Resource classification model training method, resource recommendation method and device

Abstract

The application relates to the technical field of machine learning, in particular to a resource classification model training method and a resource recommendation method and device, which can be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation and auxiliary driving; the method comprises the following steps: acquiring preset multimedia resource screening conditions; the preset multimedia resource screening conditions comprise a resource interaction condition and a resource quality grade condition; seed multimedia resources are screened out from an original multimedia resource pool based on the resource interaction condition and the resource quality grade condition; preset multimedia resources are extracted from the original multimedia resource pool; positive samples are generated based on the seed multimedia resources, and negative samples are generated based on the preset multimedia resources; a preset classification model is trained based on the positive samples and the negative samples, so that a resource classification model is obtained. The application can improve the quality of multimedia resources to be recommended and improve the retention rate of a viewing object on a multimedia resource platform.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, and particularly relates to a resource classification model training method and a resource recommendation method and device. BACKGROUND

[0002] In a multimedia resource recommendation scenario, a plurality of multimedia resources are recalled from a large-dish resource recommendation pool, the recalled multimedia resources are sent into a ranking model, and finally the multimedia resources to be recommended are obtained.

[0003] In the prior art, the quality of the multimedia resources in the large-dish resource recommendation pool is uneven, which may lead to uncontrollable quality of the multimedia resources to be recommended, and further affect the viewing experience of the viewing object and reduce the retention rate of the viewing object on the multimedia resource platform. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a resource classification model training method, a resource recommendation method and device, which can improve the quality of the multimedia resources to be recommended and improve the retention rate of the viewing object on the multimedia resource platform.

[0005] To solve the above technical problem, on the one hand, the present application provides a resource classification model training method, characterized in that it comprises:

[0006] obtaining a preset multimedia resource screening condition; the preset multimedia resource screening condition comprises a resource interaction condition and a resource quality level condition;

[0007] screening seed multimedia resources from an original multimedia resource pool based on the resource interaction condition and the resource quality level condition; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition;

[0008] randomly extracting a preset multimedia resource from the original multimedia resource pool;

[0009] generating positive samples based on the seed multimedia resources and negative samples based on the preset multimedia resources;

[0010] training a preset classification model based on the positive samples and the negative samples to obtain a resource classification model.

[0011] On the other hand, the present application provides a resource recommendation method, comprising:

[0012] extracting resource features of candidate resources;

[0013] inputting the resource feature of the candidate resource into a resource classification model for resource classification to obtain a resource classification result of the candidate resource; the resource classification model is obtained by training based on positive samples and negative samples; the positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool; the seed multimedia resources are obtained by screening from the original multimedia resource pool based on preset multimedia resource screening conditions, the preset multimedia resource screening conditions include a resource interaction condition and a resource quality level condition; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening conditions;

[0014] When the resource classification result indicates that the category of the candidate resource is a to-be-recommended category, the candidate resource is determined as a to-be-recommended resource.

[0015] In another aspect, an embodiment of the present application provides a resource classification model training device, comprising:

[0016] a screening condition acquisition module configured to acquire preset multimedia resource screening conditions; the preset multimedia resource screening conditions include a resource interaction condition and a resource quality level condition;

[0017] a resource screening module configured to screen seed multimedia resources from an original multimedia resource pool based on the resource interaction condition and the resource quality level condition; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening conditions;

[0018] a resource extraction module configured to randomly extract preset multimedia resources from the original multimedia resource pool;

[0019] a sample generation module configured to generate positive samples based on the seed multimedia resources and negative samples based on the preset multimedia resources;

[0020] a classification model training module configured to train a preset classification model based on the positive samples and the negative samples to obtain a resource classification model.

[0021] In another aspect, an embodiment of the present application provides a resource recommendation device, comprising:

[0022] a resource feature extraction module configured to extract resource features of a candidate resource;

[0023] The resource classification module is configured to input the resource feature of the candidate resource into a resource classification model to perform resource classification, and obtain a resource classification result of the candidate resource. The resource classification model is trained based on positive samples and negative samples. The positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool. The seed multimedia resources are obtained by screening the original multimedia resource pool based on preset multimedia resource screening conditions. The preset multimedia resource screening conditions include a resource interaction condition and a resource quality level condition. The seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score. The preset quality score is associated with the preset multimedia resource screening conditions.

[0024] The resource recommendation module is configured to determine the candidate resource as a resource to be recommended when the resource classification result indicates that the category of the candidate resource is a category to be recommended.

[0025] In another aspect, the present application provides an electronic device, which includes a processor and a memory. The memory stores at least one instruction or at least one program. The at least one instruction or the at least one program is loaded and executed by the processor to implement the resource classification model training method or the resource recommendation method as described above.

[0026] In another aspect, the present application provides a computer storage medium. The storage medium stores at least one instruction or at least one program. The at least one instruction or the at least one program is loaded and executed by a processor to implement the resource classification model training method or the resource recommendation method as described above.

[0027] The implementation of the embodiments of the present application has the following beneficial effects:

[0028] The application screens seed multimedia resources from an original multimedia resource pool through preset multimedia resource screening conditions; the resource quality score of the screened multimedia resources is greater than or equal to a preset quality score, thereby improving the accuracy of seed multimedia resource determination; a preset multimedia resource is extracted from the original multimedia resource pool; a positive sample is generated based on the seed multimedia resource, a negative sample is generated based on the preset multimedia resource, and then model training is performed based on the positive sample and the negative sample to obtain a corresponding resource classification model. Since the screened seed multimedia resource can represent high-quality multimedia resources, using the seed multimedia resource as a positive sample can enable the classification model to fully learn the resource characteristics of high-quality multimedia resources, thereby improving the accuracy of classification of the classification model. Further, the multimedia resources in the original resource pool are classified based on the classification model, and the multimedia resources classified as a to-be-recommended category are recommended to a viewing object based on the classification result, which can improve the quality of the to-be-recommended multimedia resources and improve the retention rate of the viewing object on the multimedia resource platform. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art and the advantages thereof, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0030] Figure 1 is an implementation environment schematic diagram provided by the embodiments of the present application;

[0031] Figure 2 is a resource classification model training method flowchart provided by the embodiments of the present application;

[0032] Figure 3 is a resource screening method flowchart provided by the embodiments of the present application;

[0033] Figure 4 is a first target resource screening method flowchart provided by the embodiments of the present application;

[0034] Figure 5 is a generation method flowchart of an interactive operation chain provided by the embodiments of the present application;

[0035] Figure 6 is a sample generation method flowchart provided by the embodiments of the present application;

[0036] Figure 7 is a resource recommendation method flowchart provided by the embodiments of the present application;

[0037] Figure 8is a multimedia resource classification process schematic diagram provided by an embodiment of the present application.

[0038] Figure 9 is a resource recommendation schematic diagram provided by an embodiment of the present application.

[0039] Figure 10 is a resource classification model training device schematic diagram provided by an embodiment of the present application.

[0040] Figure 11 is a resource recommendation device schematic diagram provided by an embodiment of the present application.

[0041] Figure 12 is an electronic device structure schematic diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0042] To make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0043] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0044] The embodiments of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, intelligent transportation, and assisted driving.

[0045] Please refer to Figure 1 which shows an implementation environment schematic diagram provided by an embodiment of the present application, which can include at least one client 110 and a resource server 120, and the client 110 and the resource server 120 can communicate data through a network.

[0046] Specifically, the client 110 can send a resource display request to the resource server 120; the resource server 120 selects a corresponding target recommended resource from the to-be-recommended resource pool and sends it to the client 110 upon receiving the resource display request. The resource server 120 can pre-classify each resource in the original resource pool and put the resources classified as the to-be-recommended category into the to-be-recommended resource pool, so as to facilitate resource recommendation based on the to-be-recommended resource pool.

[0047] The client 110 can communicate with the resource server 120 based on a browser / server mode (B / S) or a client / server mode (C / S). The client 110 can include an entity device such as a smart phone, a tablet computer, a notebook computer, a digital assistant, a smart wearable device, a vehicle terminal, etc., or a software such as an application program running on the entity device. The operating system running on the client 110 in the embodiment of the present application can include but is not limited to an Android system, an IOS system, Linux, Windows, etc.

[0048] The resource server 120 and the client 110 can establish a communication connection through a wire or wireless connection. The resource server 120 can include a standalone server, a distributed server, or a server cluster composed of multiple servers, wherein the server can be a cloud server.

[0049] To solve the problem that the quality of the to-be-recommended resources cannot be controlled in the prior art, thereby affecting the retention rate of the viewing object on the multimedia resource platform, the original multimedia resources in the original resource pool can be classified in the embodiment of the present application, and the resources are recommended based on the resource classification result. Specifically, the resource classification can be achieved through a resource classification model, please refer to Figure 2 which shows a resource classification model training method, the execution subject of which can be the resource server described above. The method can include:

[0050] S210. Obtain a preset multimedia resource screening condition; the preset multimedia resource screening condition includes a resource interaction condition and a resource quality level condition.

[0051] In the embodiment, the multimedia resource can be video, news, information, etc. The preset multimedia resource screening condition can be used to limit the conditions of the screened resources, and the preset multimedia resource screening condition can be flexibly set based on the actual application scenario.

[0052] Specifically, due to the unique properties of multimedia resources, they can have multiple dimensions of resource characteristics, so when determining the preset multimedia resource screening condition, it can be determined based on multiple dimensions of resource characteristics, that is, the preset multimedia resource screening condition corresponds to multiple dimensions of resource characteristics. Each multimedia resource can have a resource interaction feature, and a corresponding resource interaction condition can be set. The resource interaction condition can be used to screen multimedia resources whose resource interaction feature meets the preset interaction condition. For example, the resource interaction feature can include the cumulative number of plays of the multimedia resource, and the corresponding determined resource interaction condition can include that the cumulative number of plays of the multimedia resource is greater than or equal to a preset play number.

[0053] Each multimedia resource can also have a resource quality level feature, and a corresponding resource quality level condition can be set. The resource quality level condition can be used to screen multimedia resources whose resource quality level feature meets the preset quality level condition. For example, the resource quality level feature can include the quality level of the multimedia resource, and the corresponding determined resource quality level condition can include that the quality level of the multimedia resource is equal to or higher than a preset quality level. In one specific example, the resource quality level condition can be: multimedia resource publisher object level >= 4 && fan quantity > 10w, multimedia resource interaction object level >= 4 && interaction in the last 30 days.

[0054] The resource characteristics of the multimedia resources in this embodiment include but are not limited to resource interaction features and resource quality level features, and accordingly, the preset multimedia resource screening conditions include but are not limited to resource interaction conditions and resource quality level conditions.

[0055] S220. Based on the resource interaction condition and the resource quality level condition, seed multimedia resources are screened from the original multimedia resource pool; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition.

[0056] The resource interaction condition and the resource quality level condition can both be used to screen multimedia resources that meet the corresponding quality requirements. Specifically, the resource quality score of the multimedia resources obtained after screening can be greater than or equal to a preset quality score, and the screened multimedia resources can be determined as seed multimedia resources. Seed multimedia resources can to some extent represent high-quality multimedia resources, so that other high-quality multimedia resources can be mined through seed multimedia resources.

[0057] Further, the preset multimedia resource screening condition can also match the number of target seed multimedia resources; for example, the resource interaction feature can include the cumulative play count of the multimedia resource, and the corresponding determined resource interaction condition can include that the cumulative play count of the multimedia resource is greater than or equal to a preset play count. When the number of target seed multimedia resources is n1, the preset play count can be determined as m1; when the number of target seed multimedia resources is n2, the preset play count can be determined as m2, where n1 < n2 and m1 > m2. That is, the greater the number of required seed multimedia resources, the lower the corresponding preset multimedia resource screening condition, and the smaller the number of required seed multimedia resources, the higher the corresponding preset multimedia resource screening condition.

[0058] S230. Randomly extracting a preset multimedia resource from the original multimedia resource pool.

[0059] In the embodiment of the application, in order to reflect the randomness and representativeness of the extraction of the multimedia resource, the multimedia resource can be randomly extracted from the original multimedia resource pool, and the randomly extracted multimedia resource can be determined as the preset multimedia resource. Specifically, when the original multimedia resource pool is randomly extracted, a preset random algorithm can be used to achieve this, and any random algorithm that can achieve random extraction can be applied to the embodiment of the application, and the embodiment is not limited in this regard.

[0060] Further, the number of preset multimedia resources can be greater than the number of seed multimedia resources. Because generally, the workload required to obtain seed multimedia resources by screening the original multimedia resource pool is greater than the workload required to randomly select preset multimedia resources from the original multimedia resource pool, thereby saving data processing resources. On the other hand, by selecting a large number of preset multimedia resources, the training sample can be expanded, thereby improving the training effect of the model.

[0061] S240. Generating positive samples based on the seed multimedia resources and generating negative samples based on the preset multimedia resources.

[0062] The seed multimedia resource screened in the embodiment can be a multimedia resource that meets the target quality, and accordingly, the positive sample can be determined based on the seed multimedia resource. When generating the positive sample, the resource feature of the seed multimedia resource can be extracted, and the resource label of the seed multimedia resource can be determined, and then the positive sample corresponding to the seed multimedia resource can be generated based on the resource feature and the resource label. Similarly, when generating the negative sample, the resource feature of the preset multimedia resource can be extracted, and the resource label of the preset multimedia resource can be determined, and then the negative sample corresponding to the preset multimedia resource can be generated based on the resource feature and the resource label.

[0063] Specifically, for a seed multimedia resource, the corresponding resource label can be 1, or true; for a preset multimedia resource, the corresponding resource label can be 0, or false.

[0064] S250. Training a preset classification model based on the positive samples and the negative samples to obtain a resource classification model.

[0065] The resource classification model can be used for resource classification of a candidate resource, and the specific classification result can include at least two categories; thereby, based on the resource classification result of the candidate resource, a resource operation on the candidate resource can be determined.

[0066] The present application screens seed multimedia resources from an original multimedia resource pool through preset multimedia resource screening conditions; wherein the resource quality score of the screened multimedia resources is greater than or equal to a preset quality score, thereby improving the accuracy of seed multimedia resource determination; a preset multimedia resource is extracted from the original multimedia resource pool; a positive sample is generated based on the seed multimedia resource, a negative sample is generated based on the preset multimedia resource, and then a model is trained based on the positive sample and the negative sample to obtain a corresponding resource classification model. Since the screened seed multimedia resources can represent high-quality multimedia resources, using the seed multimedia resources as positive samples can enable the classification model to fully learn the resource characteristics of high-quality multimedia resources, thereby improving the accuracy of the classification model classification. Further, based on the classification model, the multimedia resources in the original resource pool are classified, and based on the classification result, the multimedia resources classified as a to-be-recommended category are recommended to the viewing object, which can improve the quality of the to-be-recommended multimedia resources and improve the retention rate of the viewing object on the multimedia resource platform.

[0067] When performing resource screening based on preset multimedia resource screening conditions, resource screening can be performed based on multiple screening conditions respectively; please refer to Figure 3 which shows a resource screening method, which can include:

[0068] S310. Screening a first target resource that meets the resource interaction condition from the original multimedia resource pool.

[0069] S320. Screening a second target resource that meets the resource quality level condition from the original multimedia resource pool.

[0070] S330. Resource fusion is performed on the first target resource and the second target resource to obtain the seed multimedia resource.

[0071] The resource interaction condition can include a resource cumulative play count condition, a resource cumulative like count condition, a resource cumulative share count condition, and the like. For example, the resource interaction condition can be: (cumulative play count of the multimedia resource >= 5000) && ((cumulative like count >= 1000 and like count top 200 under the category) or (cumulative comment count >= 500 and comment count top 200 under the category) or (cumulative share count >= 500 and share count top 200 under the category)).

[0072] The resource quality level condition can include a level condition of a multimedia resource publishing object, a fan number condition of the multimedia resource publishing object, a resource interaction condition of the multimedia resource publishing object, and the like.

[0073] In the field of multimedia resource publishing, the multimedia resources published by the authenticated resource publishing object are likely to be of high quality, so that the corresponding seed multimedia resources can be selected according to the object level of the resource publishing object. For example, the resource quality level condition can be: multimedia resource publishing object level >= 4 && fan number > 10w, and the corresponding parameters can be determined according to actual conditions.

[0074] Further, for a resource publishing object with a high level, not only the multimedia resources published by the resource publishing object are of high quality, but also the multimedia resources considered by the resource publishing object to be of high quality are likely to be of high quality, so that the multimedia resources preferred by the resource publishing object with a high level can be selected as seed multimedia resources. Accordingly, the resource quality level condition can be: multimedia resource interaction object level >= 4 && interaction in the last 30 days, and the corresponding parameters can be determined according to actual conditions.

[0075] Thus, based on the above conditions, the first target resource meeting the resource interaction condition and the second target resource meeting the resource quality level condition can be selected respectively. The first target resource and the second target resource can be fused to obtain corresponding seed multimedia resources.

[0076] The first target resource meeting the resource interaction condition and the second target resource meeting the resource quality level condition can be selected respectively, which can improve the convenience of resource selection. Further, since the first target resource and the second target resource are selected by different conditions respectively, the maximum range of selected resources is ensured, which can improve the number of selected resources. Subsequently, different operations can be performed on the selected resources based on different application scenarios to obtain corresponding seed multimedia resources, thereby improving the operability of the resources.

[0077] Specifically, when the first target resource and the second target resource are fused, a set operation can be performed on the first target resource and the second target resource, that is, the repeated resources in the first target resource and the second target resource are extracted, and one of the same resources is retained, thereby obtaining the corresponding seed multimedia resource. By extracting the repeated multimedia resource, subsequent repeated processing operations on the same multimedia resource can be avoided, and resource processing efficiency can be improved and processing resources can be saved.

[0078] In one example, when the first target resource and the second target resource are fused, an intersection operation can be performed on the first target resource and the second target resource, that is, resources that meet both the resource interaction condition and the resource quality level condition are screened out as seed multimedia resources.

[0079] Further, in the process of fusing the first target resource and the second target resource, the content recognition model can be used to identify resources containing negative content, resources affecting the viewing experience, and resources with time limitation from the first target resource and the second target resource. The resource containing negative content can be a multimedia resource with a negative comment rate greater than a preset value, a multimedia resource containing negative information, etc. The resource affecting the viewing experience can be a resource with low originality, a resource with unclear content, a resource with links, a resource suspected of being an advertisement, etc. The resource with time limitation can be a news or information resource, etc. The identified resources are excluded from the first target resource and the second target resource, and the seed multimedia resource can be obtained.

[0080] The resource interaction condition can specifically include a resource cumulative play count condition, a resource cumulative like count condition, a resource cumulative share count condition, etc., that is, separate counting of various different interaction operations; in order to reflect the continuous interaction information of the viewing object on the multimedia resource, the continuous interaction operation of the viewing object can be recorded; specifically, the resource interaction condition can include an interaction operation chain, and the interaction operation chain is used to represent the continuous interaction information between the target object and the resource to be interacted; correspondingly, please refer to Figure 4 which shows a first target resource screening method, which can include:

[0081] S410. Determine the resource with the interaction operation chain in the original multimedia resource pool as the third target resource.

[0082] S420. Determine the first target resource based on the third target resource.

[0083] The type of the interactive operation chain can be one or more, the interactive operations included in different types of interactive operation chains are different, and the interactive operations included in the same type of interactive operation chain are the same. For the same multimedia resource, because multiple viewing objects can interact with the multimedia resource, the number of interactive operation chains corresponding to the multimedia resource can be multiple.

[0084] The interactive operation chain can be used to represent that the viewing object has performed a preset interactive operation on the multimedia resource, the preset interactive operation is associated with the quality of the multimedia resource, and the multimedia resource with the interactive operation chain can be determined as a multimedia resource with high quality, and then can be determined as the third target resource.

[0085] Further, the resource screening can also be performed based on the number of interactive operation chains corresponding to each multimedia resource. The number of interactive operation chains indicates that the corresponding number of interactions with the multimedia resource is also more; specifically, the resource with a number of interactive operation chains greater than a preset number can be determined as the third target resource.

[0086] The third target resource is a screening resource corresponding to the interactive operation chain condition. Further, the resource interaction condition can also include a resource cumulative play count condition, a resource cumulative like count condition, a resource cumulative share count condition, and the like, so that the screening resource corresponding to the resource cumulative play count condition can be screened out, or the screening resource corresponding to the resource cumulative like count condition can be screened out, or the screening resource corresponding to the resource cumulative share count condition can be screened out, and the like; so that when the first target resource is determined based on the third target resource, one or more resources can be determined from the resources corresponding to the resource cumulative play count condition, the resources corresponding to the resource cumulative like count condition, the resources corresponding to the resource cumulative share count condition, and the like, and the first target resource is generated together with the third target resource.

[0087] The interactive operation chain can represent continuous interaction information, so that the interactive operation chain can improve the accuracy of expressing the interaction feature, and further improve the accuracy of multimedia resource screening.

[0088] For the generation method of the interactive operation chain, please refer to Figure 5 , which can include:

[0089] S510. When it is detected that the target object is in a non-logged-in state and a first interaction operation on the to-be-interacted resource is triggered, jump to a target login page.

[0090] S520. When the target object successfully logs in based on the target login page, obtain a second interaction operation of the target object on the to-be-interacted resource.

[0091] S530. Based on the first interactive operation in the unlogged-in state, the successful login operation based on the target login page, and the second interactive operation in the logged-in state, generate an interactive operation chain corresponding to the resource to be interacted with.

[0092] The resource server can detect the operation data of the target object through the data points embedded on the client side. When it detects that the target object has triggered the first interactive operation of the resource to be interacted with while not logged in, it will redirect the target object to the login page to facilitate the login operation. If the target object does not log in, the creation of the interactive operation chain fails. If the target object successfully logs in and triggers the second interactive operation of the resource to be interacted with after successful login, an interactive operation chain corresponding to the resource to be interacted with can be generated based on the first interactive operation in the unlogged state, the successful login operation based on the target login page, and the second interactive operation in the logged-in state. The first interactive operation and the second interactive operation can be the same or different.

[0093] In this embodiment, the interactive operation chain can include the node of successfully logging in based on the target login page. That is, the target object has logged in in order to interact with the resource to be interacted with, and is willing to consume login costs to achieve the interaction. This indicates that the target object has a high degree of preference for the resource to be interacted with, which also shows that the quality of the resource to be interacted with is high.

[0094] In specific implementation scenarios, since the target audience cannot perform interactive operations such as liking, commenting, sharing, or following without being logged in, the system will automatically redirect them to the login page and prompt them to log in before they can interact. If an unlogged target audience is redirected to the login page and logs in to like, comment, share, or follow a resource to be interacted with, it indicates that the multimedia resource is of high enough quality that the target audience is willing to pay the login cost to interact with it. Such resources are selected as the third target resource. The specific method is as follows: based on the tracking data, filter multimedia resources where the target audience, without being logged in, clicked "like," "comment," "share," or "follow" on the playback page of a resource to be interacted with, and was then redirected to the login page and completed the login and interaction.

[0095] The first interaction operation, login operation, and second interaction operation generated by the object can concretize the interaction features, reflect the coherence of the interaction operations, and further improve the accuracy of the interaction operation features. By representing the interaction features through the interaction operation chain, the feature representation capability is improved.

[0096] Further, before the preset multimedia resource screening condition is acquired, the method of the embodiment further includes determining the preset multimedia resource screening condition matching the preset quality score; the higher the preset quality score, the more stringent the preset multimedia resource screening condition set accordingly.

[0097] The preset quality score can be a specific score value or a quality score level; taking the quality score level as an example, the preset multimedia resource screening condition corresponding to each quality score level can be pre-configured, so that in a specific resource screening scenario, the corresponding preset multimedia resource screening condition can be determined according to the quality score level of the required screening resource, and then the resource screening is performed based on the determined preset multimedia resource screening condition. Thus, the screening condition can be adaptively adjusted based on the preset quality score, thereby improving the flexibility of screening condition determination.

[0098] After the corresponding resource is screened, the corresponding sample can be created; for details, please refer to Figure 6 which shows a sample generation method, which can include:

[0099] S610. Constructing resource features; the resource features include resource attribute features, resource interaction features, and resource quality level features.

[0100] S620. Sample construction based on the resource attribute features, resource interaction features, and resource quality level features of the seed multimedia resource, to obtain the positive sample.

[0101] S630. Sample construction based on the resource attribute features, resource interaction features, and resource quality level features of the preset multimedia resource, to obtain the negative sample.

[0102] Based on the unique attributes of multimedia resources, resource features of multiple dimensions representing the features of multimedia resources can be created accordingly, which can specifically include resource attribute features, resource interaction features, and resource quality level features.

[0103] The resource attribute features can include: a first classification, a second classification, a quality level, a multimedia resource viewing time length, a multimedia resource storage size, a definition, whether the cover is compliant, whether it is a multimedia resource with goods, whether it is original, whether it has a watermark, and the like; the resource interaction features can include: a negative comment rate, a cumulative play count, a cumulative like count, a cumulative comment count, a cumulative share count, a cumulative attention count, a play count in the last 7 / 14 / 30 days, a like count in the last 7 / 14 / 30 days, a comment count in the last 7 / 14 / 30 days, a share count in the last 7 / 14 / 30 days, an attention count in the last 7 / 14 / 30 days, and the like; the resource quality level features can include: a level of a resource publishing object, a publishing amount of the resource publishing object in the last 7 / 14 / 30 days, a played amount of a published resource of the resource publishing object in the last 7 / 14 / 30 days, a liked amount of the published resource of the resource publishing object in the last 7 / 14 / 30 days, a commented amount of the published resource of the resource publishing object in the last 7 / 14 / 30 days, a shared amount of the published resource of the resource publishing object in the last 7 / 14 / 30 days, an attention amount of the resource publishing object in the last 7 / 14 / 30 days, a fan amount of the resource publishing object, and whether the resource publishing object is an authenticated object.

[0104] Thus, for the seed multimedia resource and the preset multimedia resource, the resource features of the seed multimedia resource and the resource features of the preset multimedia resource can be extracted based on the constructed resource features.

[0105] Further, when the resource features of multiple dimensions are extracted, the multi-dimensional resource features can be preprocessed, such as discrete character variable integerization and discrete integer variable encoding, so as to improve the feature data processing efficiency and simplify the data processing complexity.

[0106] The above embodiments construct the resource features corresponding to the resources from multiple dimensions, which can realize comprehensive expression of the resource features, and further improve the usability and rationality of the sample data.

[0107] In the resource recommendation scenario, the resource to be recommended can be determined based on the resource classification result; for details, please refer to Figure 7 which shows a resource recommendation method, which can include:

[0108] S710. Extract resource features of the candidate resource.

[0109] S720. inputting the resource feature of the candidate resource into a resource classification model for resource classification to obtain a resource classification result of the candidate resource; the resource classification model is obtained based on positive samples and negative samples; the positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool; the seed multimedia resources are obtained by screening from the original multimedia resource pool based on preset multimedia resource screening conditions, and the preset multimedia resource screening conditions include a resource interaction condition and a resource quality level condition; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition.

[0110] S730. determining the candidate resource as a resource to be recommended when the resource classification result indicates that the category of the candidate resource is a to-be-recommended category.

[0111] According to the resource feature constructed in the above embodiment, the resource feature of the candidate resource can be extracted to obtain the resource feature of the candidate resource. The resource classification model can classify the candidate resource based on the resource feature of the candidate resource to obtain a corresponding resource classification result; the resource classification result can be specifically a prediction value of the candidate resource as a high-quality resource, which can be used to represent the similarity between the candidate resource and the seed multimedia resource, and the prediction value is a value between 0 and 1.

[0112] In one example, a preset score value can be set, and when the prediction value corresponding to the candidate resource is greater than or equal to the preset value, the candidate resource is determined as a high-quality resource, and the category of the candidate resource is correspondingly determined as a to-be-recommended category; when the prediction value corresponding to the candidate resource is less than the preset value, the candidate resource is determined as a low-quality resource, and the category of the candidate resource is correspondingly determined as a non-recommended category.

[0113] In another example, the resource classification model can be used to classify all candidate resources in the original resource pool to obtain prediction values corresponding to each candidate resource, and then the candidate resources are sorted in descending order of the prediction values, and the number of resources to be recommended is determined according to the number of resources to be recommended; for example, the number of resources to be recommended is N, then the top N candidate resources in the sorted list are determined as high-quality resources, and the category of the candidate resource is correspondingly determined as a to-be-recommended category, and the other candidate resources in the sorted list are determined as low-quality resources, and the category of the candidate resource is correspondingly determined as a non-recommended category.

[0114] The application screens seed multimedia resources from the original multimedia resource pool by presetting multimedia resource screening conditions; the resource quality score of the screened multimedia resources is greater than or equal to a preset quality score, thereby improving the accuracy of seed multimedia resource determination; the preset multimedia resources are extracted from the original multimedia resource pool; the positive samples are generated based on the seed multimedia resources, the negative samples are generated based on the preset multimedia resources, and then the model training is performed based on the positive samples and the negative samples to obtain the corresponding resource classification model. Since the screened seed multimedia resources can represent high-quality multimedia resources, using the seed multimedia resources as positive samples can enable the classification model to fully learn the resource characteristics of high-quality multimedia resources, thereby improving the accuracy of the classification model classification. Further, the multimedia resources in the original resource pool are classified based on the classification model, and the multimedia resources classified as the to-be-recommended category are recommended to the viewing object based on the classification result, which can improve the quality of the to-be-recommended multimedia resources and improve the retention rate of the viewing object on the multimedia resource platform.

[0115] Please refer to Figure 8 which shows a multimedia resource classification flowchart; wherein, based on the resource interaction condition and the resource quality level condition, resources corresponding to the respective conditions are screened from the original multimedia resource pool, then the resources screened based on the resource interaction condition are fused with the resources screened based on the resource quality level condition, and on the basis of resource fusion, low-quality resources are removed, thereby obtaining seed multimedia resources, which are used as positive samples; resources are randomly extracted from the original multimedia resource pool, and the randomly extracted resources are used as negative samples; model training is performed based on the positive samples and the negative samples to obtain a resource classification model. Then, resource classification is performed based on the resource classification model to obtain a to-be-recommended resource pool.

[0116] Further, please refer to Figure 9 After obtaining the to-be-recommended resource pool, the to-be-recommended resource pool can be used as the input of the recommendation system, and after passing through the recall model, the rough sorting model, the fine sorting model, and the mixed sorting model, the output to-be-recommended resources are obtained.

[0117] Please refer to Figure 10 The embodiment also provides a resource classification model training device, which comprises:

[0118] The screening condition acquisition module 1010 is configured to acquire a preset multimedia resource screening condition; the preset multimedia resource screening condition comprises a resource interaction condition and a resource quality level condition;

[0119] The resource screening module 1020 is configured to screen seed multimedia resources from the original multimedia resource pool based on the resource interaction condition and the resource quality level condition; the seed multimedia resource is a resource with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition.

[0120] The resource extraction module 1030 is configured to randomly extract a preset multimedia resource from the original multimedia resource pool.

[0121] The sample generation module 1040 is configured to generate a positive sample based on the seed multimedia resource and generate a negative sample based on the preset multimedia resource.

[0122] The classification model training module 1050 is configured to train a preset classification model based on the positive sample and the negative sample to obtain a resource classification model.

[0123] Further, the resource screening module 1020 includes:

[0124] The first screening module is configured to screen a first target resource meeting the resource interaction condition from the original multimedia resource pool.

[0125] The second screening module is configured to screen a second target resource meeting the resource quality level condition from the original multimedia resource pool.

[0126] The resource fusion module is configured to perform resource fusion on the first target resource and the second target resource to obtain the seed multimedia resource.

[0127] Further, the resource interaction condition includes an interaction operation chain; the interaction operation chain is used to represent continuous interaction information between a target object and a resource to be interacted.

[0128] The first screening module includes:

[0129] The third screening module is configured to determine a resource with the interaction operation chain in the original multimedia resource pool as a third target resource.

[0130] The first target resource determination module is configured to determine the first target resource based on the third target resource.

[0131] Further, the apparatus further includes:

[0132] The page jump module is configured to jump to a target login page when it is detected that the target object triggers a first interaction operation on the resource to be interacted in a non-logged-in state.

[0133] An interactive operation acquisition module is configured to acquire a second interactive operation of the target object on the to-be-interacted resource when the target object successfully logs in based on the target login page;

[0134] An interactive operation chain generation module is configured to generate an interactive operation chain corresponding to the to-be-interacted resource based on the first interactive operation in the non-logged-in state, the successful login operation based on the target login page, and the second interactive operation in the logged-in state.

[0135] Further, the apparatus further includes:

[0136] A screening condition determination module is configured to determine the preset multimedia resource screening condition matched with the preset quality score.

[0137] Further, the sample generation module 1040 includes:

[0138] A resource feature construction module is configured to construct resource features; the resource features include resource attribute features, resource interactive features, and resource quality level features;

[0139] A positive sample construction module is configured to construct samples based on the resource attribute features, the resource interactive features, and the resource quality level features of the seed multimedia resource, to obtain the positive samples.

[0140] A negative sample construction module is configured to construct samples based on the resource attribute features, the resource interactive features, and the resource quality level features of the preset multimedia resource, to obtain the negative samples.

[0141] Please refer to Figure 11 The embodiment also provides a resource recommendation apparatus, including:

[0142] A resource feature extraction module 1110 is configured to extract resource features of a candidate resource;

[0143] A resource classification module 1120 is configured to input the resource features of the candidate resource into a resource classification model to perform resource classification, to obtain a resource classification result of the candidate resource; the resource classification model is obtained by training based on positive samples and negative samples; the positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool; the seed multimedia resources are obtained by screening from the original multimedia resource pool based on preset multimedia resource screening conditions; the preset multimedia resource screening conditions include resource interactive conditions and resource quality level conditions; the seed multimedia resources are resources with a resource quality score greater than or equal to a preset quality score; the preset quality score is associated with the preset multimedia resource screening conditions.

[0144] The resource recommendation module 1130 is configured to determine the candidate resource as a resource to be recommended when the resource classification result indicates that the category of the candidate resource is a category to be recommended.

[0145] The apparatuses provided in the above embodiments can execute the methods provided in any of the embodiments of the present application, and have the corresponding function modules and advantages of executing the methods. Technical details not described in the above embodiments can be referred to the methods provided in any of the embodiments of the present application.

[0146] The embodiment further provides a computer readable storage medium, wherein the storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by a processor to perform any of the methods described above in the embodiment.

[0147] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform any of the methods described above in the embodiment.

[0148] The embodiment further provides an electronic device, and a structural diagram of the electronic device is shown in FIG. 12. Figure 12 The device 1200 can have great differences due to different configurations or performances, and can include one or more central processing units (CPUs) 1222 (for example, one or more processors) and a memory 1232, one or more storage media 1230 (for example, one or more mass storage devices) storing application programs 1242 or data 1244. The memory 1232 and the storage media 1230 can be temporary storage or persistent storage. The programs stored in the storage media 1230 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the device. Further, the central processing unit 1222 can be configured to communicate with the storage media 1230 to execute the series of instruction operations in the storage media 1230 on the device 1200. The device 1200 can further include one or more power supplies 1226, one or more wired or wireless network interfaces 1250, one or more input and output interfaces 1258, and / or one or more operating systems 1241, such as Windows Server TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TMand so on. Any of the above methods of the present embodiment can be implemented based on Figure 12 by the device shown in the figure.

[0149] The present specification provides method operation steps as described in the embodiments or flowcharts, but can include more or less operation steps based on conventional or non-inventive labor. The steps and order listed in the embodiments are only one of the many step execution orders, and do not represent the only execution order. When the system or interrupt product is executed in practice, it can be executed in sequence or in parallel (such as parallel processor or multi-threaded environment) according to the method order shown in the embodiments or the figure.

[0150] The structure shown in the present embodiment is only part of the structure related to the scheme of the present application, and does not constitute a limitation on the device to which the scheme of the present application is applied. The specific device can include more or less components than shown, or combine certain components, or have a different arrangement of components. It should be understood that the methods, devices, etc. disclosed in the present embodiment can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interface, device or unit module.

[0151] Based on such understanding, the technical scheme of the present application or the part essentially contributing to the prior art or the whole or part of the technical scheme can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0152] Those skilled in the art can further understand that the units and algorithm steps of each example described in combination with the embodiments disclosed in the specification can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the foregoing description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0153] The above-described embodiments are merely used to illustrate the technical solutions of the present application, but not limit the present application; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions recorded in the foregoing embodiments, or equivalent replacements can be made to part of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for training a resource classification model, characterized in that, The method comprises the following steps: obtaining a preset multimedia resource screening condition; the preset multimedia resource screening condition comprises a resource interaction condition of a multimedia resource and a resource quality level condition; the resource quality level condition comprises a level condition of a multimedia resource publishing object, a fan number condition of the multimedia resource publishing object, and a resource interaction condition of the multimedia resource publishing object; the resource interaction condition of the multimedia resource comprises an interaction operation chain; the interaction operation chain is used to represent continuous interaction information between a target object and a resource to be interacted with; the generation method of the interaction operation chain comprises the following steps: when it is detected that the target object is in an unlogged state and a first interaction operation of the resource to be interacted with is triggered, jumping to a target login page; when the target object successfully logs in based on the target login page, obtaining a second interaction operation of the target object on the resource to be interacted with; based on the first interaction operation in the unlogged state, the successful login operation based on the target login page, and the second interaction operation in the logged state, an interaction operation chain corresponding to the resource to be interacted with is generated; determining a third target resource from an original multimedia resource pool, which has the interaction operation chain and the number of interaction operation chains is greater than a preset number; based on the third target resource, a first target resource is determined; screening a second target resource meeting the resource quality level condition from the original multimedia resource pool; performing resource fusion on the first target resource and the second target resource to obtain a seed multimedia resource; the seed multimedia resource is a resource with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition; randomly extracting a preset multimedia resource from the original multimedia resource pool; generating a positive sample based on the seed multimedia resource and a negative sample based on the preset multimedia resource; when generating the positive sample, extracting a resource feature of the seed multimedia resource and determining a resource label of the seed multimedia resource, generating a positive sample corresponding to the seed multimedia resource based on the resource feature of the seed multimedia resource and the resource label of the seed multimedia resource; when generating the negative sample, extracting a resource feature of the preset multimedia resource and determining a resource label of the preset multimedia resource, generating a negative sample corresponding to the preset multimedia resource based on the resource feature of the preset multimedia resource and the resource label of the preset multimedia resource; training a preset classification model based on the positive sample and the negative sample to obtain a resource classification model.

2. The method of claim 1, wherein, Before the step of obtaining the preset multimedia resource screening condition, the method further comprises the following steps: determining the preset multimedia resource screening condition matching the preset quality score.

3. The method of claim 1, wherein, The step of generating a positive sample based on the seed multimedia resource and a negative sample based on the preset multimedia resource comprises the following steps: constructing a resource feature; the resource feature comprises a resource attribute feature, a resource interaction feature, and a resource quality level feature; Sample construction is performed based on the resource attribute features, resource interaction features, and resource quality level features of the seed multimedia resource, to obtain the positive sample; Sample construction is performed based on the resource attribute features, resource interaction features, and resource quality level features of the preset multimedia resource, to obtain the negative sample.

4. A resource recommendation method characterized by, The method comprises: extracting resource features of a candidate resource; inputting the resource features of the candidate resource into a resource classification model to perform resource classification, to obtain a resource classification result of the candidate resource; the resource classification model is trained based on positive samples and negative samples; the positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool; The seed multimedia resource determination method comprises: screening a first target resource from an original multimedia resource pool, the first target resource meeting a resource interaction condition of the multimedia resource; screening a second target resource from the original multimedia resource pool, the second target resource meeting a resource quality level condition; performing resource fusion on the first target resource and the second target resource to obtain the seed multimedia resource; the resource quality level condition comprises a level condition of a multimedia resource publishing object, a fan quantity condition of the multimedia resource publishing object, and a resource interaction condition of the multimedia resource publishing object; the seed multimedia resource is a resource with a resource quality score greater than or equal to a preset quality score, the preset quality score being associated with the preset multimedia resource screening condition; the resource interaction condition of the multimedia resource comprises an interaction operation chain; the interaction operation chain is used to represent continuous interaction information between a target object and a resource to be interacted; the generation method of the interaction operation chain comprises: when it is detected that the target object is in an unlogged state and a first interaction operation on the resource to be interacted is triggered, jumping to a target login page; when the target object successfully logs in based on the target login page, obtaining a second interaction operation of the target object on the resource to be interacted; based on the first interaction operation in the unlogged state, the successful login operation based on the target login page, and the second interaction operation in the logged state, generating an interaction operation chain corresponding to the resource to be interacted; the screening of the first target resource from the original multimedia resource pool comprises: determining a third target resource in the original multimedia resource pool, the third target resource having the interaction operation chain and the number of interaction operation chains being greater than a preset number; determining the first target resource based on the third target resource; in generating the positive sample, extracting resource features of the seed multimedia resource and determining resource labels of the seed multimedia resource, generating a positive sample corresponding to the seed multimedia resource based on the resource features of the seed multimedia resource and the resource labels of the seed multimedia resource; in generating the negative sample, extracting resource features of the preset multimedia resource and determining resource labels of the preset multimedia resource, generating a negative sample corresponding to the preset multimedia resource based on the resource features of the preset multimedia resource and the resource labels of the preset multimedia resource; When the resource classification result indicates that the category of the candidate resource is a to-be-recommended category, the candidate resource is determined as a to-be-recommended resource.

5. A resource classification model training apparatus characterized by comprising: Comprise: A screening condition acquisition module is configured to acquire a preset multimedia resource screening condition; The preset multimedia resource screening condition comprises a resource interaction condition of a multimedia resource and a resource quality level condition; the resource quality level condition comprises a level condition of a multimedia resource publishing object, a fan quantity condition of the multimedia resource publishing object, and a resource interaction condition of the multimedia resource publishing object; the resource interaction condition of the multimedia resource comprises an interaction operation chain; The interactive operation chain is used for representing continuous interaction information between the target object and the resource to be interacted with; The page jump module is configured to, when detecting that the target object is in a non-logged-in state and triggering a first interaction operation on the resource to be interacted with, jump to a target login page; The interaction operation acquisition module is configured to, when the target object successfully logs in based on the target login page, acquire a second interaction operation of the target object on the resource to be interacted with; The interactive operation chain generation module is configured to generate an interactive operation chain corresponding to the resource to be interacted with based on the first interaction operation in the non-logged-in state, the successful login operation based on the target login page, and the second interaction operation in the logged-in state; The resource screening module is configured to screen a seed multimedia resource from an original multimedia resource pool based on the resource interaction condition and the resource quality level condition. The seed multimedia resource is a resource with a resource quality score greater than or equal to a preset quality score, and the preset quality score is associated with the preset multimedia resource screening condition. The resource screening module includes: The first screening module is configured to screen a first target resource that meets the resource interaction condition from the original multimedia resource pool. The second screening module is configured to screen a second target resource that meets the resource quality level condition from the original multimedia resource pool. The resource fusion module is configured to perform resource fusion on the first target resource and the second target resource to obtain the seed multimedia resource. The resource extraction module is configured to randomly extract a preset multimedia resource from the original multimedia resource pool. The sample generation module is configured to generate a positive sample based on the seed multimedia resource and a negative sample based on the preset multimedia resource. In generating the positive sample, resource features of the seed multimedia resource are extracted, a resource label of the seed multimedia resource is determined, and a positive sample corresponding to the seed multimedia resource is generated based on the resource features and the resource label.

6. The apparatus of claim 5, wherein, The classification model training module is configured to train a preset classification model based on the positive sample and the negative sample to obtain a resource classification model. The device further includes:

7. The apparatus of claim 5, wherein, The screening condition determination module is configured to determine the preset multimedia resource screening condition that matches the preset quality score. The sample generation module includes: The resource feature construction module is configured to construct resource features, wherein the resource features comprise resource attribute features, resource interaction features, and resource quality level features. The positive sample construction module is configured to construct samples based on the resource attribute features, the resource interaction features, and the resource quality level features of the seed multimedia resources, to obtain the positive samples. The negative sample construction module is configured to construct samples based on the resource attribute features, the resource interaction features, and the resource quality level features of the preset multimedia resources, to obtain the negative samples.

8. A resource recommendation apparatus characterized by comprising: The resource feature extraction module is configured to extract resource features of candidate resources. The resource classification module is configured to input the resource features of the candidate resources into a resource classification model to perform resource classification, to obtain resource classification results of the candidate resources. The resource classification model is trained based on positive samples and negative samples. The positive samples are generated based on seed multimedia resources, and the negative samples are generated based on preset multimedia resources randomly extracted from an original multimedia resource pool. ​ The seed multimedia resource determination method comprises: screening a first target resource from an original multimedia resource pool, the first target resource meeting a multimedia resource resource interaction condition; screening a second target resource from the original multimedia resource pool, the second target resource meeting a resource quality level condition; performing resource fusion on the first target resource and the second target resource to obtain the seed multimedia resource; the resource quality level condition comprises a level condition of a multimedia resource publishing object, a fan quantity condition of the multimedia resource publishing object, and a resource interaction condition of the multimedia resource publishing object; the seed multimedia resource is a resource with a resource quality score greater than or equal to a preset quality score, the preset quality score being associated with the preset multimedia resource screening condition; the resource interaction condition of the multimedia resource comprises an interaction operation chain; the interaction operation chain is used to represent continuous interaction information between a target object and a resource to be interacted; the generation method of the interaction operation chain comprises: when it is detected that the target object is in an unlogged state and a first interaction operation on the resource to be interacted is triggered, jumping to a target login page; when the target object successfully logs in based on the target login page, obtaining a second interaction operation of the target object on the resource to be interacted; based on the first interaction operation in the unlogged state, the successful login operation based on the target login page, and the second interaction operation in the logged state, generating an interaction operation chain corresponding to the resource to be interacted; the screening of the first target resource from the original multimedia resource pool comprises: determining a third target resource in the original multimedia resource pool, the third target resource having the interaction operation chain and the number of interaction operation chains being greater than a preset number; determining the first target resource based on the third target resource; in generating the positive sample, extracting resource features of the seed multimedia resource and determining resource labels of the seed multimedia resource, generating a positive sample corresponding to the seed multimedia resource based on the resource features of the seed multimedia resource and the resource labels of the seed multimedia resource; in generating the negative sample, extracting resource features of the preset multimedia resource and determining resource labels of the preset multimedia resource, generating a negative sample corresponding to the preset multimedia resource based on the resource features of the preset multimedia resource and the resource labels of the preset multimedia resource; The resource recommendation module is configured to determine the candidate resource as a to-be-recommended resource when the resource classification result indicates that the category of the candidate resource is a to-be-recommended category.

9. An electronic device, comprising: The device comprises a processor and a memory, and the memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the resource classification model training method of any one of claims 1 to 3 or the resource recommendation method of claim 4.

10. A computer storage medium, characterized in that, The storage medium stores at least one instruction or at least one program, the at least one instruction or the at least one program is loaded by the processor and executes the resource classification model training method of any one of claims 1 to 3 or the resource recommendation method of claim 4.

11. A computer program product, characterised in that, The computer program product includes computer instructions stored in a computer readable storage medium; the processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the resource classification model training method of any one of claims 1 to 3 or the resource recommendation method of claim 4.

Images