Training method, device, equipment and storage medium for genre resource allocation model

Abstract

This disclosure provides a training method, apparatus, device, and storage medium for a genre resource allocation model. It relates to the technical fields of resource distribution, resource recommendation, and artificial intelligence. The specific implementation scheme is as follows: A training dataset is collected based on users' historical consumption information; a satisfaction evaluation model is trained based on the training dataset; the satisfaction evaluation model is used to predict the user's satisfaction level with resources recommended based on the training resource candidate set; based on the satisfaction evaluation model, multiple target training data points with the highest satisfaction levels are selected from the training dataset; the genre resource allocation model is trained using these multiple target training data points. The technology of this disclosure can effectively train a genre resource allocation model, thereby accurately and efficiently predicting the proportion of genre resources among multiple recommended resources.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, specifically to the fields of resource distribution, resource recommendation, and artificial intelligence, and particularly to a training method, apparatus, device, and storage medium for a genre resource allocation model. Background Technology

[0002] Driven by the growing demand for consumption efficiency and the increasing popularity of immersive consumption habits, more and more users prefer to consume resources in an immersive way.

[0003] In various resource recommendation scenarios, such as immersive recommendation, multiple resources can be recommended for each user request. Specifically, multiple resources can be filtered based on their content popularity. Summary of the Invention

[0004] This disclosure provides a training method, apparatus, device, and storage medium for a genre resource allocation model.

[0005] According to one aspect of this disclosure, a training method for a genre resource allocation model is provided, comprising:

[0006] Based on users' historical consumption information, a training dataset is collected. The training dataset includes multiple training data points, each of which includes training user features, training resource candidate set features, and training user behavior features. It also includes: features of multiple recommended resources recommended based on the training resource candidate set, consumption information of multiple recommended resources, and the actual genre ratio of multiple recommended resources.

[0007] Based on the training dataset, a satisfaction evaluation model is trained; the satisfaction evaluation model is used to predict the satisfaction level of training users with resources recommended based on the training resource candidate set.

[0008] Based on the satisfaction evaluation model, select multiple target training data with the highest satisfaction from the training dataset;

[0009] The genre resource allocation model was trained using multiple sets of target training data.

[0010] According to another aspect of this disclosure, a method for predicting resource proportion is provided, comprising:

[0011] Obtain user characteristics, resource candidate set characteristics, and user behavior characteristics respectively;

[0012] Based on the user characteristics, the resource candidate set characteristics, and the user behavior characteristics, a pre-trained genre resource allocation model is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set.

[0013] According to another aspect of this disclosure, a training apparatus for a genre resource allocation model is provided, comprising:

[0014] The data acquisition module is used to collect a training dataset based on the user's historical consumption information. The training dataset includes multiple training data points, each of which includes training user features, training resource candidate set features, and training user behavior features. It also includes features of multiple recommended resources recommended based on the training resource candidate set, as well as consumption information of the multiple recommended resources and the actual genre ratio of the multiple recommended resources.

[0015] The first training module is used to train a satisfaction evaluation model based on the training dataset; the satisfaction evaluation model is used to predict the satisfaction level of training users with resources recommended based on the training resource candidate set.

[0016] The filtering module is used to filter multiple target training data with the highest satisfaction from the training dataset based on the satisfaction evaluation model.

[0017] The second training module is used to train the genre resource allocation model using multiple target training data.

[0018] According to another aspect of this disclosure, a resource proportion prediction device is provided, comprising:

[0019] The acquisition module is used to acquire user characteristics, resource candidate set characteristics, and user behavior characteristics, respectively.

[0020] The prediction module is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set, using a pre-trained genre resource allocation model, based on the user characteristics, the resource candidate set characteristics, and the user behavior characteristics.

[0021] According to another aspect of this disclosure, an electronic device is provided, comprising:

[0022] At least one processor; and

[0023] A memory communicatively connected to the at least one processor; wherein,

[0024] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described above and any possible implementations.

[0025] According to yet another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described above and any possible implementation thereof.

[0026] According to another aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the aspects and any possible implementations described above.

[0027] According to the technology disclosed herein, a genre resource allocation model can be effectively trained, thereby accurately and efficiently predicting the proportion of genre resources among multiple recommended resources.

[0028] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0029] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0030] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure;

[0031] Figure 2 This is a schematic diagram according to the second embodiment of the present disclosure;

[0032] Figure 3 This is an architectural diagram of a resource recommendation device provided in an embodiment of this disclosure.

[0033] Figure 4 This is a schematic diagram according to the third embodiment of the present disclosure;

[0034] Figure 5 This is a schematic diagram according to the fourth embodiment of the present disclosure;

[0035] Figure 6 This is a schematic diagram according to the fifth embodiment of the present disclosure;

[0036] Figure 7 This is a schematic diagram according to the sixth embodiment of the present disclosure;

[0037] Figure 8 This is a schematic diagram according to the seventh embodiment of the present disclosure;

[0038] Figure 9 This is a schematic diagram according to the eighth embodiment of the present disclosure;

[0039] Figure 10 This is a schematic diagram according to the ninth embodiment of the present disclosure;

[0040] Figure 11 This is a schematic diagram according to the tenth embodiment of the present disclosure;

[0041] Figure 12 This is a schematic diagram according to the eleventh embodiment of the present disclosure;

[0042] Figure 13 This is a block diagram of an electronic device used to implement the methods of the embodiments of this disclosure. Detailed Implementation

[0043] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0044] Obviously, the described embodiments are only some, not all, of the embodiments disclosed herein. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0045] It should be noted that the terminal devices involved in the embodiments of this disclosure may include, but are not limited to, smart devices such as mobile phones, personal digital assistants (PDAs), wireless handheld devices, and tablet computers; the display devices may include, but are not limited to, personal computers, televisions, and other devices with display functions.

[0046] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0047] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure; as shown Figure 1 As shown, this embodiment provides a resource recommendation method applied to immersive resource recommendation scenarios, which may specifically include the following steps:

[0048] S101. Based on the user's request, recall resources of various genres respectively to obtain the recall results of resources of various genres.

[0049] The resource recommendation method in this embodiment can be implemented by a resource recommendation device, which can be an electronic entity or a software-integrated application used to provide immersive resource recommendations to users. This resource recommendation device can be applied in a resource recommendation system.

[0050] The immersive resource recommendation in this embodiment refers to the resource recommendation device autonomously and continuously recommending resources to the user without the user having a specific request. For the user, simply swiping up and down the screen allows them to continuously consume different resources, achieving an immersive experience.

[0051] In this embodiment, the genre of a resource can refer to the style or format of the resource.

[0052] In this embodiment, when a user clicks on an entry resource to enter an immersive recommendation scenario, it can be considered that the user has initiated a resource request; or, in an immersive resource recommendation scenario, if a user consumes the next resource by swiping down, and if the user has already consumed a preset proportion of the currently recommended resources, then swiping up again can also be considered that the user has initiated a resource request.

[0053] Since the technical solution of this embodiment is applied in an immersive resource recommendation scenario, the user's request can be an implicit request that does not include any specific request content or request limiting conditions.

[0054] S102. Based on the recall results of resources of various genres, sort the multiple resources of the corresponding genres to obtain the first sorting result of resources of various genres.

[0055] Specifically, the resource sorting in step 102 involves sorting multiple resources within each genre. These multiple resources can be selected from only the better-quality resources in the resource retrieval results for that genre. After this step, each genre's resources will have a first sorting result.

[0056] S103. Based on a pre-trained genre resource allocation model, configure the proportion of resources for each genre;

[0057] The genre resource allocation model in this embodiment can reasonably and accurately allocate the proportion of various genre resources, which can cater to the personalized needs of users, so as to make more targeted immersive resource recommendations to users in the future, and effectively improve the effect of immersive resource recommendations.

[0058] S104. Based on the proportion of resources of various genres and the first ranking result of resources of various genres, a comprehensive ranking of a preset number of resources of various genres is performed to obtain a second ranking result.

[0059] The sorting in step S102 involves sorting resources of various genres separately, making it easier to select high-quality resources from them. In contrast, the sorting in step S104 is a comprehensive sorting of resources from various genres, facilitating the subsequent selection of the few resources to be recommended.

[0060] S105. Based on the second ranking result, several resources are pushed to the user's client in an immersive resource recommendation manner.

[0061] In this embodiment, the resources can be the first few resources in the second sorting result. The number of resources can be set based on requirements, such as 6, 8, or other numbers, and is not limited here.

[0062] The resource recommendation method in this embodiment, by employing the above steps, can recommend resources of various genres together. This effectively overcomes the limitation of existing immersive resource recommendation scenarios, which can only recommend video resources. It effectively enriches the types of resources recommended and improves resource recommendation efficiency. Furthermore, in this embodiment, the proportion of various genres can be configured before recommendation, effectively optimizing the content of resource recommendations, improving the accuracy of resource recommendations, and thus enhancing resource recommendation efficiency.

[0063] Figure 2 This is a schematic diagram based on the second embodiment of this disclosure; the resource recommendation method of this embodiment, in the above... Figure 1 Based on the technical solutions of the illustrated embodiments, the technical solutions of this disclosure will be described in further detail. For example... Figure 2 As shown, the resource recommendation method in this embodiment may specifically include the following steps:

[0064] S201. Based on the user's request, recall resources of various genres respectively to obtain the recall results of resources of various genres.

[0065] In this embodiment, to improve the matching degree between recalled resources and users, recall can be based on user information when recalling resources of various genres. This user information may include user attribute information, such as age, gender, and occupation. Furthermore, user information may also include user interests, such as finance, entertainment, and sports. User interests can be identified in the user's attribute information in the form of tags.

[0066] Alternatively, in one embodiment of this disclosure, when recalling resources of various genres, the popularity of each resource can also be taken into account, and resources with high popularity can be recalled first.

[0067] Further optionally, in one embodiment of this disclosure, if a user is a new user and enters the immersive recommendation scenario for the first time, the resource recommendation device can recall resources of various genres according to the popularity of each resource based on the user's request, and obtain the recall results of resources of various genres.

[0068] It should be noted that the above resource recall rules are examples in the embodiments of this disclosure. In actual applications, other rules can also be used to recall resources, such as prioritizing the recall of resources whose published locations are close to the user's location based on the distance between the resource's published location information and the user's location, etc., which will not be listed in detail here.

[0069] The technical solution of this embodiment can be applied to the mixed recommendation of resources of various genres in immersive recommendation scenarios. The various genres in this embodiment can include video and text / images; wherein video can include short videos and / or mini-videos; and wherein the playback duration of mini-videos is shorter than that of short videos. Text / images include animated text / images and / or text / images. Animated text / images must include images and may include a small amount of text to describe the images. Text / images must include text and may include a small amount of images to explain the text.

[0070] S202. Obtain the top-ranked resources from the recall results of resources of various genres respectively;

[0071] In practical applications, the number of resources included in the resource retrieval results for various genres is enormous. To improve resource recommendation efficiency, the retrieved resources for various genres can be initially sorted. This sorting can be called coarse sorting. The specific sorting mechanism and method can refer to the coarse sorting in existing resource recommendation architectures, and will not be elaborated here. In this embodiment, step S202 can select several resources with high ranking based on the coarse sorting results.

[0072] S203. For each resource among multiple resources in each genre, obtain the characteristics of the resource and the user characteristics;

[0073] In this embodiment, the characteristics of a resource may include at least one of the following: resource length, resource identifier, resource tag, resource genre, resource entry genre, and estimated effect characteristics of the resource.

[0074] When the resource is a video resource, the resource length refers to the duration of the video; when the resource is a text and image resource, the resource length refers to the length of the text and the number of images included in the resource.

[0075] The entry genre feature of the resource in this embodiment can refer to the genre feature of the resource entry resource, which can identify the genre of the resource entry resource as video or text / images.

[0076] The predicted performance features of the resources in this embodiment may include at least one of the following: predicted user playback duration, predicted likes, predicted attention, predicted reposts, predicted completion rate, and predicted post-release value.

[0077] The estimated posterior value is used to characterize whether a user, after consuming a resource, will continue to want to view other resources from the same author, or other related resources, such as related content resources or resources with related tags.

[0078] The user characteristics in this embodiment may include at least one of the user's attribute characteristics and the user's historical consumption characteristics.

[0079] User attributes can include at least one of the following: basic user attributes and user preference attributes. Basic user attributes include age, gender, and occupation. User preference attributes include hobbies and interests. These preferences can be tagged in the user's attribute information; for example, hobbies might include travel, entertainment, and football. All user attributes are correlated with resources and can provide a reference for predicting user satisfaction with those resources. For example, different age groups may have different resource preferences; for instance, middle-aged and elderly people prefer video resources; middle-aged office workers prefer text and image resources; middle-aged women prefer fashion-related resources, while middle-aged men prefer financial resources; and the elderly prefer health and wellness resources, and so on.

[0080] A user's historical consumption characteristics can include at least one of the following: the percentage of consumption for various resource genres across multiple historical time periods prior to the current time period; the duration of consumption for each resource genre; and the completion rate for each resource genre. If the currently predicted resource genre has a high consumption percentage across multiple historical time periods, the corresponding satisfaction score will be higher. Conversely, if the currently predicted resource genre has the lowest consumption percentage across multiple historical time periods, the corresponding satisfaction score will be lower. Similarly, if the currently predicted resource genre has a high completion rate across multiple historical time periods, the corresponding satisfaction score will be higher. Conversely, if the currently predicted resource genre has a low completion rate across multiple historical time periods, the corresponding satisfaction score will be lower.

[0081] The length of the current time period can be a preset time length defined in the resource recommendation scenario, such as the current day, the current week, etc. Multiple historical time periods can be configured according to needs, such as historical one-day, historical three-day, historical seven-day, etc. The specific length and number of historical time periods are not limited here. The consumption percentage of various types of resources consumed by the user within each historical time period, the duration of consumption of various types of resources, and the completion rate of various types of resources can be obtained by statistically analyzing the user's historical consumption information.

[0082] Optionally, in this embodiment, scene features can also be acquired. These scene features can identify whether the current immersive resource recommendation scene belongs to a recommendation scene or a discovery scene. In this embodiment, the recommendation scene and the discovery scene can be two different sections of the application. Different users may have different scene preferences. Based on these scene features, the accuracy of predicting user satisfaction with resources can be improved.

[0083] In practical implementation, for each predicted effect feature, a corresponding prediction model can be pre-trained. When using it, at least one of the aforementioned resource features, user features, scene features, and resource entry genre features can be input into the prediction model. The prediction model can then predict and output the predicted effect. Using this method, it is possible to predict the user's estimated playback duration, estimated likes, estimated follows, estimated shares, estimated completion rate, and estimated post-release value. Each feature can be represented by a numerical value, for example, a value between 0 and 1, with higher values ​​indicating a greater probability of the corresponding effect.

[0084] S204. For each resource among multiple resources in each genre, based on the characteristics of the resource and the user characteristics, a pre-trained satisfaction prediction model is used to predict the satisfaction value of the resource.

[0085] In the specific prediction process, for each resource, its features and user characteristics are input into the satisfaction prediction model. This model can then predict and output a satisfaction value. In practical applications, this satisfaction value can be a number between 0 and 1; the higher the value, the higher the level of satisfaction with the resource, and the more worthy it is of recommendation.

[0086] Optionally, if scene features are also obtained, the scene features, along with resource features and user features, are also input into the satisfaction prediction model to improve the accuracy of the satisfaction value predicted by the satisfaction prediction model.

[0087] S205. Based on the satisfaction value of each resource, sort the multiple resources of each genre to obtain the first ranking result of each genre resource.

[0088] For example, the specific implementation of step S205 may include the following steps:

[0089] (a1) For each resource among multiple resources in each genre, multiply the satisfaction value of the resource by the score of the resource already obtained, and use the updated score of the resource.

[0090] (b1) For each resource in a genre, sort them based on the updated scores of the resources to obtain the first sorting result of the resources in each genre.

[0091] In this embodiment, the obtained resource score can refer to the score obtained during the coarse ranking of recalled resources after resource recall. The specific acquisition method can refer to the relevant schemes of the coarse and fine ranking layers of the resource recommendation device, and is not limited here.

[0092] In this embodiment, by estimating the satisfaction value of each resource, the resources can be ranked more accurately, thereby effectively improving the accuracy and efficiency of subsequent resource recommendations.

[0093] Steps S202-S205 are as described above. Figure 1 This is a specific implementation of step S102 in the illustrated embodiment. This method enables a more reasonable, accurate, and effective sorting of resources across various genres.

[0094] S206. Obtain user attribute characteristics, resource candidate set characteristics, and user behavior characteristics;

[0095] S207. Based on user attribute characteristics, resource candidate set characteristics, and user behavior characteristics, a genre resource allocation model is adopted to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set.

[0096] The user characteristics in this step may include at least one of the user's attribute characteristics and the user's historical consumption characteristics. For details, please refer to the relevant records mentioned above; they will not be repeated here.

[0097] The resource candidate set features in this embodiment include the features of multiple resources ranked at the top from the first ranking results of resources of various genres.

[0098] In this embodiment, the number of resource candidate sets can be pre-configured, such as 400. To enrich the resource genres in the candidate sets, some top-ranked resources need to be selected from the first ranking results of resources of various genres. For example, if multi-genre resources include short videos, animated graphics, and text-based graphics, the top 100 resources of each genre can be selected. Alternatively, a minimum limit can be set for each genre, such as selecting a minimum of 50 resources, with the remaining number supplemented by resources of other genres. Regardless of the method used, as long as the resource candidate sets include various resource genres and the number reaches the required number of resource candidate sets, the selection is acceptable.

[0099] The resource candidate set features include feature representations of multiple resources; each resource's feature representation is obtained by vector representation and pooling operations based on at least one of the resource's title, resource's tag, resource's genre, and resource's length; it also includes the average feature representation of the resource candidate set.

[0100] User behavior characteristics include users' historical consumption characteristics, users' satisfaction sequence characteristics, and / or the context characteristics of users' consumption resources;

[0101] A user's historical consumption characteristics include at least one of the following: the percentage of consumption of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres.

[0102] The user satisfaction sequence features include the feature representations of the multiple resources that the user was most satisfied with within a preset time period prior to the current time period. The specific number of resources included can be set according to needs or experience, such as 10 or 20, and is not limited here. Specifically, the feature representations of the resources are obtained in the same way as the feature representations of the resources in the resource candidate set features mentioned above.

[0103] Steps S206-S207 are as described above. Figure 1 This is a specific implementation of step S103 in the illustrated embodiment. This method can accurately predict the proportion of resources in various genres, facilitating more targeted resource recommendations and effectively improving the accuracy and efficiency of resource recommendations.

[0104] S208. Based on the proportion of resources of various genres, obtain the second preset number of resources ranked first from the first ranking result of resources of various genres; the second preset number of resources includes resources of various genres.

[0105] S209. Perform a comprehensive sorting of the second preset number of resources to obtain the second sorting result;

[0106] Specifically, for resources of various genres, the quantity of resources for that genre can be obtained by multiplying a second preset quantity by the proportion of resources of that genre. Then, from the first sorting result of resources of that genre, the corresponding quantity of resources is selected in order from front to back. In this way, a total of the second preset quantity of resources can be obtained, providing effective resource support for subsequent immersive resource recommendations.

[0107] Further optionally, step S209 of this embodiment may include the following steps in its specific implementation:

[0108] (a2) For each resource in the second preset number of resources, obtain at least one of the following: the score of the obtained resource, the contextual relevance factor of the resource and the previous recommended resource, the quality score of the resource, the predicted effect characteristics of the resource, the predicted decline rate, and the predicted exit rate.

[0109] The contextual relevance factor between the current resource and the recommended previous resource can be calculated using a pre-trained relevance factor prediction model. Specifically, the model is fed the feature representations of the current resource and the previous resource, and it can predict and output the contextual relevance factor between the current resource and the recommended previous resource.

[0110] The score of the acquired resources can refer to the score updated in step S205.

[0111] The quality score of a resource can be used to comprehensively characterize the quality of the resource, and can be obtained by using quality assessment strategies or quality assessment models.

[0112] The predicted effect characteristics of the resources can be referred to the relevant descriptions in the above embodiments, and will not be repeated here.

[0113] The estimated decline rate and estimated exit rate of a resource are used to predict the probability that a user will scroll down and consume the next resource after clicking on the current resource, and the probability that a user will exit when clicking on the current resource. They can respectively represent the user's preference for the current resource and the resulting aftereffects.

[0114] Specifically, the estimated decline rate and estimated exit rate of resources can also be predicted using a pre-trained prediction model, referencing the methods for obtaining the predicted performance characteristics of resources. In practice, accurate predictions can be made based on user characteristics and resource characteristics.

[0115] (b2) Obtain the weight factors of each feature based on the Evolution Strategies (ES) algorithm; the weight factors are different for resources of different genres;

[0116] Specifically, the ES algorithm is used to optimize and find the optimal weight factors for each feature. For details, please refer to the implementation principles of the ES algorithm.

[0117] Different genres of resources have different focuses on the characteristics considered. In order to improve accuracy, different weighting factors should be found for resources of different genres.

[0118] (c2) Based on the weight factors of each feature, and at least one of the scores of each resource, the context relevance factors of each resource, the quality score of each resource, the predicted effect features of each resource, the predicted decline rate and the predicted exit rate, the second preset number of resources are merged and sorted to obtain the second sorting result.

[0119] For example, in a specific implementation, for each resource in the second preset number of resources, at least one of the resource's score, the resource's context relevance factor, the resource's quality score, the estimated effect feature of each resource, the estimated decline rate, and the estimated exit rate can be multiplied by the weight factor of the corresponding feature to obtain the comprehensive score of the resource; then, based on the comprehensive score of each resource, the second preset number of resources are merged and sorted to obtain the second sorting result.

[0120] Similarly, the more features referenced during fusion ranking, the higher the accuracy of the ranking. Preferably, during fusion ranking, all features mentioned above are referenced, including resource scores, resource context relevance factors, resource quality scores, resource predicted performance features, predicted decline rate, and predicted exit rate.

[0121] Steps S208-S209 are as described above. Figure 1 This is a specific implementation of step S104 in the illustrated embodiment. This method allows for the accurate acquisition of the weight factors of each feature of each resource, thereby enabling a more reasonable and accurate sorting of the second preset number of resources.

[0122] S210. From the second sorting result, obtain the top-ranked resources;

[0123] S211. Push several resources to the user's client in an immersive resource recommendation manner.

[0124] In this embodiment, since the second sorting result is a comprehensive sorting of resources of various genres, the genres of the several resources pushed to the user can be the same or different, which can effectively enrich the types of resources recommended in the immersive resource recommendation scenario.

[0125] The immersive resource recommendation in this embodiment specifically refers to proactively and seamlessly recommending several resources to the user without any user request. After consuming one resource, the user can simply scroll down to continue consuming the next resource.

[0126] The resource recommendation method in this embodiment, taking the recommendation of multiple resources in a single request as an example, in practical applications, if a user consumes a preset proportion of resources in an immersive recommendation scenario, and continues to scroll down to consume more resources, a next request will be triggered. The resource recommendation device will then continue to recommend resources to the user according to the resource recommendation method of this embodiment. This process is imperceptible to the user; they only experience the immersive resource consumption.

[0127] Figure 3This is an architecture diagram of a resource recommendation device provided in this embodiment, which may specifically include a recall layer, a coarse-to-fine ranking layer, a resource proportion configuration layer, and a re-ranking layer. Specifically, the above-mentioned step S201 of this embodiment can be performed in the recall layer of the resource recommendation device. The steps S202-S205 can be performed in the coarse-to-fine ranking layer of the resource recommendation device. The steps S208-S209 can be performed in the re-ranking layer of the resource recommendation device. Steps S206-S207 can be considered as predicting the proportion of various genre resources based on the resources in the coarse-to-fine ranking layer, thereby effectively providing data support for the resources in the re-ranking layer, specifically in the resource proportion configuration layer. Finally, based on the ranking result of the re-ranking layer, i.e., the second ranking result, the top-ranked resources can be selected and pushed to the user.

[0128] The resource recommendation method in this embodiment, by adopting the above technical solution, can effectively enrich the genre of resource recommendation in immersive recommendation scenarios.

[0129] Furthermore, in this embodiment, in the coarse-fine ranking layer, a pre-trained satisfaction model can be used to predict the satisfaction of each resource, enabling accurate ranking of resources for each genre within each genre.

[0130] Furthermore, in this embodiment, a pre-trained genre resource allocation model can be used to accurately predict the proportion of various genre resources, which can improve the funnel efficiency of the resource allocation stage, provide more accurate resources to the downstream, and thus effectively improve the efficiency of resource recommendation.

[0131] Furthermore, in this embodiment, the ES algorithm can be used to find the optimal feature weight factor in the reordering layer, so as to calculate the score of each resource more reasonably and accurately, and thus sort the resources of each genre more reasonably and accurately.

[0132] Figure 4 This is a schematic diagram based on the third embodiment of this disclosure; as shown Figure 4 As shown, this embodiment provides the above-mentioned Figure 1 or Figure 2 The training method for the satisfaction prediction model in the illustrated embodiment may specifically include the following steps:

[0133] S401. Configure the training data set, which includes the characteristics of the training user, the characteristics of the two training resources, and the true satisfaction relationship between the training user and the two training resources.

[0134] S402. Based on the characteristics of the training users and the two training resources in the training data set, a satisfaction prediction model is used to predict the predicted satisfaction values ​​of the training users for the two training resources respectively.

[0135] S403. Based on the predicted satisfaction values ​​of the training users for the two training resources and the actual satisfaction relationship between the training users for the two training resources, the parameters of the satisfaction prediction model are adjusted.

[0136] The training device for the satisfaction prediction model in this embodiment is the entity that performs the training method for the satisfaction prediction model.

[0137] In the training process of the satisfaction prediction model in this embodiment, training data sets are used to train the model. Specifically, each training data set includes the features of a training user, the features of two training resources, and the training user's true satisfaction relationship with the two training resources. Moreover, the training user's satisfaction with the two training resources cannot be at the same level, so the true satisfaction relationship cannot be equal; it must be greater than or less than the training resource.

[0138] During the training process, for each training resource, the characteristics of the training user and the characteristics of the training resource are input into the satisfaction prediction model. This model can predict and output the predicted satisfaction value of the training user for that training resource. The predicted satisfaction value can be a value between 0 and 1, with a larger value indicating higher satisfaction and vice versa.

[0139] In this embodiment, the training only uses the actual satisfaction relationship between training users and two training resources as supervision, which can effectively learn the relationship between the user's satisfaction with different resources, effectively reducing the training difficulty of the satisfaction prediction model and improving the training efficiency of the model.

[0140] The training method of the satisfaction prediction model in this embodiment involves configuring a training data set and adjusting the parameters of the satisfaction prediction model based on the predicted satisfaction values ​​of the training users for the two training resources and the actual satisfaction relationship between the training users and the two training resources. This makes the satisfaction prediction model more accurate.

[0141] Figure 5 This is a schematic diagram based on the fourth embodiment of the present disclosure; as shown Figure 5 As shown, the training method of the satisfaction prediction model in this embodiment is based on the above... Figure 4 Based on the technical solutions of the illustrated embodiments, the technical solutions of this disclosure will be described in more detail, which may specifically include the following steps:

[0142] S501. Collect the characteristics of the training user and the consumption information of the training user from the user's historical consumption information;

[0143] S502. Based on the pre-configured resource satisfaction strategy table, extract information on two training resources with different consumption satisfaction levels from the consumption information of training users.

[0144] S503. Based on the information extracted from the two training resources, obtain the features of the two training resources;

[0145] S504. Based on the consumption satisfaction levels of the two training resources, configure the true satisfaction relationship between the two training resources.

[0146] For example, before step S502, the system may further include: configuring a resource satisfaction strategy table, which includes a strategy for identifying satisfied consumption; a strategy for identifying seamless consumption; and a strategy for identifying unsatisfactory consumption; wherein the satisfaction level of satisfied consumption is greater than the satisfaction level of seamless consumption; and the satisfaction level of seamless consumption is greater than the satisfaction level of unsatisfactory consumption.

[0147] Optionally, in one embodiment of this disclosure, configuring the resource satisfaction strategy table may specifically include the following steps:

[0148] (a3) Based on the preset satisfaction behavior judgment conditions, configure the identification strategy for satisfied consumption and the identification strategy for imperceptible consumption;

[0149] For example, if a user's consumption behavior meets the preset satisfaction criteria, the user's consumption level is determined to be a satisfied consumption level; if the user's consumption behavior does not meet the preset satisfaction criteria, the user's consumption level is determined to be an indifferent consumption level.

[0150] For example, the criteria for judging satisfaction behavior may include: the user's consumption time is greater than a first preset time or a preset time ratio, and the user has at least one interactive behavior, in which case the user is considered satisfied.

[0151] The preset duration and percentage can be set according to the actual scenario, and are not limited here. For example, the duration can be 10 seconds, 15 seconds, or 20 seconds. The preset duration percentage can also be set to 60%, 70%, or 80% based on experience, and is not limited here.

[0152] User interactions can include at least one of the following: liking, commenting, following, and sharing.

[0153] (b3) Configure an identification strategy for unsatisfactory consumption based on preset unsatisfactory behavior judgment conditions.

[0154] In this embodiment, a preset condition for judging dissatisfaction needs to be configured separately. When this condition is met, the user's consumption level can be determined to be an unsatisfactory consumption level. For example, the condition for judging dissatisfaction could be: if the consumption duration exceeds a second preset duration threshold, or if the user exits, it is determined that the user is dissatisfied with the resources; the second preset duration threshold is less than a first preset duration threshold. For example, the second preset duration threshold could be 2 seconds, 3 seconds, or other time lengths.

[0155] The features of the training user in this embodiment include at least one of the training user's attribute features and the training user's historical consumption features;

[0156] The attribute characteristics of the training users include at least one of the basic attribute characteristics and the preference characteristics of the training users; the historical consumption characteristics of the training users include at least one of the consumption ratio of various genres of resources consumed by the training users in multiple historical time periods before the consumption period of the corresponding training resources, and the completion rate of various genres of resources.

[0157] The characteristics of training resources include at least one of the following: length of training resources, identifier of training resources, label of training resources, genre of training resources, entry genre of training resources, and predicted effect characteristics of training resources.

[0158] When the training resource is a video resource, the length of the training resource refers to the duration of the video; when the training resource is a text and image resource, the length of the training resource refers to the length of the text and the number of images included in the resource.

[0159] The entry genre of training resources refers to the genre of the entry resource. The genre that can identify the entry resource of training resources is video or text / images.

[0160] The effectiveness characteristics of training resources include at least one of the following: training user playback duration, likes, follows, shares, completion rate, and post-viewing value.

[0161] Steps S501-S504 are as described above. Figure 4 One specific implementation of step S401 in the illustrated embodiment.

[0162] S505. Based on the characteristics of the training users and the two training resources in the training data set, a satisfaction prediction model is used to predict the predicted satisfaction values ​​of the training users for the two training resources respectively.

[0163] S506. Based on the predicted satisfaction values ​​of training users for the two training resources, obtain the predicted satisfaction relationship between training users and the two training resources.

[0164] S507. Based on the predicted satisfaction relationship between training users and the actual satisfaction relationship between training users and the two training resources, adjust the parameters of the satisfaction prediction model.

[0165] Specifically, the model checks whether the predicted satisfaction relationship between training users and the actual satisfaction relationship between training users and the two training resources are consistent. If they are inconsistent, the parameters of the satisfaction prediction model are adjusted so that the model can learn to predict the actual satisfaction relationship between users and the two training resources.

[0166] The training method of the satisfaction prediction model in this embodiment involves configuring a training data set and adjusting the parameters of the satisfaction prediction model based on the predicted satisfaction values ​​of the training users for the two training resources and the actual satisfaction relationship between the training users and the two training resources. This makes the satisfaction prediction model more accurate.

[0167] Figure 6 This is a schematic diagram according to the fifth embodiment of this disclosure; as shown Figure 6 As shown, this embodiment provides a training method for a genre resource allocation model, which may specifically include the following steps:

[0168] S601. Collect a training dataset based on users' historical consumption information;

[0169] The training dataset includes multiple training data sets, each containing training user features, training resource candidate set features, and training user behavior features; it also includes: features of multiple recommended resources based on the training resource candidate set, consumption information of multiple recommended resources, and the actual genre ratio of multiple recommended resources;

[0170] One training data point corresponds to a single resource recommendation session, or the data from a single user request for resource recommendations. Each feature in the training data can be calculated or statistically derived based on the user's historical consumption information.

[0171] S602. Based on the training dataset, train a satisfaction evaluation model; the satisfaction evaluation model is used to predict the degree of satisfaction of training users with resources recommended based on the training resource candidate set;

[0172] S603. Based on the satisfaction evaluation model, select multiple target training data with the highest satisfaction from the training dataset;

[0173] S604. Use multiple target training data to train the genre resource allocation model.

[0174] The training method of the genre resource allocation model in this embodiment is based on the structure of the generator-evaluator (GE) model.

[0175] In this embodiment, by using a training dataset to train the satisfaction assessment model, the model can accurately score the satisfaction level of each training data point in the dataset. Then, using this satisfaction assessment model, multiple target training data points with the highest satisfaction levels are selected from the training dataset. For example, the top 30% or 20% of training data points with the highest satisfaction levels can be selected as target training data. The genre resource allocation model is then trained using this target training data, enabling it to learn the proportion of genre resources allocated to high-quality samples.

[0176] The training method of the genre resource allocation model in this embodiment can select high-quality training data with very high user satisfaction to train the genre resource allocation model, thereby effectively improving the accuracy of the trained genre resource allocation model.

[0177] Further optionally, step S602 of the above embodiment may include the following steps in its specific implementation:

[0178] (a4) For each of the training data in the training dataset, based on the training user characteristics, training resource candidate set characteristics, training user behavior characteristics and multiple recommended resources in the training data, the satisfaction evaluation model is used to predict the predicted satisfaction of the multiple recommended resources.

[0179] (b4) Based on the consumption information of multiple recommended resources and the preset satisfaction behavior judgment conditions, obtain the true satisfaction of the multiple resources;

[0180] For example, based on a user's consumption information for multiple recommended resources, it can be determined whether the user is satisfied with each individual recommended resource. Individual satisfaction can refer to the satisfaction judgment conditions for a single resource in the aforementioned embodiments. For example, it can include at least one of the following: the user's consumption time exceeds a first preset time or a preset time ratio, and the user exhibits interactive behavior. In such cases, the user is considered satisfied.

[0181] If a user's satisfaction rate for a single recommended resource within a single request (i.e., a refresh of multiple resources) is greater than or equal to a preset threshold (e.g., 70% or 75%), the user is considered satisfied with the recommended resources. Conversely, if the user's satisfaction rate for a single recommended resource within a single refresh is less than the preset threshold, the user is considered dissatisfied with the recommended resources.

[0182] (c4) Adjust the parameters of the satisfaction prediction model based on the predicted and actual satisfaction of multiple recommended resources.

[0183] Further optionally, step S604 of the above embodiment may include the following steps in its specific implementation:

[0184] (a5) For each target training data in the target training dataset, based on the training user characteristics, training resource candidate set characteristics and training user behavior characteristics in the target training data, a genre resource allocation model is adopted to predict the predicted genre proportion of multiple recommended resources.

[0185] (b5) Adjust the parameters of the genre resource allocation model based on the predicted genre proportion and the actual genre proportion of multiple recommended resources.

[0186] By adjusting the parameters of the genre resource allocation model, the predicted genre proportions predicted by the model are made consistent with the actual genre proportions.

[0187] Using multiple training datasets, the genre resource allocation model is continuously trained in the manner described above until the cutoff condition is met, thus obtaining the genre resource allocation model.

[0188] Further optionally, in one embodiment of this disclosure, the training user features include the attribute features of the training user; the attribute features of the training user include at least one of the basic attribute features of the training user and the user's preference features.

[0189] Further optionally, in one embodiment of this disclosure, the training resource candidate set includes feature representations of multiple training resources; the feature representation of each training resource is obtained by performing vector representation and pooling operations based on at least one of the title, label, genre, and length of the training resource; and also includes the average feature representation of the training resource candidate set.

[0190] Further optionally, in one embodiment of this disclosure, the training user behavior characteristics include training the user's historical consumption characteristics, training the user's satisfaction sequence characteristics, and / or training the user's consumption resource scenario characteristics;

[0191] The historical consumption characteristics of training users include at least one of the following: the proportion of users consuming resources of various genres in multiple historical time periods before the current time period, the duration of consuming resources of various genres, and the completion rate of resources of various genres.

[0192] The training user satisfaction sequence features include the feature representations of multiple resources that the user is most satisfied with within a preset time period before the current time period.

[0193] It should be noted that, although the target training data is also selected from the training dataset to select the training data with the highest satisfaction, in this embodiment, the number of training resources included in the training resource candidate set used in training the satisfaction evaluation model can be less than the number of training resources included in the training resource candidate set used in training the genre resource allocation model.

[0194] However, both models are trained using data from the same resource recommendation process. When training the satisfaction assessment model, the training resources in the candidate resource set can be obtained from the results of the re-ranking layer of the resource recommendation device. Conversely, when training the genre resource allocation model, the training resources in the candidate resource set can be obtained from the data after the coarse-fine ranking layer of that resource recommendation process.

[0195] In the above embodiments, by employing the rich and comprehensive features described above, the genre resource allocation model can be effectively trained. Furthermore, the above embodiments also employ a GE (Generic Object Optimization) training mode, which can filter out training data with very high user satisfaction to train the genre resource allocation model, effectively improving the accuracy of the trained model.

[0196] Figure 7 This is a schematic diagram according to the sixth embodiment of this disclosure; as shown Figure 7 As shown, this embodiment provides a method for predicting resource proportion, which may specifically include the following steps:

[0197] S701. Obtain user characteristics, resource candidate set characteristics, and user behavior characteristics respectively;

[0198] S702. Based on user characteristics, resource candidate set characteristics, and user behavior characteristics, a pre-trained genre resource allocation model is used to predict the proportion of various genre resources among multiple resources recommended to users based on the resource candidate set.

[0199] In practical use, user characteristics, resource candidate set characteristics, and user behavior characteristics are input into the genre resource allocation model. This genre resource allocation model can predict and output the proportion of various genre resources among multiple resources recommended to users based on the resource candidate set.

[0200] The solution in this embodiment can be applied after the coarse ranking layer and before the re-ranking layer of the resource recommendation device. It can effectively predict the proportion of various genre resources among multiple resources recommended to the user, and then make targeted resource recommendations to the user based on the proportion, thereby improving the accuracy and efficiency of resource recommendation.

[0201] Further optionally, in step S701 of the above embodiments, obtaining user characteristics includes:

[0202] Obtain at least one of the user's attribute characteristics and the user's historical consumption characteristics; the user's attribute characteristics include at least one of the user's basic attribute characteristics and the user's preference characteristics; the user's historical consumption characteristics include at least one of the user's consumption ratio of various genres of resources in multiple historical time periods before the current time period and the completion rate of various genres of resources.

[0203] Further optionally, in step S701 of the above embodiments, obtaining the resource candidate set features includes:

[0204] Obtain feature representations of multiple resources in the resource candidate set; each resource's feature representation is obtained by vector representation and pooling operation based on at least one of the resource's title, resource's tag, resource's genre, and resource's length.

[0205] Based on the feature representations of multiple resources in the resource candidate set, the average feature representation of the resource candidate set is obtained.

[0206] Further optionally, in step S701 of the above embodiments, obtaining user behavior characteristics includes:

[0207] Obtain users' historical consumption characteristics, users' satisfaction sequence characteristics, and / or the scenario characteristics of users' resource consumption;

[0208] A user's historical consumption characteristics include at least one of the following: the percentage of consumption of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres.

[0209] The user satisfaction sequence features include the feature representations of the multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

[0210] In the above embodiments, by employing the rich and comprehensive features and using the genre resource allocation model, the proportion of various genre resources in multiple resources can be accurately and effectively predicted.

[0211] The resource proportion prediction method in this embodiment can be used in addition to the methods described above. Figure 1 or Figure 2 The scenario shown in the embodiment can also be applied to other multi-genre resource recommendation scenarios, and is not limited here.

[0212] Figure 8 This is a schematic diagram according to the seventh embodiment of the present disclosure; as shown Figure 8 As shown, this embodiment provides a resource recommendation device 800, applied in an immersive resource recommendation scenario, including:

[0213] The recall module 801 is used to recall resources of various genres based on user requests, and obtain the recall results of resources of various genres.

[0214] The first sorting module 802 is used to sort multiple resources of a corresponding genre based on the recall results of resources of various genres, and obtain the first sorting result of resources of various genres.

[0215] The proportion configuration module 803 is used to configure the proportion of various genre resources based on a pre-trained genre resource allocation model.

[0216] The second sorting module 804 is used to comprehensively sort a preset number of resources of various genres based on the proportion of resources of various genres and the first sorting result of resources of various genres, and obtain the second sorting result.

[0217] The recommendation module 805 is used to push several resources to the user's client in an immersive resource recommendation manner based on the second ranking result.

[0218] The resource recommendation device 800 in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0219] Figure 9 This is a schematic diagram based on the eighth embodiment of the present disclosure; as shown Figure 9 As shown, this embodiment provides a resource recommendation device 900, in the above... Figure 8 Based on the technical solutions of the illustrated embodiments, the technical solutions of this disclosure will be described in further detail. For example... Figure 9 As shown, the resource recommendation device 900 of this embodiment includes the above-described... Figure 8 The modules with the same name and function shown are: recall module 901, first sorting module 902, proportion configuration module 903, second sorting module 904, and recommendation module 905.

[0220] like Figure 9 As shown, in the resource recommendation device 900 of this embodiment, the first sorting module 902 includes:

[0221] The first resource acquisition unit 9021 is used to acquire multiple resources ranked at the top from the recall results of resources of various genres.

[0222] The first feature acquisition unit 9022 is used to acquire, for each resource among multiple resources of each genre, at least one of the following: resource features, user features, scene features, entry genre features of the resource, and estimated effect features of the resource.

[0223] The first prediction unit 9023 is used to predict the satisfaction value of each resource among multiple resources of each genre, based on at least one of the features of the resource, the user features, the scene features, the entry genre features of the resource, and the estimated effect features of the resource, using a pre-trained satisfaction model satisfaction prediction model.

[0224] The first sorting unit 9024 is used to sort multiple resources of each genre based on the satisfaction value of each resource, and obtain the first sorting result of resources of each genre.

[0225] Further optionally, in one embodiment of this disclosure, the first sorting unit 9024 is used for:

[0226] For each resource among multiple resources in each genre, the satisfaction value of the resource is multiplied by the score of the resource already obtained, and the result is the updated score of the resource.

[0227] For each resource in a given genre, the resources are sorted based on their updated scores to obtain the first sorting result for each genre.

[0228] Further optionally, in one embodiment of this disclosure, the features of the resource include at least one of the following: the length of the resource, the identifier of the resource, the tag of the resource, and the genre of the resource.

[0229] When the resource is a video resource, the length of the resource refers to the duration of the video; when the resource is a text and image resource, the length of the resource refers to the length of the text and the number of images included in the resource.

[0230] The user characteristics include at least one of the user's attribute characteristics and the user's historical consumption characteristics; the user's attribute characteristics include at least one of the user's basic attribute characteristics and the user's preference characteristics; the user's historical consumption characteristics include at least one of the user's consumption ratio of various genres of resources in multiple historical time periods before the current time period and the completion rate of various genres of resources.

[0231] The scene feature indicates that it currently belongs to a recommended scene or a discovered scene;

[0232] The entry genre feature of the resource refers to the genre feature of the resource entry point, which can identify whether the resource entry point is video or text / image.

[0233] The predicted performance characteristics of the resource include at least one of the following: predicted playback duration, predicted likes, predicted follower count, predicted share count, predicted completion rate, and predicted post-release value.

[0234] Further optional, such as Figure 9 As shown, in one embodiment of this disclosure, the proportion configuration module 903 includes:

[0235] The second feature acquisition unit 9031 is used to acquire user attribute features, resource candidate set features and user behavior features;

[0236] The second prediction unit 9032 is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the user's attribute characteristics, the resource candidate set characteristics, and the user's behavior characteristics, using a pre-trained genre resource allocation model.

[0237] Further optionally, in one embodiment of this disclosure, the resource candidate set features include feature representations of multiple resources; each resource feature representation is obtained by performing vector representation and pooling operations based on at least one of the resource title, resource tag, resource genre, and resource length; and also includes the average feature representation of the resource candidate set;

[0238] The user behavior characteristics include the user's historical consumption characteristics, the user's satisfaction sequence characteristics, and / or the user's consumption resource scenario characteristics;

[0239] The user's historical consumption characteristics include at least one of the following: the user's consumption percentage of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres.

[0240] The user's satisfaction sequence features include the feature representations of the multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

[0241] Further optional, such as Figure 9 As shown, in one embodiment of this disclosure, the second sorting module 904 includes:

[0242] The second resource acquisition unit 9041 is used to acquire a second preset number of resources ranked first from the first ranking result of various genre resources according to the proportion of resources of various genres; the second preset number of resources includes resources of various genres.

[0243] The second sorting unit 9042 is used to perform a comprehensive sorting of the second preset number of resources to obtain the second sorting result.

[0244] Further optionally, in one embodiment of this disclosure, the second sorting unit 9042 is used for:

[0245] For each of the resources in the second preset number of resources, obtain at least one of the following: the score of the obtained resource, the contextual relevance factor of the resource and the previous recommended resource, the quality score of the resource, the estimated performance feature of the resource, the estimated decline rate, and the estimated exit rate.

[0246] Based on the evolutionary strategy algorithm, the weight factors of each feature are obtained; the weight factors are different for resources of different genres.

[0247] Based on the weighting factors of each feature, and at least one of the scores of each resource, the context relevance factors of each resource, the quality score of each resource, the predicted performance features of each resource, the predicted decline rate, and the predicted exit rate, the second preset number of resources are fused and sorted to obtain the second sorting result.

[0248] Further optionally, in one embodiment of this disclosure, the second sorting unit 9042 is used for:

[0249] For each of the resources in the second preset number of resources, the resource's score, the resource's context relevance factor, the resource's quality score, the estimated performance feature of each resource, the estimated decline rate, and the estimated exit rate are multiplied by the weight factor of the corresponding feature to obtain the comprehensive score of the resource.

[0250] Based on the comprehensive score of each resource, the second preset number of resources are merged and sorted to obtain the second sorting result.

[0251] Further optional, such as Figure 9 As shown, in one embodiment of this disclosure, the recommendation module 905 includes:

[0252] The third resource acquisition unit 9051 is used to acquire the top-ranked resources from the second sorting result;

[0253] The push unit 9052 is used to push several resources to the user's client in an immersive resource recommendation manner.

[0254] Further optionally, in one embodiment of this disclosure, the various genres include video and text / images; the video includes short videos and / or mini videos; the text / images include animated text / images and / or text / images.

[0255] The resource recommendation device 900 in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0256] Figure 10 This is a schematic diagram based on the eighth embodiment of the present disclosure; as shown Figure 10 As shown, this embodiment provides a training device 1000 for a satisfaction prediction model, including:

[0257] The configuration module 1001 is used to configure a training data set, which includes the characteristics of the training user, the characteristics of two training resources, and the true satisfaction relationship between the training user and the two training resources; and the true satisfaction relationship is greater than or less than.

[0258] The prediction module 1002 is used to predict the predicted satisfaction values ​​of the training users for the two training resources based on the characteristics of the training users in the training data group and the characteristics of the two training resources, using a satisfaction prediction model.

[0259] The adjustment module 1003 is used to adjust the parameters of the satisfaction prediction model based on the predicted satisfaction values ​​of the training users for the two training resources and the actual satisfaction relationship between the training users for the two training resources.

[0260] The training device 1000 for the satisfaction prediction model in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0261] Further, alternatively, in the above Figure 10 Based on the technical solution of the illustrated embodiment, in one embodiment of this disclosure, the adjustment module 1003 is used for:

[0262] Based on the predicted satisfaction values ​​of the training users for the two training resources, the predicted satisfaction relationship between the training users and the two training resources is obtained.

[0263] Based on the predicted satisfaction relationship between the training users and the actual satisfaction relationship between the training users and the two training resources, the parameters of the satisfaction prediction model are adjusted.

[0264] Further optionally, in one embodiment of this disclosure, the configuration module 1001 is configured to:

[0265] The characteristics of the training user and the consumption information of the training user are collected from the user's historical consumption information.

[0266] Based on the pre-configured resource satisfaction strategy table, information on two training resources with different consumption satisfaction levels is extracted from the consumption information of the training users.

[0267] Based on the information extracted from the two training resources, the features of the two training resources are obtained.

[0268] Based on the consumer satisfaction levels of the two training resources, configure the true satisfaction relationship between the two training resources.

[0269] Further optionally, in one embodiment of this disclosure, the configuration module 1001 is also configured to:

[0270] Configure the resource satisfaction strategy table, which includes a strategy for identifying satisfied consumption, a strategy for identifying seamless consumption, and a strategy for identifying unsatisfactory consumption; the satisfaction level of satisfied consumption is higher than the satisfaction level of seamless consumption; the satisfaction level of seamless consumption is higher than the satisfaction level of unsatisfactory consumption.

[0271] Further optionally, in one embodiment of this disclosure, the configuration module 1001 is configured to:

[0272] Based on preset satisfaction behavior judgment conditions, configure the identification strategy for the satisfied consumption and the identification strategy for the imperceptible consumption;

[0273] Based on preset criteria for judging unsatisfactory behavior, a strategy for identifying unsatisfactory consumption is configured.

[0274] Further optionally, in one embodiment of this disclosure, the features of the training user include at least one of the training user's attribute features and the training user's historical consumption features;

[0275] The attribute characteristics of the training users include at least one of the basic attribute characteristics and the preference characteristics of the training users; the historical consumption characteristics of the training users include at least one of the consumption ratio of various genres of resources consumed by the training users in multiple historical time periods before the corresponding training resource consumption period, and the completion rate of resources of various genres.

[0276] Further optionally, in one embodiment of this disclosure, the features of the training resource include at least one of the following: length of the training resource, identifier of the training resource, label of the training resource, genre of the training resource, entry genre of the training resource, and effect features of the training resource.

[0277] When the training resource is a video resource, the length of the training resource refers to the duration of the video; when the training resource is a text and image resource, the length of the training resource refers to the length of the text included in the resource and the number of images included.

[0278] The entry genre of the training resource refers to the genre of the entry resource of the training resource, which can identify the entry resource of the training resource as video or text and images.

[0279] The effectiveness characteristics of the training resources include at least one of the following: playback duration, likes, follows, shares, completion rate, and post-viewing value of the training resources by the training users.

[0280] In the above embodiments, the implementation principle and technical effect of training the satisfaction prediction model by using the above modules are the same as those in the above related method embodiments. For details, please refer to the description of the above related method embodiments, which will not be repeated here.

[0281] Figure 11 This is a schematic diagram based on the tenth embodiment of this disclosure; as shown Figure 11 As shown, this embodiment provides a training device 1100 for a genre resource allocation model, including:

[0282] The acquisition module 1101 is used to acquire a training dataset based on the user's historical consumption information. The training dataset includes multiple training data, each of which includes training user features, training resource candidate set features, and training user behavior features. It also includes features of multiple recommended resources recommended based on the training resource candidate set, as well as consumption information of the multiple recommended resources and the actual genre ratio of the multiple recommended resources.

[0283] The first training module 1102 is used to train a satisfaction evaluation model based on the training dataset; the satisfaction evaluation model is used to predict the satisfaction level of training users with resources recommended based on the training resource candidate set.

[0284] The filtering module 1103 is used to filter multiple target training data with the highest satisfaction from the training dataset based on the satisfaction evaluation model.

[0285] The second training module 1104 is used to train the genre resource allocation model using multiple target training data.

[0286] The training device 1100 for the genre resource allocation model in this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0287] Further optionally, in one embodiment of this disclosure, the first training module 1102 is used for:

[0288] For each training data in the training dataset, based on the training user characteristics, training resource candidate set characteristics, training user behavior characteristics, and multiple recommended resources in the training data, the satisfaction evaluation model is used to predict the predicted satisfaction of the multiple recommended resources.

[0289] Based on the consumption information of the multiple recommended resources and the preset satisfaction strategy, the true satisfaction level of the multiple resources is obtained.

[0290] Based on the predicted and actual satisfaction levels of the multiple recommended resources, the parameters of the satisfaction prediction model are adjusted.

[0291] Further optionally, in one embodiment of this disclosure, the second training module 1104 is used for:

[0292] For each target training data in the target training dataset, based on the training user characteristics, training resource candidate set characteristics and training user behavior characteristics in the target training data, a genre resource allocation model is adopted to predict the predicted genre ratio of multiple recommended resources.

[0293] Based on the predicted and actual genre proportions of the multiple recommended resources, the parameters of the genre resource allocation model are adjusted.

[0294] Further optionally, in one embodiment of this disclosure, the training user features include the attribute features of the training user; the attribute features of the training user include at least one of the basic attribute features of the training user and the user's preference features.

[0295] Further optionally, in one embodiment of this disclosure, the training resource candidate set includes feature representations of multiple training resources; each training resource feature representation is obtained by performing vector representation and pooling operations based on at least one of the training resource title, training resource label, training resource genre, and training resource length; and also includes the average feature representation of the training resource candidate set.

[0296] Further optionally, in one embodiment of this disclosure, the training user behavior characteristics include the training user's historical consumption characteristics, the training user's satisfaction sequence characteristics, and / or the training user's resource consumption scenario characteristics;

[0297] The historical consumption characteristics of the training users include at least one of the following: the proportion of users consuming resources of various genres in multiple historical time periods before the current time period, the duration of consuming resources of various genres, and the completion rate of resources of various genres.

[0298] The training user's satisfaction sequence features include the feature representations of multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

[0299] The training device 1100 for the genre resource allocation model in the above embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0300] Figure 12 This is a schematic diagram according to the eleventh embodiment of this disclosure; as shown Figure 12 As shown, this embodiment provides a resource proportion prediction device 1200, including:

[0301] The acquisition module 1201 is used to acquire user features, resource candidate set features, and user behavior features respectively;

[0302] The prediction module 1202 is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set, using a pre-trained genre resource allocation model, based on the user characteristics, the resource candidate set characteristics, and the user behavior characteristics.

[0303] The resource proportion prediction device 1200 of this embodiment achieves the same implementation principle and technical effect as the above-mentioned related method embodiments by using the above-mentioned modules. For details, please refer to the description of the above-mentioned related method embodiments, which will not be repeated here.

[0304] Further optionally, in one embodiment of this disclosure, the acquisition module 1201 is configured to:

[0305] Obtain at least one of the user's attribute characteristics and the user's historical consumption characteristics; the user's attribute characteristics include at least one of the user's basic attribute characteristics and the user's preference characteristics; the user's historical consumption characteristics include at least one of the user's consumption ratio of various genres of resources in multiple historical time periods before the current time period and the completion rate of various genres of resources.

[0306] Further optionally, in one embodiment of this disclosure, the acquisition module 1201 is configured to:

[0307] Obtain feature representations of multiple resources in the resource candidate set; each resource's feature representation is obtained by vector representation and pooling operations based on at least one of the resource's title, resource's tag, resource's genre, and resource's length.

[0308] Based on the feature representations of multiple resources in the resource candidate set, the average feature representation of the resource candidate set is obtained.

[0309] Further optionally, in one embodiment of this disclosure, the acquisition module 1201 is configured to:

[0310] Obtain users' historical consumption characteristics, users' satisfaction sequence characteristics, and / or the scenario characteristics of users' resource consumption;

[0311] The user's historical consumption characteristics include at least one of the following: the user's consumption percentage of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres.

[0312] The user's satisfaction sequence features include the feature representations of the multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

[0313] The resource proportion prediction device 1200 of the above embodiment achieves the same implementation principle and technical effect of predicting resource proportion by using the above module as the implementation principle of the above related method embodiment. For details, please refer to the description of the above related method embodiment, which will not be repeated here.

[0314] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0315] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0316] Figure 13 A schematic block diagram of an example electronic device 1300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0317] like Figure 13 As shown, device 1300 includes a computing unit 1301, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1302 or a computer program loaded from storage unit 1308 into random access memory (RAM) 1303. The RAM 1303 may also store various programs and data required for the operation of device 1300. The computing unit 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Input / output (I / O) interface 1305 is also connected to bus 1304.

[0318] Multiple components in device 1300 are connected to I / O interface 1305, including: input unit 1306, such as keyboard, mouse, etc.; output unit 1307, such as various types of monitors, speakers, etc.; storage unit 1308, such as disk, optical disk, etc.; and communication unit 1309, such as network card, modem, wireless transceiver, etc. Communication unit 1309 allows device 1300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0319] The computing unit 1301 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1301 performs the various methods and processes described above, such as the methods described above in this disclosure. For example, in some embodiments, the methods described above in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 13011. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1300 via ROM 1302 and / or communication unit 1309. When the computer program is loaded into RAM 1303 and executed by the computing unit 1301, one or more steps of the methods described above in this disclosure can be performed. Alternatively, in other embodiments, the computing unit 1301 may be configured to perform the methods described above in this disclosure by any other suitable means (e.g., by means of firmware).

[0320] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0321] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0322] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0323] 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 CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or 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).

[0324] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0325] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0326] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0327] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A training method for a genre resource allocation model, comprising: Based on the user's historical consumption information, a training dataset is collected, which includes multiple training data points; Based on the attribute features of the training users, the features of the training resource candidate set, the behavioral features of the training users, the features of multiple recommended resources, and the consumption information of multiple recommended resources in each of the training data, a satisfaction evaluation model is trained; one piece of training data is the data during a training user's request for resource recommendation; the multiple recommended resources are multiple resources selected and recommended from the training resource candidate set during a training user's request for resource recommendation; the satisfaction evaluation model is used to predict the training user's satisfaction level with the multiple recommended resources recommended based on the training resource candidate set during a resource recommendation process; the features of the training resource candidate set include the features of the multiple training resources included in the training resource candidate set. The features of each training resource are obtained by vector representation and pooling operation based on at least one of the training resource title, training resource label, training resource genre, and training resource length. The consumption information of the multiple recommended resources is used to determine the true satisfaction level of the multiple resources by combining the preset satisfaction behavior judgment conditions. Based on the satisfaction evaluation model, select multiple target training data with the highest satisfaction from the training dataset; The genre resource allocation model is trained using the attribute features of training users, the features of training resource candidate sets, the behavioral features of training users, and the actual genre proportions of multiple recommended resources from each target training data set.

2. The method according to claim 1, wherein, Based on the attribute features of training users, the features of the candidate set of training resources, the behavioral features of training users, the features of multiple recommended resources, and the consumption information of multiple recommended resources in the training data, a satisfaction evaluation model is trained, including: Based on the attribute features of training users, the features of candidate sets of training resources, the behavioral features of training users, and the features of multiple recommended resources in the training data, the satisfaction evaluation model is used to predict the predicted satisfaction of the multiple recommended resources. Based on the consumption information of the multiple recommended resources and the preset satisfaction behavior judgment conditions, the true satisfaction of the multiple recommended resources is obtained. Based on the predicted and actual satisfaction levels of the multiple recommended resources, the parameters of the satisfaction assessment model are adjusted.

3. The method according to claim 1, wherein, The genre resource allocation model was trained using multiple sets of target training data, including: For each target training data in the target training dataset, based on the attribute features of the training users, the candidate set features of the training resources, and the behavioral features of the training users in the target training data, a genre resource allocation model is adopted to predict the predicted genre proportion of multiple recommended resources. Based on the predicted and actual genre proportions of the multiple recommended resources, the parameters of the genre resource allocation model are adjusted.

4. The method according to any one of claims 1-3, wherein, The attribute features of the training users include at least one of the basic attribute features of the training users and the user's preference features.

5. The method according to any one of claims 1-3, wherein, It also includes the average feature representation of the training resource candidate set.

6. The method according to any one of claims 1-3, wherein, The training user behavior characteristics include the training user's historical consumption characteristics, the training user's satisfaction sequence characteristics, and / or the training user's consumption resource scenario characteristics; The historical consumption characteristics of the training users include at least one of the following: the proportion of users consuming resources of various genres in multiple historical time periods before the current time period, the duration of consuming resources of various genres, and the completion rate of resources of various genres. The training user's satisfaction sequence features include the feature representations of multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

7. A method for predicting resource proportion, comprising: Obtain user attribute features, resource candidate set features, and user behavior features respectively; Based on the user attribute features, the resource candidate set features, and the user behavior features, a pre-trained genre resource allocation model is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set; the genre resource allocation model is the genre resource allocation model trained as described in any one of claims 1-6 above.

8. The method according to claim 7, wherein, Obtain user attribute characteristics, including: Obtain at least one of the user's basic attribute characteristics and the user's preference characteristics.

9. The method according to claim 7, wherein, Obtain the characteristics of the resource candidate set, including: Obtain feature representations of multiple resources in the resource candidate set; each resource's feature representation is obtained by vector representation and pooling operations based on at least one of the resource's title, resource's tag, resource's genre, and resource's length. Based on the feature representations of multiple resources in the resource candidate set, the average feature representation of the resource candidate set is obtained.

10. The method according to claim 7, wherein, Obtain user behavior characteristics, including: Obtain users' historical consumption characteristics, users' satisfaction sequence characteristics, and / or the scenario characteristics of users' resource consumption; The user's historical consumption characteristics include at least one of the following: the user's consumption percentage of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres. The user's satisfaction sequence features include the feature representations of the multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

11. A training device for a genre resource allocation model, comprising: The acquisition module is used to acquire a training dataset based on the user's historical consumption information. The training dataset includes multiple training data points. The first training module is used to train a satisfaction evaluation model based on the attribute features of the training user, the features of the training resource candidate set, the behavioral features of the training user, the features of multiple recommended resources, and the consumption information of multiple recommended resources in each training data set. One piece of training data is data from a single resource recommendation request by a training user. The multiple recommended resources are resources selected and recommended from the training resource candidate set during a single resource recommendation request by the training user. The satisfaction evaluation model is used to predict the training user's satisfaction level with the multiple recommended resources recommended based on the training resource candidate set during a single resource recommendation process. The features of the training resource candidate set include the features of the multiple training resources included in the training resource candidate set. The features of each training resource are obtained by vector representation and pooling operation based on at least one of the training resource title, training resource label, training resource genre, and training resource length. The consumption information of the multiple recommended resources is used to determine the true satisfaction level of the multiple resources by combining the preset satisfaction behavior judgment conditions. The filtering module is used to filter multiple target training data with the highest satisfaction from the training dataset based on the satisfaction evaluation model. The second training module is used to train the genre resource allocation model by using the attribute features of training users, the features of training resource candidate sets, the behavioral features of training users, and the actual genre proportions of multiple recommended resources from each target training data.

12. The apparatus according to claim 11, wherein, The first training module is used for: Based on the attribute features of training users, the features of candidate sets of training resources, the behavioral features of training users, and the features of multiple recommended resources in the training data, the satisfaction evaluation model is used to predict the predicted satisfaction of the multiple recommended resources. Based on the consumption information of the multiple recommended resources and the preset satisfaction behavior judgment conditions, the true satisfaction of the multiple recommended resources is obtained. Based on the predicted and actual satisfaction levels of the multiple recommended resources, the parameters of the satisfaction assessment model are adjusted.

13. The apparatus according to claim 11, wherein, The second training module is used for: For each target training data in the target training dataset, based on the attribute features of the training users, the candidate set features of the training resources, and the behavioral features of the training users in the target training data, a genre resource allocation model is adopted to predict the predicted genre proportion of multiple recommended resources. Based on the predicted and actual genre proportions of the multiple recommended resources, the parameters of the genre resource allocation model are adjusted.

14. The apparatus according to any one of claims 11-13, wherein, The attribute features of the training users include at least one of the basic attribute features of the training users and the user's preference features.

15. The apparatus according to any one of claims 11-13, wherein, It also includes the average feature representation of the training resource candidate set.

16. The apparatus according to any one of claims 11-13, wherein, The training user behavior characteristics include the training user's historical consumption characteristics, the training user's satisfaction sequence characteristics, and / or the training user's consumption resource scenario characteristics; The historical consumption characteristics of the training users include at least one of the following: the proportion of users consuming resources of various genres in multiple historical time periods before the current time period, the duration of consuming resources of various genres, and the completion rate of resources of various genres. The training user's satisfaction sequence features include the feature representations of multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

17. A resource proportion prediction device, comprising: The acquisition module is used to acquire user attribute features, resource candidate set features, and user behavior features, respectively. The prediction module is used to predict the proportion of various genre resources among multiple resources recommended to the user based on the resource candidate set, using a pre-trained genre resource allocation model, based on the user attribute features, the resource candidate set features, and the user behavior features; the genre resource allocation model is the genre resource allocation model trained as described in any one of claims 11-16 above.

18. The apparatus according to claim 17, wherein, The acquisition module is used for: Obtain at least one of the user's basic attribute characteristics and the user's preference characteristics.

19. The apparatus according to claim 17, wherein, The acquisition module is used for: Obtain feature representations of multiple resources in the resource candidate set; each resource's feature representation is obtained by vector representation and pooling operations based on at least one of the resource's title, resource's tag, resource's genre, and resource's length. Based on the feature representations of multiple resources in the resource candidate set, the average feature representation of the resource candidate set is obtained.

20. The apparatus according to claim 17, wherein, The acquisition module is used for: Obtain users' historical consumption characteristics, users' satisfaction sequence characteristics, and / or the scenario characteristics of users' resource consumption; The user's historical consumption characteristics include at least one of the following: the user's consumption percentage of resources of various genres in multiple historical time periods prior to the current time period, the duration of consumption of resources of various genres, and the completion rate of resources of various genres. The user's satisfaction sequence features include the feature representations of the multiple resources that the user is most satisfied with within a preset time period prior to the current time period.

21. 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 executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-6 or 7-10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6 or 7-10.

23. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-6 or 7-10.

Images