the various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the resource recommendation method described above can be executed. Alternatively, in other embodiments, the computing unit 701 can be configured to execute the resource recommendation method in any other suitable way (e.g., by means of firmware).

[0144] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), systems-on-a-chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from a storage system, at least one input device, and at least one output device, and transmits the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0145] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0146] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection 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 include, 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 a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0147] To provide for interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide for 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 acoustic, speech, or tactile input).

[0148] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0149] A computer system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0150] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and no limitations are imposed herein.

[0151] The above specific embodiments do not constitute a limitation on the protection scope 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 principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A resource recommendation method, comprising: Filtering out historical accessed resources that meet preset conditions from the historical accessed resource set of the target object to obtain a reference resource set; The preset conditions are used to screen out resources that the target object is not interested in; this step includes: Obtaining a first operation parameter of the target object for the historical accessed resources; the first operation parameter includes reading duration, completion rate, and negative feedback parameter; the negative feedback parameter is used to represent that the target object is not interested in the historical accessed resource; Filtering out historical accessed resources with a reading duration lower than the duration threshold to obtain a first resource subset; Screening out historical accessed resources with a completion rate higher than the set threshold from the first resource subset to obtain a second resource subset; Screening out historical accessed resources with negative feedback parameters from the historical accessed resource set to obtain a third resource subset that the user is not interested in; wherein, the union of the second resource subset and the third resource subset is the historical accessed resources that ultimately need to be filtered out; Based on the user operation data of the target object for each reference resource in the reference resource set, determining the degree of interest of the target object in each reference resource; Based on the degree of interest of the target object in each reference resource, screening out multiple resources to be recommended from the similar resource sets of each reference resource; Recommending the multiple resources to be recommended to the target object.

2. The method according to claim 1, wherein The determining the degree of interest of the target object in each reference resource based on the user operation data of the target object for each reference resource in the reference resource set includes: Performing the following operations for each reference resource respectively to obtain the degree of interest of each reference resource: Obtaining the user operation data of the reference resource, where the user operation data includes at least one second operation parameter; Quantifying each of the second operation parameters into a feature value recognizable by the interest degree prediction model; Inputting the feature values of each second operation parameter into the interest degree prediction model to obtain the degree of interest of the target object in the reference resource.

3. The method according to claim 2, further comprising training an initial model based on the following method to obtain the interest degree prediction model: Obtaining the feature values of the second operation parameters of the sample resources; Inputting the feature values of the second operation parameters of the sample resources into the initial model to obtain the degree of interest of the sample resources; Based on the degree of interest, performing binary classification on the sample resources to obtain the classification result of the sample resources; wherein, the classification result is a positive sample or a negative sample; Based on the classification result of the sample resources and the classification label of the sample resources, determining the loss value; Adjusting the initial model based on the loss value to obtain the interest degree prediction model.

4. The method according to claim 2, wherein, Quantifying each of the second operation parameters into a feature value recognizable by the interest degree prediction model based on at least one of the following methods, including: In the case where the value range of the second operation parameter is a continuous value range, normalizing the value of the second operation parameter to obtain the feature value of the second operation parameter; In the case where the parameter value of the second operation parameter includes characters, performing semantic analysis on the characters to obtain a semantic analysis result; based on the semantic analysis result, determining the feature value of the second operation parameter.

5. The method according to any one of claims 1-4, wherein, Screening out resources to be recommended from the similar resource sets of each reference resource based on the degree of interest of the target object in each reference resource, including: Obtaining the recall value of the similar resource based on the degree of interest of the target object in the reference resource and the similarity between the similar resource and the reference resource; Screening out multiple resources to be recommended based on the screening principle of preferentially selecting similar resources with high recall values.

6. The method according to claim 1 or 2, wherein determining the degree of interest of the target object in each reference resource based on the user operation data of the target object for each reference resource in the reference resource set includes: For any reference resource, obtaining the resource type of the reference resource; Obtaining the number of clicks of the target object on the resource type; Determining the degree of interest of the target object in the reference resource based on the number of clicks and the user operation parameters of the reference resource.

7. A resource recommendation device, including: A filtering module, configured to filter out historical accessed resources that meet preset conditions from the historical accessed resource set of the target object to obtain a reference resource set; The preset conditions are used to screen out resources that the target object is not interested in; The filtering module includes: A first filtering sub-module, configured to obtain the first operation parameters of the target object for the historical accessed resources; the first operation parameters include reading duration, completion rate, and negative feedback parameters; the negative feedback parameters are used to characterize that the target object is not interested in the historical accessed resources; A second filtering sub-module, configured to filter out historical accessed resources with a reading duration lower than a duration threshold to obtain a first resource subset; Screening out historical accessed resources with a completion rate higher than a set threshold from the first resource subset to obtain a second resource subset; Screening out historical accessed resources with negative feedback parameters from the historical accessed resource set to obtain a third resource subset of resources that the user is not interested in; wherein, the union of the second resource subset and the third resource subset is the historical accessed resources that finally need to be filtered out; A determination module, configured to determine the degree of interest of the target object in each reference resource based on the user operation data of the target object for each reference resource in the reference resource set; A screening module, configured to screen out multiple resources to be recommended from the similar resource sets of each reference resource based on the degree of interest of the target object in each reference resource; A recommendation module, configured to recommend the multiple resources to be recommended to the target object.

8. The apparatus according to claim 7, wherein The determination module is configured to: Perform the following operations for each reference resource respectively to obtain the degree of interest of each reference resource: Obtaining the user operation data of the reference resource, where the user operation data includes at least one second operation parameter; Quantifying each of the second operation parameters into a feature value recognizable by an interest degree prediction model; Inputting the feature values of each second operation parameter into the interest degree prediction model to obtain the degree of interest of the target object in the reference resource.

9. The device according to claim 8, further including: A training module, configured to train an initial model based on the following method to obtain the interest degree prediction model: Obtaining the feature values of the second operation parameters of the sample resources; Input the eigenvalue of the second operation parameter of the sample resource into the initial model to obtain the interest degree of the sample resource; Based on the interest degree, perform binary classification on the sample resource to obtain the classification result of the sample resource; wherein, the classification result is a positive sample or a negative sample; Determine the loss value based on the classification result of the sample resource and the classification label of the sample resource; Adjust the initial model based on the loss value to obtain the interest degree prediction model.

10. The apparatus according to claim 8, wherein, The determination module quantizes each of the second operation parameters into a feature value recognizable by the interest degree prediction model based on at least one of the following methods, including: When the value range of the second operation parameter is a continuous value range, normalize the value of the second operation parameter to obtain the eigenvalue of the second operation parameter; When the parameter value of the second operation parameter includes characters, perform semantic analysis on the characters to obtain a semantic analysis result; based on the semantic analysis result, determine the eigenvalue of the second operation parameter.

11. The device according to any one of claims 7-10, wherein The screening module is used for: Based on the interest degree of the target object in the reference resource and the similarity between the similar resource and the reference resource, obtain the recall value of the similar resource; Based on the screening principle of preferentially selecting similar resources with a high recall value, screen out a plurality of resources to be recommended.

12. The apparatus according to claim 7 or 8, wherein the determination module is further configured to: For any reference resource, obtain the resource type of the reference resource; Obtain the number of clicks of the target object on the resource type; Based on the number of clicks and the user operation parameter of the reference resource, determine the interest degree of the target object in the reference resource.

13. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1-6.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the method according to any one of claims 1-6.

15. A computer program product, comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-6.

Images