Fatigue feeling pre-estimation model training method and device and fatigue feeling pre-estimation method and device

Abstract

The invention provides a fatigue feeling prediction model training method and device and a fatigue feeling prediction method and device, and relates to the artificial intelligence fields of deep learning, large models, natural language processing, intelligent recommendation and the like. The fatigue feeling prediction model training method can comprise the steps of obtaining a historical consumption record corresponding to a predetermined resource recommendation application, wherein the historical consumption record comprises operation information of a recommendation object for opening the resource recommendation application for a recommended resource object; generating a training sample set according to the historical consumption record; according to the training sample set, a fatigue feeling prediction model is trained, and the fatigue feeling prediction model is used for determining a fatigue feeling prediction result of a target recommendation object for candidate resources to be recommended when the resources are recommended to the target recommendation object to be processed.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to fatigue prediction model training and fatigue prediction methods and devices in the fields of deep learning, large models, natural language processing, and intelligent recommendation. Background Art

[0002] At present, short video platforms are loved by more and more users. When recommending videos to users, it may involve multiple stages such as recall, rough sorting, fine sorting, and re-sorting. Among them, for the multi-objective fusion operation in the fine sorting stage, the modeled targets can be divided into two categories: positive feedback targets and negative feedback targets. Among them, positive feedback targets include satisfaction, completion rate, and completion of broadcasts, etc., while there are relatively few studies on negative feedback target modeling, which affects the accuracy of the fusion results. Summary of the invention

[0003] The present disclosure provides fatigue prediction model training and fatigue prediction method and device.

[0004] A fatigue prediction model training method, comprising:

[0005] Acquire a historical consumption record corresponding to a predetermined resource recommendation application, wherein the historical consumption record includes: operation information of opening a recommended object of the resource recommendation application for the recommended resource object;

[0006] Generate a training sample set according to the historical consumption records;

[0007] The fatigue prediction model is trained according to the training sample set, and the fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resource to be recommended when recommending resources to the target recommendation object to be processed.

[0008] A fatigue estimation method, comprising:

[0009] Determine candidate resources to be recommended for a target recommendation object of opening a resource recommendation application;

[0010] The fatigue prediction model is used to determine the fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on the historical consumption records corresponding to the resource recommendation application. The historical consumption records include: the operation information of the recommended object who opens the resource recommendation application for the recommended resource object.

[0011] A fatigue prediction model training device, comprising: an information acquisition module, a sample construction module and a model training module;

[0012] The information acquisition module is used to acquire the historical consumption record corresponding to the predetermined resource recommendation application, wherein the historical consumption record includes: operation information of opening the recommended object of the resource recommendation application for the recommended resource object;

[0013] The sample construction module is used to generate a training sample set according to the historical consumption records;

[0014] The model training module is used to train the fatigue prediction model according to the training sample set. The fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resources to be recommended when recommending resources to the target recommendation object to be processed.

[0015] A fatigue prediction device includes: a resource determination module and a result determination module;

[0016] The resource determination module is used to determine candidate resources to be recommended for a target recommendation object of opening a resource recommendation application;

[0017] The result determination module is used to use a fatigue prediction model to determine the fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on historical consumption records corresponding to the resource recommendation application. The historical consumption records include: operation information of the recommended object of the resource recommendation application for the recommended resource object.

[0018] An electronic device, comprising:

[0019] at least one processor; and

[0020] a memory communicatively connected to the at least one processor; wherein,

[0021] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0022] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the method as described above.

[0023] A computer program product comprises a computer program / instruction, wherein the computer program / instruction implements the method described above when executed by a processor.

[0024] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended 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

[0025] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0026] Figure 1 This is a flow chart of an embodiment of the fatigue prediction model training method disclosed in the present invention;

[0027] Figure 2 This is a flow chart of an embodiment of the fatigue prediction method disclosed in the present invention;

[0028] Figure 3 A schematic diagram of the overall implementation process of the fatigue prediction model training and fatigue prediction method disclosed in the present invention;

[0029] Figure 4 Schematic diagram of the structure of the fatigue prediction model training device embodiment 400 of the present disclosure;

[0030] Figure 5 Schematic diagram of the structure of the fatigue prediction device embodiment 500 of the present disclosure;

[0031] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0032] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0033] In addition, it should be understood that the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0034] Figure 1 Flow chart of an embodiment of the fatigue prediction model training method disclosed in the present invention. Figure 1 As shown, the following specific implementation methods are included.

[0035] In step 101, a historical consumption record corresponding to a predetermined resource recommendation application is obtained, wherein the historical consumption record includes: operation information of opening a recommended object of the resource recommendation application for the recommended resource object.

[0036] In step 102, a training sample set is generated according to the historical consumption records.

[0037] In step 103, a fatigue prediction model is trained according to the training sample set. The fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resource to be recommended when recommending resources to the target recommendation object to be processed.

[0038] By adopting the scheme described in the above method embodiment, the fatigue prediction model can be trained using the training sample set, and then when recommending resources to the target recommendation object, the trained fatigue prediction model can be used to determine the fatigue prediction result of the target recommendation object for the candidate resources to be recommended. The fatigue prediction result can be used as a negative feedback target, thereby enriching the content of the existing negative feedback target and improving the accuracy of the fusion result.

[0039] The recommended object usually refers to the user, and accordingly, the operation information refers to the user behavior information.

[0040] For a predetermined resource recommendation application, historical consumption records can be obtained. Generally speaking, after a recommended object opens a resource recommendation application, a resource object will be recommended to the recommended object. For each recommended resource object, a corresponding consumption record will be generated. The consumption record may include the recommended object's operation information on the resource object. The specific content of the operation information can be determined according to actual needs.

[0041] The historical consumption record refers to the consumption record generated before the current time. In addition, obtaining the historical consumption record may refer to obtaining the historical consumption record generated within the most recent predetermined time period, and the specific value of the most recent predetermined time period is determined according to actual needs.

[0042] According to the acquired historical consumption records, a training sample set corresponding to the fatigue prediction model can be generated, and the training sample set includes positive samples and negative samples.

[0043] In some embodiments of the present disclosure, the resource recommendation application may include: a short video platform, and accordingly, the resource object type may be a video type. Each historical consumption record may correspond to a recommendation object and a recommended first video, respectively. The method of generating a training sample set according to the historical consumption records may include: in response to determining that there is a second video in the first video that meets any one of the first negative sample filtering condition, the second negative sample filtering condition, and the third negative sample filtering condition, determining that the negative sample video includes the second video, in response to determining that there is a third video in the first video that meets the positive sample filtering condition, determining that the positive sample video includes the third video, generating negative samples in the training sample set according to the negative sample videos, and generating positive samples in the training sample set according to the positive sample videos.

[0044] That is, the negative sample videos and the positive sample videos can be screened out from the first video corresponding to each historical consumption record, and then the negative samples and the positive samples in the training sample set can be generated according to the negative sample videos and the positive sample videos, respectively. The whole process is simple and convenient to implement, thus laying a good foundation for subsequent processing.

[0045] Among them, the screening methods of negative sample videos may include method one, method two and method three, which are introduced below respectively.

[0046] 1) Method 1

[0047] In some embodiments of the present disclosure, any fourth video (a video corresponding to any historical consumption record) can be determined in the first video, and the first playback time of the fourth video can be determined based on the corresponding historical consumption record. In response to determining that the first playback time is less than the first threshold, it can be determined that the fourth video meets the first negative sample screening condition, and the fourth video can be determined as a negative sample video.

[0048] The historical consumption record includes the video playback time information, that is, the recommended object's viewing time information for the video. If it is determined that the playback time of the fourth video is less than the first threshold, it can be determined that the fourth video meets the first negative sample screening condition, and the fourth video can be determined as a negative sample video.

[0049] The specific value of the first threshold can be determined according to actual needs, such as 3 seconds. If the playback time of the fourth video is less than 3 seconds, the fourth video can be considered as a fast-scrolling video. Fast-scrolling is usually caused by the recommendation object not being interested in the fourth video. Therefore, the fourth video in this case can be determined as a negative sample video.

[0050] 2) Method 2

[0051] In some embodiments of the present disclosure, any fourth video can be determined in the first video, and a sixth video whose video duration belongs to the same duration interval as the fourth video can be obtained from the fifth video, the fifth video includes other videos in the first video except the fourth video, and the target playback duration of each video in the video set is determined according to the corresponding historical consumption records. The video set includes the fourth video and the sixth video, and the sixth video whose video duration belongs to the same duration interval as the fourth video can be selected from the videos corresponding to other historical consumption records, and the video set is composed of the selected sixth video and the fourth video. The target playback time of each video in the video set can be determined according to the corresponding historical consumption records, and the videos in the video set can be sorted in descending order according to the target playback time. Accordingly, in response to determining that there is a jump-out operation for the fourth video, and determining that the sorting position number of the fourth video is greater than the product of M and x, it can be determined that the fourth video meets the second negative sample screening condition, and the fourth video can be determined as a negative sample video, M represents the number of videos in the video set, x represents the first percentile greater than 0 and less than 1, and the existence of a jump-out operation means that the recommended object corresponding to the historical consumption record exits the short video platform when playing the fourth video.

[0052] According to the video duration of each video, multiple different duration intervals can be set in advance, and there is no overlap between any two duration intervals, and the video duration of each video belongs to a duration interval. The video duration is usually measured in seconds.

[0053] Assuming that 99 sixth videos are selected in total, then these 99 sixth videos and the fourth video group video set can be used, and then the 100 videos in the video set can be sorted in descending order according to the playback time recorded in the historical consumption records corresponding to each of the 100 videos. Assuming that the value of the first percentage is 20%, then if it is determined that the corresponding recommended object has exited the short video platform when the fourth video is being played, and it is determined that the ranking position of the fourth video is after the 20th (100*20%) position, then it can be determined that the fourth video meets the second negative sample screening condition, and the fourth video can be determined as a negative sample video.

[0054] Negative sample videos that meet the second negative sample screening condition can be called confident pop-out videos. In other words, confident pop-out videos refer to videos with pop-out operations and unsatisfactory distribution. Satisfactory distribution can be defined as: the playback time ranking is in the TOP20% (i.e. the top 20%) among all videos with the same video duration range. Correspondingly, unsatisfactory distribution refers to the bottom 80%.

[0055] 3) Method 3

[0056] In some embodiments of the present disclosure, any fourth video can be determined in the first video, and a video set can be obtained and the videos in the video set can be sorted in the manner described in 2). In addition, the number of videos belonging to the same category as the fourth video that have been recommended before the fourth video in the same session can be counted, and the session refers to the consumption process of the recommended object corresponding to the historical consumption record from opening the short video platform to exiting the short video platform. Accordingly, in response to determining that the following conditions are met at the same time: the number of videos is greater than the second threshold, the sorting position number of the fourth video is greater than the product of M and y, and the video length of the fourth video is less than the third threshold, it can be determined that the fourth video meets the third negative sample screening condition, and the fourth video can be determined as a negative sample video, and y represents a second percentile greater than 0 and less than 1.

[0057] The specific values ​​of the second threshold, the third threshold and the second percentage can be determined according to actual needs.

[0058] A session refers to the consumption process of the recommendation object from opening the short video platform to exiting the short video platform. The consumption process refers to the complete session process. During this process, the recommendation object may watch multiple recommended videos and generate corresponding operation information for each of the recommended videos.

[0059] In addition, how to classify each video into different categories and which specific categories are included can also be determined according to actual needs. For example, the categories may include sports, entertainment, military, etc.

[0060] Assuming that the recommended object corresponding to the fourth video is recommended object a, and assuming that the value of the second threshold is 3, the value of the third threshold is 100 seconds, and the second percentage is 50%, then the number of videos belonging to the same category as the fourth video that have been recommended to recommended object a before the fourth video is recommended to recommended object a in the same session can be counted, assuming that it is 4, and assuming that the sorting position number of the fourth video is greater than M*50%, and assuming that the video length of the fourth video is less than 100 seconds, then since the following conditions are met at the same time: the number of videos 4 is greater than 3, the sorting position number of the fourth video is greater than M*50%, and the video length of the fourth video is less than 100 seconds, it can be determined that the fourth video meets the third negative sample screening condition, and the fourth video can be determined as a negative sample video.

[0061] Negative sample videos that meet the third negative sample screening condition can be called fatigue videos. In other words, fatigue videos refer to videos of the same category that are consumed forward in the same session and exceed the second threshold and are unsatisfactory. Dissatisfaction can be defined as: the playback time ranking is after TOP50% of all videos in the same duration range and the video duration is less than 100 seconds.

[0062] By adopting the processing methods in the above-mentioned method one, method two and method three, fast-swiping videos, confident pop-up videos and fatigue videos can all be determined as negative sample videos, that is, negative sample videos can be determined by combining significant negative operations such as fast-swiping operations, pop-up operations, and fatigue operations after multiple consumption of the same category videos in the same session, thereby realizing joint modeling of multiple negative operations, thereby enriching the expression ability of fatigue, improving the learning ability of the model, and correspondingly improving the accuracy of fatigue estimation results generated by the subsequent model.

[0063] In addition, in some embodiments of the present disclosure, any fourth video can be determined in the first video, and a video set can be obtained and the videos in the video set can be sorted in the manner described in 2). Accordingly, in response to determining that a predetermined negative operation has not occurred for the fourth video, and determining that the sorting position number of the fourth video is less than the product of M and z, it can be determined that the fourth video meets the positive sample screening condition, and the fourth video can be determined as a positive sample video, and z represents the third percentile greater than 0 and less than 1.

[0064] The specific value of the third percentile can be determined according to actual needs, such as 50%. If it is determined that the predetermined negative operation has not occurred for the fourth video, and it is determined that the ranking position number of the fourth video is less than the product of M*50%, it can be determined that the fourth video meets the positive sample screening condition, and the fourth video can be determined as a positive sample video. The predetermined negative operation may include the aforementioned fast sliding operation, jump operation, and fatigue operation.

[0065] Based on the determined negative sample videos, negative samples can be generated. In addition, through the above processing, positive sample videos can be determined from the videos corresponding to each historical consumption record, and positive samples can be generated based on the positive sample videos. The model can then be trained in combination with the positive samples and negative samples, thereby improving the training accuracy of the model, and thereby improving the accuracy of the fatigue prediction results subsequently generated by the model.

[0066] In addition, from the above introduction, it can be seen that the fourth video can be any video in the first video. When the fourth video meets the first negative sample screening condition, the second negative sample screening condition or the third negative sample screening condition, the fourth video is the second video. When the fourth video meets the positive sample screening condition, the fourth video is the third video.

[0067] Accordingly, in some embodiments of the present disclosure, corresponding negative samples can be generated for each negative sample video, and each negative sample can include: first feature information and a first label corresponding to the corresponding negative sample video, and the first label is used to indicate that the corresponding sample is a negative sample. In addition, corresponding positive samples can be generated for each positive sample video, and each positive sample can include: second feature information and a second label corresponding to the corresponding positive sample video, and the second label is used to indicate that the corresponding sample is a positive sample.

[0068] In practical applications, before generating positive and negative samples, optimization processing can be performed on the determined negative sample videos and positive sample videos, such as discarding some negative sample videos or discarding some positive sample videos according to the ratio requirement of positive and negative samples.

[0069] Furthermore, a corresponding negative sample can be generated for each negative sample video, and a corresponding positive sample can be generated for each positive sample video. Each negative sample includes first feature information and a first label, and each positive sample includes second feature information and a second label. For example, the value of the first label can be 0, and the value of the second label can be 1, or vice versa, as long as the model can understand the meaning of the corresponding label.

[0070] In some embodiments of the present disclosure, the first feature information and the second feature information may be obtained in the same manner, such as may include: for a target video, the target video is any negative sample video or any positive sample video, obtaining object features of a recommended object corresponding to the target video, the recommended object corresponding to the target video is a recommended object corresponding to the same historical consumption record as the target video, obtaining video features of the target video, obtaining cross-features of the target video and the recommended object corresponding to the target video, obtaining request features, wherein the request is a request to trigger the recommendation of the target video, obtaining historical session features of the recommended object corresponding to the target video, the session refers to the consumption process of the recommended object corresponding to the target video from opening the short video platform to exiting the short video platform, and determining the obtained object features, video features, cross-features, request features and historical session features as feature information corresponding to the target video.

[0071] The target video may refer to a negative sample video or a positive sample video, and the corresponding feature information is obtained in the same way.

[0072] The specific contents included in each of the characteristic information may be determined according to actual needs. In addition, the specific methods for obtaining the various characteristic information may also be determined according to actual needs.

[0073] It should be noted that the various characteristic information and historical consumption records involved in the solution described in this disclosure can be obtained through various public, legal and compliant methods, such as obtaining from the user with the user's authorization, and obtaining with the user's knowledge and consent. In the technical solution of this disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0074] Among them, the object characteristics may include: the recommended object's age, gender, education level, life stage (such as going to school, working, etc.), activity level in the short video platform (such as high, medium, and low), and whether the user is a new user.

[0075] Video features may include: the length of the target video, the category to which the target video belongs, the semantic features of the category, and the market experience of the category (such as average satisfaction, average completion rate, average playback time, etc.).

[0076] Cross-features may include: whether the category to which the target video belongs is a category that the recommendation object is interested in, etc. For example, the category of interest may be determined based on the user profile of the recommendation object.

[0077] The request may refer to a user swiping up on the screen, etc. Accordingly, the request features may include: which request is the nth request in the same session that triggers the recommendation of the target video, the device information used by the recommendation object (such as a mobile phone or a tablet, etc.), the time period when the request is issued to trigger the recommendation of the target video (such as morning, evening, etc.), and the date type when the request is issued to trigger the recommendation of the target video (such as a weekday, holiday, etc.).

[0078] The historical conversation features may include: some features determined based on each historical conversation of the recommendation object, such as user historical interaction, user participation, etc.

[0079] It can be seen that the above feature information contains features of different dimensions such as recommended objects, videos, requests, and historical conversations, thereby enriching the content of the feature information. Moreover, in practical applications, in addition to feature information such as object features, video features, cross features, request features, and historical conversation features, positive and negative samples can also further include some other features as needed, which is very flexible and convenient. For example, it can further include features used in existing methods to estimate feedback targets such as satisfaction, completion of broadcast, and fast scrolling.

[0080] The generated positive samples and negative samples can be used to form a training sample set, and the training sample set can be used to train the fatigue prediction model. The fatigue prediction model can adopt an extreme gradient boosting tree (XGBoost, eXtremeGradient Boosting) structure.

[0081] After the training of the fatigue prediction model is completed, it can be applied to actual fatigue prediction, which is described below through an embodiment.

[0082] Figure 2 Flow chart of an embodiment of the fatigue prediction method disclosed in the present invention. Figure 2 As shown, the following specific implementation methods are included.

[0083] In step 201, candidate resources to be recommended are determined for a target recommendation object of opening a resource recommendation application.

[0084] In step 202, a fatigue prediction model is used to determine fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on historical consumption records corresponding to the resource recommendation application. The historical consumption records include: operation information of the recommended object who opens the resource recommendation application for the recommended resource object.

[0085] By adopting the scheme described in the above method embodiment, when recommending resources to the target recommendation object, the fatigue prediction model can be used to determine the fatigue prediction results of the candidate resources to be recommended. The fatigue prediction results can be used as negative feedback targets, thereby enriching the content of existing negative feedback targets and further improving the accuracy of fusion results.

[0086] In some embodiments of the present disclosure, for any candidate resource, third feature information corresponding to the candidate resource may be obtained, and the third feature information may be input into a fatigue prediction model to obtain an output fatigue prediction result.

[0087] The fatigue prediction result can be 1 or 0. The fatigue prediction model can be based on Figure 1 The fatigue prediction model is obtained by training using the method in the illustrated embodiment.

[0088] In addition, in some embodiments of the present disclosure, the resource recommendation application may include: a short video platform, and accordingly, the resource object type may be a video type. For any candidate video, a method for obtaining the third feature information corresponding to the candidate resource may include: obtaining object features of the target recommendation object, obtaining video features of the candidate video, obtaining cross-features between the candidate video and the target recommendation object, obtaining request features, wherein the request is a request for triggering the recommendation of the candidate video, and obtaining historical session features of the target recommendation object, wherein the session refers to the consumption process of the target recommendation object from opening the short video platform to exiting the short video platform, and determining the object features, video features, cross-features, request features, and historical session features as the third feature information corresponding to the candidate video.

[0089] Combined with the above introduction, Figure 3 Schematic diagram of the overall implementation process of the fatigue prediction model training and fatigue prediction method described in the present disclosure. Figure 3 As shown, for a certain short video platform, historical consumption records can be obtained, wherein, for any historical consumption record, in response to determining that the video corresponding to the historical consumption record meets any of the first negative sample screening condition, the second negative sample screening condition and the third negative sample screening condition, the video corresponding to the historical consumption record can be determined as a negative sample video, and for any historical consumption record, in response to determining that the video corresponding to the historical consumption record meets the positive sample screening condition, the video corresponding to the historical consumption record can be determined as a positive sample video, and then the first feature information corresponding to each negative sample video and the second feature information corresponding to each positive sample video can be obtained respectively by performing feature extraction, and a negative sample can be generated based on the extracted feature information. and positive samples to form a training sample set, which can then be used to train the fatigue prediction model. After the training is completed, for any target recommendation object that opens the short video platform, when it is necessary to recommend a video to it, feature extraction can be performed on each determined candidate resource, and the extracted third feature information and the fatigue prediction model can be used to determine the fatigue prediction result of each candidate resource, and the fatigue prediction result can be used for subsequent related processing, such as multi-target fusion, etc., and finally determine the recommended video and recommend it to the target recommendation object. Furthermore, a new historical consumption record can be generated according to the target recommendation object's operation information on the recommended video, which can be used for subsequent optimization of the fatigue prediction model.

[0090] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the described order of actions, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present disclosure. In addition, for parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0091] The above is an introduction to the method embodiment. The following is a further explanation of the scheme disclosed in the present invention through an apparatus embodiment.

[0092] Figure 4 FIG. 4 is a schematic diagram of the structure of the fatigue prediction model training device embodiment 400 of the present disclosure. Figure 4 As shown, it includes: an information acquisition module 401, a sample construction module 402 and a model training module 403.

[0093] The information acquisition module 401 is used to acquire the historical consumption record corresponding to the predetermined resource recommendation application, wherein the historical consumption record includes: operation information of opening the recommended object of the resource recommendation application for the recommended resource object.

[0094] The sample construction module 402 is used to generate a training sample set according to historical consumption records.

[0095] The model training module 403 is used to train the fatigue prediction model according to the training sample set. The fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resource to be recommended when recommending resources to the target recommendation object to be processed.

[0096] By adopting the scheme described in the above-mentioned device embodiment, the fatigue prediction model can be trained using the training sample set, and then when recommending resources to the target recommendation object, the trained fatigue prediction model can be used to determine the fatigue prediction results of the candidate resources to be recommended. The fatigue prediction results can be used as negative feedback targets, thereby enriching the content of existing negative feedback targets and improving the accuracy of fusion results.

[0097] In some embodiments of the present disclosure, the resource recommendation application may include: a short video platform, and accordingly, the resource object type may be a video type, each historical consumption record corresponds to a recommendation object and a recommended first video, and the sample construction module 402 generates a training sample set according to the historical consumption records. The method may include: in response to determining that there is a second video in the first video that meets any of the first negative sample filtering condition, the second negative sample filtering condition, and the third negative sample filtering condition, determining that the negative sample video includes the second video; in response to determining that there is a third video in the first video that meets the positive sample filtering condition, determining that the positive sample video includes the third video, generating negative samples in the training sample set according to the negative sample videos, and generating positive samples in the training sample set according to the positive sample videos.

[0098] In some embodiments of the present disclosure, the sample construction module 402 can determine any fourth video in the first video, and can determine the first playback time of the fourth video based on the corresponding historical consumption record. In response to determining that the first playback time is less than the first threshold, it can be determined that the fourth video meets the first negative sample screening condition, and the fourth video can be determined as a negative sample video.

[0099] In some embodiments of the present disclosure, for any historical consumption record, the sample construction module 402 can determine any fourth video in the first video, and can obtain a sixth video from the fifth video whose video duration belongs to the same duration interval as the fourth video, the fifth video includes other videos in the first video except the fourth video, and the target playback duration of each video in the video set is determined according to the corresponding historical consumption record. The video set includes the fourth video and the sixth video, and the videos in the video set are sorted in descending order according to the target playback duration. Accordingly, in response to determining that there is a jump-out operation on the fourth video, and determining that the sorting position number of the fourth video is greater than the product of M and x, it can be determined that the fourth video meets the second negative sample screening condition, and the fourth video can be determined as a negative sample video, M represents the number of videos in the video set, x represents the first percentile greater than 0 and less than 1, and the existence of a jump-out operation means that the recommended object corresponding to the historical consumption record exits the short video platform when playing the fourth video.

[0100] In some embodiments of the present disclosure, the sample construction module 402 can also count the number of videos belonging to the same category as the fourth video that have been recommended before the fourth video in the same session, where the session refers to the consumption process of the recommended object corresponding to the historical consumption record from opening the short video platform to exiting the short video platform. Accordingly, in response to determining that the following conditions are met at the same time: the number of videos is greater than the second threshold, the sorting position number of the fourth video is greater than the product of M and y, and the video length of the fourth video is less than the third threshold, it can be determined that the fourth video meets the third negative sample screening condition, and the fourth video can be determined as a negative sample video, and y represents the second percentile greater than 0 and less than 1.

[0101] In addition, in some embodiments of the present disclosure, the sample construction module 402 determines that the fourth video meets the positive sample screening conditions in response to determining that no predetermined negative operation has occurred with respect to the fourth video and determining that the sorting position number of the fourth video is less than the product of M and z, and the fourth video may be determined as a positive sample video, where z represents the third percentile greater than 0 and less than 1.

[0102] Furthermore, in some embodiments of the present disclosure, the sample construction module 402 can generate corresponding negative samples for each negative sample video, and each negative sample can include: first feature information and a first label corresponding to the corresponding negative sample video, and the first label is used to indicate that the corresponding sample is a negative sample. In addition, a corresponding positive sample can be generated for each positive sample video, and each positive sample can include: second feature information and a second label corresponding to the corresponding positive sample video, and the second label is used to indicate that the corresponding sample is a positive sample.

[0103] In some embodiments of the present disclosure, the first feature information and the second feature information may be obtained in the same manner. For example, the sample construction module 402 may perform the following processing: for a target video, the target video is any negative sample video or any positive sample video, the object features of the recommended object corresponding to the target video are obtained, the recommended object corresponding to the target video is the recommended object corresponding to the same historical consumption record as the target video, the video features of the target video are obtained, the cross features of the target video and the recommended object corresponding to the target video are obtained, the request features are obtained, the request is a request to trigger the recommendation of the target video, the historical session features of the recommended object corresponding to the target video are obtained, the session refers to the consumption process of the recommended object corresponding to the target video from opening the short video platform to exiting the short video platform, and the obtained object features, video features, cross features, request features and historical session features are determined as the feature information corresponding to the target video.

[0104] Figure 5 FIG. 5 is a schematic diagram of the structure of the fatigue prediction device embodiment 500 of the present disclosure. Figure 5 As shown, it includes: a resource determination module 501 and a result determination module 502.

[0105] The resource determination module 501 is used to determine candidate resources to be recommended for a target recommendation object of opening a resource recommendation application.

[0106] The result determination module 502 is used to use the fatigue prediction model to determine the fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on the historical consumption records corresponding to the resource recommendation application. The historical consumption records include: the operation information of the recommended object who opens the resource recommendation application for the recommended resource object.

[0107] By adopting the scheme described in the above-mentioned device embodiment, when recommending resources to the target recommendation object, the fatigue prediction model can be used to determine the fatigue prediction results of the candidate resources to be recommended. The fatigue prediction results can be used as negative feedback targets, thereby enriching the content of existing negative feedback targets and further improving the accuracy of fusion results.

[0108] In some embodiments of the present disclosure, the result determination module 502 may respectively obtain the third feature information corresponding to any candidate resource, and may input the third feature information into a fatigue prediction model to obtain an output fatigue prediction result.

[0109] In addition, in some embodiments of the present disclosure, the resource recommendation application may include: a short video platform, and accordingly, the resource object type may be a video type. For any candidate video, the result determination module 502 may obtain the third feature information corresponding to the candidate resource in a manner that may include: obtaining object features of the target recommendation object, obtaining video features of the candidate video, obtaining cross-features between the candidate video and the target recommendation object, obtaining request features, wherein the request is a request for triggering the recommendation of the candidate video, and obtaining historical session features of the target recommendation object, wherein the session refers to the consumption process of the target recommendation object from opening the short video platform to exiting the short video platform, and determining the object features, video features, cross-features, request features, and historical session features as the third feature information corresponding to the candidate video.

[0110] Figure 4 and Figure 5 The specific working process of the illustrated device embodiment can refer to the relevant description in the aforementioned method embodiment and will not be described in detail.

[0111] In summary, the scheme described in the present disclosure can enrich the content of existing negative feedback targets, thereby improving the accuracy of fusion results and recommendation results, etc. Moreover, the scheme described in the present disclosure can be applied to various resource recommendation applications, that is, it has wide applicability.

[0112] The solution disclosed in the present disclosure can be applied to the field of artificial intelligence, especially to the fields of deep learning, large models, natural language processing, and intelligent recommendation. Artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.). It has both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing. Artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0113] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0114] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as 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 examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0115] like Figure 6 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 to a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0116] Multiple components in the electronic device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the electronic device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0117] The computing unit 601 may be a variety of general and / or special 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 dedicated artificial intelligence (AI, Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSP, Digital Signal Processing), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as the methods described in the present disclosure. For example, in some embodiments, the methods described in the present disclosure may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the methods described in the present disclosure may be executed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the method described in the present disclosure in any other appropriate manner (for example, by means of firmware).

[0118] 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 parts (ASSPs), system on chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a dedicated or general programmable processor, which can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0120] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM, Electronically Programmable Read-Only Memory), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM, Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0121] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 input, voice input, or tactile input).

[0122] The systems and techniques described herein can be implemented in a computing system that includes backend 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 frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0123] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0124] 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 recorded 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 this document does not limit this.

[0125] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A fatigue prediction model training method, comprising: Acquire a historical consumption record corresponding to a predetermined resource recommendation application, wherein the historical consumption record includes: operation information of opening a recommended object of the resource recommendation application for the recommended resource object; Generate a training sample set according to the historical consumption records; The fatigue prediction model is trained according to the training sample set, and the fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resource to be recommended when recommending resources to the target recommendation object to be processed.

2. The method according to claim 1, wherein: The resource recommendation application includes: a short video platform; the resource object type is a video type; Each historical consumption record corresponds to a recommended object and a first video recommended; Generating a training sample set according to the historical consumption records includes: In response to determining that there is a second video in the first video that meets any one of the first negative sample screening condition, the second negative sample screening condition, and the third negative sample screening condition, determining that the negative sample video includes the second video; In response to determining that a third video that meets the positive sample screening condition exists in the first video, determining that the positive sample video includes the third video; Generate negative samples in the training sample set according to the negative sample video; A positive sample in the training sample set is generated according to the positive sample video.

3. The method according to claim 2, wherein: Generating a training sample set according to the historical consumption records includes: Determine any fourth video in the first video; Determining a first playback duration of the fourth video according to the corresponding historical consumption record; In response to determining that the first playback duration is less than a first threshold, determining that the fourth video meets the first negative sample screening condition.

4. The method according to claim 2, wherein: Generating a training sample set according to the historical consumption records includes: Determine any fourth video in the first video; Acquire a sixth video from the fifth video, the video duration of which belongs to the same duration interval as the video duration of the fourth video, wherein the fifth video includes other videos in the first video except the fourth video; Determine the target playback duration of each video in the video set according to the corresponding historical consumption records, wherein the video set includes the fourth video and the sixth video; Sorting the videos in the video set in descending order according to the target playback duration; In response to determining that a jump-out operation has occurred for the fourth video, and determining that the sorting position number of the fourth video is greater than the product of M and x, it is determined that the fourth video meets the second negative sample screening condition, M represents the number of videos in the video set, x represents the first percentile greater than 0 and less than 1, and the jump-out operation refers to the recommendation object corresponding to the historical consumption record exiting the short video platform when playing the fourth video.

5. The method according to claim 4, wherein: In response to determining that the following conditions are met at the same time, determining that the fourth video meets the third negative sample screening condition: the number of videos is greater than the second threshold, the ranking position number of the fourth video is greater than the product of M and y, and the video length of the fourth video is less than the third threshold, and y represents a second percentile greater than 0 and less than 1; Among them, the number of videos is the number of videos belonging to the same category as the fourth video that have been recommended before the fourth video is sent in the same session, and the session refers to the consumption process of the recommended object corresponding to the historical consumption record from opening the short video platform to exiting the short video platform.

6. The method according to claim 4, wherein: In response to determining that no predetermined negative operation occurs for the fourth video, and determining that the sorting position number of the fourth video is less than the product of M and z, it is determined that the fourth video meets the positive sample screening condition, and z represents a third percentile greater than 0 and less than 1.

7. The method according to any one of claims 2 to 6, wherein: Generating negative samples in the training sample set according to the negative sample video comprises: For each negative sample video, a corresponding negative sample is generated respectively, wherein the negative sample includes: first feature information corresponding to the corresponding negative sample video and a first label, wherein the first label is used to indicate that the corresponding sample is a negative sample; Generating the positive samples in the training sample set according to the positive sample video comprises: For each positive sample video, a corresponding positive sample is generated respectively, and the positive sample includes: second feature information corresponding to the corresponding positive sample video and a second label, and the second label is used to indicate that the corresponding sample is a positive sample.

8. The method according to claim 7, wherein: The first characteristic information and the second characteristic information are both obtained by: For a target video in any of the negative sample videos or any of the positive sample videos, obtaining object features of a recommended object corresponding to the target video, wherein the recommended object corresponding to the target video is a recommended object corresponding to the same historical consumption record as the target video; Acquire video features of the target video; Obtaining cross-features of the target video and the recommended object corresponding to the target video; Acquire a request feature, where the request is a request for triggering a recommendation of the target video; Acquire historical session features of the recommended object corresponding to the target video, where the session refers to the consumption process of the recommended object corresponding to the target video from opening the short video platform to exiting the short video platform; The object feature, the video feature, the cross feature, the request feature, and the historical session feature are determined as feature information corresponding to the target video.

9. A method for predicting fatigue, comprising: Determine candidate resources to be recommended for a target recommendation object of opening a resource recommendation application; The fatigue prediction model is used to determine the fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on the historical consumption records corresponding to the resource recommendation application. The historical consumption records include: the operation information of the recommended object who opens the resource recommendation application for the recommended resource object.

10. The method according to claim 9, wherein: The step of using the fatigue prediction model to respectively determine the fatigue prediction results of the target recommendation object for each candidate resource includes: For any candidate resource, respectively obtain the third feature information corresponding to the candidate resource; The third characteristic information is input into the fatigue prediction model to obtain the fatigue prediction result as output.

11. The method according to claim 10, wherein: The resource recommendation application includes: a short video platform; the resource object type is a video type; The obtaining of the third feature information corresponding to the candidate video includes: Acquire object features of the target recommended object; Obtaining video features of the candidate video; Obtaining cross-features of the candidate video and the target recommendation object; Obtaining a request feature, where the request is a request for triggering a recommendation of the candidate video; Acquire historical session features of the target recommendation object, where the session refers to the consumption process of the target recommendation object from opening the short video platform to exiting the short video platform; The object feature, the video feature, the cross feature, the request feature, and the historical session feature are determined as the third feature information.

12. A fatigue prediction model training device, comprising: Information acquisition module, sample construction module and model training module; The information acquisition module is used to acquire the historical consumption record corresponding to the predetermined resource recommendation application, wherein the historical consumption record includes: operation information of opening the recommended object of the resource recommendation application for the recommended resource object; The sample construction module is used to generate a training sample set according to the historical consumption records; The model training module is used to train the fatigue prediction model according to the training sample set. The fatigue prediction model is used to determine the fatigue prediction result of the target recommendation object for the candidate resources to be recommended when recommending resources to the target recommendation object to be processed.

13. The device according to claim 12, wherein: The resource recommendation application includes: a short video platform; the resource object type is a video type; Each historical consumption record corresponds to a recommended object and a first video recommended; In response to determining that there is a second video in the first video that meets any one of the first negative sample screening condition, the second negative sample screening condition and the third negative sample screening condition, the sample construction module determines that the negative sample video includes the second video; in response to determining that there is a third video in the first video that meets the positive sample screening condition, the sample construction module determines that the positive sample video includes the third video; negative samples in the training sample set are generated based on the negative sample videos; and positive samples in the training sample set are generated based on the positive sample videos.

14. The device according to claim 13, wherein: The sample construction module determines any fourth video in the first video, determines a first playback duration of the fourth video according to the corresponding historical consumption record, and in response to determining that the first playback duration is less than a first threshold, determines that the fourth video meets the first negative sample screening condition.

15. The device according to claim 13, wherein: The sample construction module determines any fourth video in the first video, obtains a sixth video from the fifth video whose video duration belongs to the same duration interval as the fourth video, the fifth video includes other videos in the first video except the fourth video, and determines the target playback duration of each video in the video set according to the corresponding historical consumption records. The video set includes the fourth video and the sixth video, and sorts the videos in the video set in descending order according to the target playback duration. In response to determining that there is a jump-out operation on the fourth video, and determining that the sorting position number of the fourth video is greater than the product of M and x, it is determined that the fourth video meets the second negative sample screening condition, M represents the number of videos in the video set, x represents the first percentile greater than 0 and less than 1, and the jump-out operation refers to the recommendation object corresponding to the historical consumption record exiting the short video platform when playing the fourth video.

16. The device according to claim 15, wherein: The sample construction module determines that the fourth video meets the third negative sample screening condition in response to determining that the following conditions are met at the same time: the number of videos is greater than the second threshold, the ranking position number of the fourth video is greater than the product of M and y, and the video length of the fourth video is less than the third threshold, and y represents a second percentile greater than 0 and less than 1; Among them, the number of videos is the number of videos belonging to the same category as the fourth video that have been recommended before the fourth video is sent in the same session, and the session refers to the consumption process of the recommended object corresponding to the historical consumption record from opening the short video platform to exiting the short video platform.

17. The device according to claim 15, wherein: The sample construction module determines that the fourth video meets the positive sample screening condition in response to determining that no predetermined negative operation has occurred for the fourth video and determining that the sorting position number of the fourth video is less than the product of M and z, where z represents the third percentile greater than 0 and less than 1.

18. The device according to any one of claims 13 to 17, wherein: The sample construction module generates a corresponding negative sample for each negative sample video, and the negative sample includes: first feature information and a first label corresponding to the corresponding negative sample video, and the first label is used to indicate that the corresponding sample is a negative sample; and generates a corresponding positive sample for each positive sample video, and the positive sample includes: second feature information and a second label corresponding to the corresponding positive sample video, and the second label is used to indicate that the corresponding sample is a positive sample.

19. The device according to claim 18, wherein: The sample construction module obtains the first feature information and the second feature information in the same manner, including: for a target video in any of the negative sample videos or any of the positive sample videos, obtaining the object features of the recommended object corresponding to the target video, the recommended object corresponding to the target video being the recommended object corresponding to the same historical consumption record as the target video; obtaining the video features of the target video; obtaining the cross-features of the target video and the recommended object corresponding to the target video; obtaining the request features, the request being a request that triggers the recommendation of the target video; obtaining the historical session features of the recommended object corresponding to the target video, the session referring to the consumption process of the recommended object corresponding to the target video from opening the short video platform to exiting the short video platform; determining the object features, the video features, the cross-features, the request features and the historical session features as the feature information corresponding to the target video.

20. A fatigue prediction device, comprising: a resource determination module and a result determination module; The resource determination module is used to determine candidate resources to be recommended for a target recommendation object of opening a resource recommendation application; The result determination module is used to use a fatigue prediction model to determine the fatigue prediction results of the target recommendation object for each candidate resource. The fatigue prediction model is trained based on a training sample set, and the training sample set is generated based on historical consumption records corresponding to the resource recommendation application. The historical consumption records include: operation information of the recommended object of the resource recommendation application for the recommended resource object.

21. The device according to claim 20, wherein: The result determination module obtains the third feature information corresponding to any candidate resource, respectively, inputs the third feature information into the fatigue prediction model, and obtains the output fatigue prediction result.

22. The device according to claim 21, wherein The resource recommendation application includes: a short video platform; the resource object type is a video type; The result determination module obtains the object features of the target recommendation object, obtains the video features of the candidate video, obtains the cross features of the candidate video and the target recommendation object, obtains the request features, the request is a request for triggering the recommendation of the candidate video, obtains the historical session features of the target recommendation object, the session refers to the consumption process of the target recommendation object from opening the short video platform to exiting the short video platform, and determines the object features, the video features, the cross features, the request features and the historical session features as the third feature information.

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, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 11.

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

25. A computer program product, comprising a computer program / instruction, which implements the method according to any one of claims 1 to 11 when executed by a processor.

Images