Multimedia Resource Recommendation Method, Device, Server, Storage Medium and Product

Abstract

The present disclosure relates to a method, apparatus, server, storage medium and product for recommending multimedia resources, and relates to the field of Internet technologies. An embodiment of the present disclosure provides a method for recommending multimedia resources. Based on the chronological order, the first resource types of the historically recommended multimedia resources and the second resource types of the recently recommended multimedia resources are combined to form a first type sequence. Since the resource types in the first type sequence are arranged in chronological order, and the arrangement order of these resource types can reflect the change in the user's recent interest in resource types, therefore, this method can predict the resource types that the user is interested in, and then recommend the multimedia resources corresponding to the resource types that the user is interested in, thereby improving the recommendation effect of the multimedia resources.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technologies, and in particular, to a method, apparatus, server, storage medium, and product for recommending multimedia resources. Background Art

[0002] Currently, with the continuous development of Internet technologies, short-video applications have emerged as the times require. Short-video applications include various types of short videos. Short-video applications can recommend different types of short videos to users, thereby satisfying users' short-video viewing experiences. And how to recommend short videos to users has become a research hotspot in short-video recommendation.

[0003] In related technologies, when recommending short videos to users, mainly first obtain the short videos browsed by the user within the historical time range, then determine the video types of these short videos, and then determine the number of views of each video type. The more the number of views, the more interested the user is in this video type, and then recommend short videos of this video type to the user.

[0004] However, according to the method in related technologies, if the user browses a certain video type more frequently, short videos of this video type will be recommended to the user. As a result, over time, the video types recommended to the user will become more convergent. In the long run, users will feel tired, resulting in poor short-video recommendation effects. Summary of the Invention

[0005] The present disclosure provides a method, apparatus, server, storage medium, and product for recommending multimedia resources, which can improve the multimedia resource recommendation effect. The technical solution of the present disclosure is as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, there is provided a method for recommending multimedia resources, the method including:

[0007] Obtain a plurality of first multimedia resources and a plurality of second multimedia resources, where the plurality of first multimedia resources are multimedia resources recommended for a target account within a first historical time range, and the plurality of second multimedia resources are multimedia resources recommended for the target account within a second historical time range, and the start time of the second historical time range is not earlier than the end time of the first historical time range;

[0008] Determine a plurality of first resource types corresponding to the plurality of first multimedia resources and a plurality of second resource types corresponding to the plurality of second multimedia resources;

[0009] Based on the time sequence, determine a first type sequence composed of the plurality of first resource types and the plurality of second resource types;

[0010] Recommend multimedia resources to the target account based on the association relationship between each resource type and time in the first type of sequence.

[0011] In some embodiments, the step of recommending multimedia resources to the target account based on the association relationship between each resource type and time in the first type of sequence includes:

[0012] Determine the label type of each resource type in the first type of sequence, where the label type is used to represent the degree of interest of the target account in the resource type;

[0013] Based on the association relationship between each resource type and time in the first type of sequence, determine the association relationship between the label type of each resource type and time;

[0014] Recommend multimedia resources to the target account based on the association relationship between the label type of each resource type and time.

[0015] In some embodiments, the step of determining the label type of each resource type in the first type of sequence includes:

[0016] If the multimedia resource corresponding to the resource type meets the first preset condition, determine that the label type of the resource type is the first label;

[0017] If the multimedia resource corresponding to the resource type meets the second preset condition, determine that the resource label of the resource type is the second label, where the first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

[0018] In some embodiments, the step of recommending multimedia resources to the target account based on the association relationship between each resource type and time in the first type of sequence includes:

[0019] Input each resource type in the first type of sequence into a resource type prediction model in chronological order to obtain a predicted resource type, where the resource type prediction model is used to predict the resource type that the target account is interested in;

[0020] Obtain the multimedia resources corresponding to the predicted resource type and recommend the multimedia resources to the target account.

[0021] In some embodiments, the process of determining the multiple second resource types corresponding to the multiple second multimedia resources includes:

[0022] Divide the multiple second multimedia resources into n multimedia resource sets in chronological order with m as a unit, where each multimedia resource set includes m second multimedia resources, and m is an integer greater than zero;

[0023] For each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, and obtain the multiple second resource types.

[0024] In some embodiments, the process of determining the multiple second resource types corresponding to the multiple second multimedia resources includes:

[0025] Divide the second historical time range into multiple third historical time ranges in units of a preset duration;

[0026] Determine multiple third multimedia resources corresponding to each third historical time range in chronological order;

[0027] For each third historical time range, divide the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets in units of m, where p is an integer greater than zero;

[0028] For each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, and obtain the multiple second resource types.

[0029] In some embodiments, the determining, in chronological order, a first type sequence composed of the multiple first resource types and the multiple second resource types includes:

[0030] Sort the multiple first resource types based on the chronological order of the multiple first resource types to obtain a second type sequence;

[0031] Add the second resource types different from the multiple first resource types among the multiple second resource types to the second type sequence based on the chronological order of the multiple second resource types to obtain the first type sequence.

[0032] According to a second aspect of the embodiments of the present disclosure, there is provided a multimedia resource recommendation method, the method including:

[0033] Obtain multiple first sample multimedia resources and multiple second sample multimedia resources, where the multiple first sample multimedia resources are multimedia resources recommended for a sample account within a first sample historical time range, the multiple second sample multimedia resources are multimedia resources recommended for the sample account within a second sample historical time range, and the start time of the second sample historical time range is not earlier than the end time of the first sample historical time range;

[0034] Determine multiple first sample resource types corresponding to the multiple first sample multimedia resources and multiple second sample resource types corresponding to the multiple second sample multimedia resources;

[0035] Determine a first sample type sequence composed of the multiple first sample resource types and the multiple second sample resource types based on the chronological order.

[0036] Based on the association relationship between each sample resource type in the first sample type sequence and time, perform model training to obtain a resource type prediction model, which is used to predict the resource types that the target account is interested in, so as to recommend multimedia resources corresponding to the predicted resource types to the target account.

[0037] In some embodiments, the performing model training based on the association relationship between each sample resource type in the first sample type sequence and time to obtain a resource type prediction model includes:

[0038] Taking s and z as units respectively, based on the chronological order of each sample resource type in the first sample type sequence, obtain s sample resource types and z sample resource types from the first sample type sequence, where the z sample resource types are the sample resource types after the s sample resource types.

[0039] Based on the s sample resource types and the z sample resource types, perform model training for the current iteration.

[0040] Re-obtain s sample resource types and z sample resource types from the first sample type sequence. The re-obtained s sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the s sample resource types in the current iteration, and the re-obtained z sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the z sample resource types in the current iteration. s, z, and y are all integers greater than zero.

[0041] Based on the re-obtained s sample resource types and the re-obtained z sample resource types, perform model training for the next iteration until the convergence condition is met to obtain the resource type prediction model.

[0042] In some embodiments, the performing model training for the current iteration based on the s sample resource types and the z sample resource types includes:

[0043] Obtain the sample label types of the z sample resource types.

[0044] Based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types, perform model training for the current iteration.

[0045] According to a third aspect of the embodiments of the present disclosure, there is provided a multimedia resource recommendation device, which includes:

[0046] A first acquisition unit configured to acquire a plurality of first multimedia resources and a plurality of second multimedia resources, where the plurality of first multimedia resources are multimedia resources recommended for a target account within a first historical time range, and the plurality of second multimedia resources are multimedia resources recommended for the target account within a second historical time range, and the start time of the second historical time range is not earlier than the end time of the first historical time range;

[0047] A first determination unit configured to determine a plurality of first resource types corresponding to the plurality of first multimedia resources and a plurality of second resource types corresponding to the plurality of second multimedia resources;

[0048] A second determination unit configured to determine a first type sequence composed of the plurality of first resource types and the plurality of second resource types based on the time sequence;

[0049] A recommendation unit configured to recommend multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time.

[0050] In some embodiments, the recommendation unit is configured to determine the label type of each resource type in the first type sequence, where the label type is used to represent the degree of interest of the target account in the resource type; determine the association relationship between the label type of each resource type and time based on the association relationship between each resource type in the first type sequence and time; and recommend multimedia resources to the target account based on the association relationship between the label type of each resource type and time.

[0051] In some embodiments, the recommendation unit is configured to determine that the label type of the resource type is a first label if the multimedia resource corresponding to the resource type meets a first preset condition; and determine that the resource label of the resource type is a second label if the multimedia resource corresponding to the resource type meets a second preset condition, where the first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

[0052] In some embodiments, the recommendation unit is configured to input each resource type in the first type sequence into a resource type prediction model based on the time sequence to obtain a predicted resource type, where the resource type prediction model is used to predict the resource type that the target account is interested in; acquire the multimedia resource corresponding to the predicted resource type, and recommend the multimedia resource to the target account.

[0053] In some embodiments, the first determination unit is configured to divide the multiple second multimedia resources into n multimedia resource sets based on the time sequence in units of m, where each multimedia resource set includes m second multimedia resources, and m is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, so as to obtain the multiple second resource types.

[0054] In some embodiments, the first determination unit is configured to divide the second historical time range into multiple third historical time ranges in units of a preset duration; determine multiple third multimedia resources corresponding to each third historical time range according to the time sequence; for each third historical time range, divide the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets in units of m, where p is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, so as to obtain the multiple second resource types.

[0055] In some embodiments, the second determination unit is configured to sort the multiple first resource types based on the time sequence of the multiple first resource types to obtain a second type sequence; add the second resource types different from the multiple first resource types among the multiple second resource types to the second type sequence based on the time sequence of the multiple second resource types, so as to obtain the first type sequence.

[0056] According to a fourth aspect of the embodiments of the present disclosure, there is provided a multimedia resource recommendation device, where the device includes:

[0057] A second acquisition unit configured to acquire multiple first sample multimedia resources and multiple second sample multimedia resources, where the multiple first sample multimedia resources are multimedia resources recommended for a sample account within a first sample historical time range, and the multiple second sample multimedia resources are multimedia resources recommended for the sample account within a second sample historical time range, and the start time of the second sample historical time range is not earlier than the end time of the first sample historical time range;

[0058] A third determination unit configured to determine multiple first sample resource types corresponding to the multiple first sample multimedia resources and multiple second sample resource types corresponding to the multiple second sample multimedia resources;

[0059] A fourth determination unit configured to determine a first sample type sequence composed of the multiple first sample resource types and the multiple second sample resource types based on the time sequence;

[0060] A model training unit, configured to perform model training based on the association relationship between each sample resource type and time in the first sample type sequence, to obtain a resource type prediction model, where the resource type prediction model is used to predict the resource types that a target account is interested in, so as to recommend multimedia resources corresponding to the predicted resource types to the target account.

[0061] In some embodiments, the model training unit is configured to, in units of s and z respectively, based on the time sequence of each sample resource type in the first sample type sequence, obtain s sample resource types and z sample resource types from the first sample type sequence, where the z sample resource types are the sample resource types after the s sample resource types; perform model training for the current iteration based on the s sample resource types and the z sample resource types; re-obtain s sample resource types and z sample resource types from the first sample type sequence, where the re-obtained s sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the s sample resource types in the current iteration, and the re-obtained z sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the z sample resource types in the current iteration, and s, z, and y are all integers greater than zero; perform model training for the next iteration based on the re-obtained s sample resource types and the re-obtained z sample resource types until a convergence condition is satisfied, to obtain the resource type prediction model.

[0062] In some embodiments, the model training unit is configured to obtain the sample label types of the z sample resource types; perform model training for the current iteration based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types.

[0063] According to a fifth aspect of the embodiments of the present disclosure, there is provided a server, where the server includes:

[0064] A processor;

[0065] A memory for storing executable instructions of the processor;

[0066] Wherein, the processor is configured to execute the instructions to implement the multimedia resource recommendation method described in the first aspect or the second aspect above.

[0067] According to a sixth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by a processor of a server, enabling the server to execute the multimedia resource recommendation method described in the first aspect or the second aspect above.

[0068] According to a seventh aspect of the embodiments of the present disclosure, there is provided a computer program product, including a computer program which, when executed by a processor, implements the multimedia resource recommendation method described in the first aspect or the second aspect above.

[0069] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0070] The embodiments of the present disclosure provide a multimedia resource recommendation method. Based on the chronological order, the first resource types of the historically recommended multimedia resources and the second resource types of the recently recommended multimedia resources are combined to form a first type sequence. Since the resource types in the first type sequence are arranged in chronological order, and the arrangement order of these resource types can reflect the change in the user's recent interest in resource types, therefore, this method can predict the resource types that the user is interested in, and then recommend the multimedia resources corresponding to the resource types that the user is interested in, thereby improving the recommendation effect of the multimedia resources.

[0071] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure, and do not constitute an improper limitation of the present disclosure.

[0073] Figure 1 is a schematic diagram of an implementation environment of a multimedia resource recommendation method shown according to an exemplary embodiment.

[0074] Figure 2 is a flowchart of a multimedia resource recommendation method shown according to an exemplary embodiment.

[0075] Figure 3 is a flowchart of a multimedia resource recommendation method shown according to an exemplary embodiment.

[0076] Figure 4 is a schematic diagram of determining a first type sequence shown according to an exemplary embodiment.

[0077] Figure 5 is a schematic diagram of determining a first type sequence shown according to an exemplary embodiment.

[0078] Figure 6 is a flowchart of training a resource type prediction model shown according to an exemplary embodiment.

[0079] Figure 7It is a schematic diagram showing a method for determining a first sample type sequence according to an exemplary embodiment.

[0080] Figure 8 It is a schematic diagram showing a prediction based on a first sample type sequence according to an exemplary embodiment.

[0081] Figure 9 It is a schematic diagram showing model training by a DQN algorithm according to an exemplary embodiment.

[0082] Figure 10 It is a schematic diagram showing a training resource type prediction model and a prediction by the resource type prediction model according to an exemplary embodiment.

[0083] Figure 11 It is a block diagram of a multimedia resource recommendation device according to an exemplary embodiment.

[0084] Figure 12 It is a block diagram of a multimedia resource recommendation device according to an exemplary embodiment.

[0085] Figure 13 It is a block diagram of the structure of a server according to an exemplary embodiment. Detailed implementation manners

[0086] To enable those of ordinary skill in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0087] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data used may be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0088] The user information involved in the present disclosure may be information authorized by the user or fully authorized by all parties.

[0089] Figure 1 It is an implementation environment diagram of a multimedia resource recommendation method according to an exemplary embodiment. Refer to Figure 1 , this implementation environment includes: a terminal 101 and a server 102. The terminal 101 is connected to the server 102 through a wireless or wired network.

[0090] The target application is installed on the terminal 101, and the server 102 is the background server 102 of the target application. The user logs in to the target application through an account, and the server 102 of the target application can recommend multimedia resources to the user. The multimedia resources can be videos, short videos, pictures, articles, etc., and no specific limitation is made on the multimedia resources here.

[0091] In the embodiments of the present disclosure, the multimedia resource recommendation method can be applied to various scenarios. If the multimedia resource is a short video, the method can be applied to the short video recommendation scenario. In this scenario, the server 102 can predict the type of short video that the user is interested in according to the user's recent browsing situation of short videos, and then recommend short videos of this type to the user. For example, if the server 102 predicts, according to the user's recent browsing situation of short videos, that the type of short video that the user is interested in is sports, then short videos of the sports type are recommended to the user; if the server 102 predicts, according to the user's recent browsing situation of short videos, that the type of short video that the user is interested in is beauty makeup, then short videos of the beauty makeup type are recommended to the user.

[0092] If the multimedia resource is a picture, the method can be applied to the picture recommendation scenario. In this scenario, the server 102 can predict the type of picture that the user is interested in according to the user's recent browsing situation of pictures, and then recommend pictures of this type to the user. If the multimedia resource is an article, the method can be applied to the article recommendation scenario. In this scenario, the server 102 can predict the type of article that the user is interested in according to the user's recent browsing situation of articles, and then recommend articles of this type to the user.

[0093] The terminal 101 is at least one of devices such as mobile phones, tablet computers, and PC (Personal Computer) devices. The server 102 can be at least one of a single server, a server cluster composed of multiple servers, a cloud server, a cloud computing platform, and a virtualization center.

[0094] Figure 2 It is a flowchart of a multimedia resource recommendation method shown according to an exemplary embodiment. As Figure 2 shown, the method includes:

[0095] 201. Obtain a plurality of first multimedia resources and a plurality of second multimedia resources.

[0096] The plurality of first multimedia resources are the multimedia resources recommended for the target account within the first historical time range, and the plurality of second multimedia resources are the multimedia resources recommended for the target account within the second historical time range. The start time of the second historical time range is not earlier than the end time of the first historical time range.

[0097] 202. Determine multiple first resource types corresponding to multiple first multimedia resources and multiple second resource types corresponding to multiple second multimedia resources.

[0098] 203. Based on the time sequence, determine a first type sequence composed of multiple first resource types and multiple second resource types.

[0099] 204. Recommend multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time.

[0100] In some embodiments, recommending multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time includes:

[0101] Determine the label type of each resource type in the first type sequence, where the label type is used to represent the degree of interest of the target account in the resource type;

[0102] Based on the association relationship between each resource type in the first type sequence and time, determine the association relationship between the label type of each resource type and time;

[0103] Recommend multimedia resources to the target account based on the association relationship between the label type of each resource type and time.

[0104] In some embodiments, determining the label type of each resource type in the first type sequence includes:

[0105] If the multimedia resource corresponding to the resource type meets the first preset condition, determine the label type of the resource type as the first label;

[0106] If the multimedia resource corresponding to the resource type meets the second preset condition, determine the resource label of the resource type as the second label, where the first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

[0107] In some embodiments, recommending multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time includes:

[0108] Based on the time sequence, input each resource type in the first type sequence into a resource type prediction model to obtain a predicted resource type, where the resource type prediction model is used to predict the resource types that the target account is interested in;

[0109] Obtain the multimedia resources corresponding to the predicted resource type and recommend the multimedia resources to the target account.

[0110] In some embodiments, the process of determining multiple second resource types corresponding to multiple second multimedia resources includes:

[0111] In units of m, based on the chronological order, multiple second multimedia resources are divided into n multimedia resource sets, and each multimedia resource set includes m second multimedia resources, where m is an integer greater than zero;

[0112] For each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, and obtain multiple second resource types.

[0113] In some embodiments, the process of determining multiple second resource types corresponding to multiple second multimedia resources includes:

[0114] Divide the second historical time range into multiple third historical time ranges in units of a preset duration;

[0115] Determine multiple third multimedia resources corresponding to each third historical time range in chronological order;

[0116] For each third historical time range, in units of m, based on the chronological order, divide the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets, where p is an integer greater than zero;

[0117] For each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, and obtain multiple second resource types.

[0118] In some embodiments, based on the chronological order, determining a first type sequence composed of multiple first resource types and multiple second resource types includes:

[0119] Sort the multiple first resource types based on the chronological order of the multiple first resource types to obtain a second type sequence;

[0120] Add the second resource types different from the multiple first resource types among the multiple second resource types to the second type sequence based on the chronological order of the multiple second resource types to obtain a first type sequence.

[0121] In the embodiments of the present disclosure, a first type sequence is composed of the first resource types of the historically recommended multimedia resources and the second resource types of the recently recommended multimedia resources based on the chronological order. Since the resource types in the first type sequence are arranged in chronological order, and the arrangement order of these resource types can reflect the recent change in the user's interest in resource types, therefore, this method can predict the resource types that the user is interested in, and then recommend the multimedia resources corresponding to the resource types that the user is interested in, thereby improving the recommendation effect of the multimedia resources.

[0122] Figure 3It is a flowchart of a multimedia resource recommendation method shown according to an exemplary embodiment. As Figure 3 shown, it is executed by a server. The method includes:

[0123] 301. The server obtains a plurality of first multimedia resources and a plurality of second multimedia resources.

[0124] The plurality of first multimedia resources are multimedia resources recommended for a target account within a first historical time range. The plurality of second multimedia resources are multimedia resources recommended for the target account within a second historical time range. The start time of the second historical time range is not earlier than the end time of the first historical time range. It can be seen from this that although both the first multimedia resources and the second multimedia resources are multimedia resources recommended for the target account in history, compared with the first multimedia resources, the second multimedia resources are multimedia resources recommended for the target account recently. Among them, the multimedia resource can be a video, a short video, a picture, an article, etc. There is no specific limitation on the multimedia resource here.

[0125] In a possible implementation manner, a target application program is installed on a terminal, and the target account can log in to the target application program through the terminal. When the terminal detects a login operation of the target account logging in to the target application program, it sends a recommendation request to the server. The recommendation request carries the account identifier, a first time identifier, and a second time identifier of the target account. Among them, the first time identifier is used to obtain the first multimedia resources within the first historical time range before the current time, and the second time identifier is used to obtain the second multimedia resources within the second historical time range before the current time.

[0126] In another possible implementation manner, when the terminal detects a refresh operation in the case where the target account logs in to the target application program, it sends a recommendation request to the server. The recommendation request carries the account identifier, a first time identifier, and a second time identifier of the target account.

[0127] The server receives the recommendation request and obtains the first multimedia resources and the second multimedia resources from the multimedia resource library based on the account identifier, the first time identifier, and the second time identifier. The multimedia resource library stores the account identifiers of multiple accounts and the multimedia resources corresponding to each moment. The server obtains the first multimedia resources within the first historical time range and the second multimedia resources within the second historical time range corresponding to the account identifier from the multimedia resource library based on the account identifier, the first time identifier, and the second time identifier.

[0128] Among them, both the first historical time range and the second historical time range can be set and changed as needed, and no specific limitation is made thereto. For example, the second historical time range is one week before the current time, and the first historical time range is one month before that one week. Another example is that the second historical time range is two weeks before the current time, and the first historical time range is one month before those two weeks.

[0129] It should be noted that the multiple first multimedia resources and the multiple second multimedia resources recommended for the target account are the multimedia resources that appear when the target account browses the target application, that is, the exposed multimedia resources.

[0130] 302. The server determines multiple first resource types corresponding to the multiple first multimedia resources and multiple second resource types corresponding to the multiple second multimedia resources.

[0131] In this step, the server can determine the multiple second resource types through any of the following implementation manners.

[0132] The first implementation manner: The server determines the second resource type corresponding to each second multimedia resource one by one, and then removes the duplicate second resource types based on the time sequence to obtain multiple second resource types.

[0133] In this implementation manner, the server can sort the multiple second multimedia resources based on the time sequence of the multiple second multimedia resources to obtain a first queue. Among them, the server can sort the multiple second multimedia resources in the time sequence from the earliest to the latest, which can reflect the sequence of appearance of the multiple second multimedia resources, and then determine the second resource type corresponding to each second multimedia resource one by one, and finally obtain multiple second resource types.

[0134] The second implementation manner: The server divides the multiple second multimedia resources into n multimedia resource sets based on the time sequence in units of m, and each multimedia resource set includes m second multimedia resources; for each multimedia resource set, it determines at least one second resource type corresponding to the m second multimedia resources to obtain multiple second resource types.

[0135] In this implementation manner, the server can also first sort the multiple second multimedia resources to obtain a first queue. Then it determines the quantity of the multiple second multimedia resources, and then determines the ratio of this quantity to m, and divides based on this ratio. If the ratio is an integer, then n is this ratio; if the ratio is not an integer and there is a remainder, then the second multimedia resources corresponding to the remainder are taken as a separate multimedia resource set, that is, n is this ratio plus 1.

[0136] In the embodiments of the present disclosure, m is an integer greater than zero, and m can be set and changed as needed, and no specific limitation is made thereto. For example, m is 10, 30, or 50.

[0137] See Figure 4 , the number of multiple second multimedia resources is t*m. The server can sequentially determine at least one second resource type corresponding to every m second multimedia resources based on the time sequence, and finally obtain multiple second resource types. Since there may be the same second resource type among these m second multimedia resources, the number of second resource types is less than or equal to m.

[0138] In the embodiments of the present disclosure, when there are many recently appeared multimedia resources, these multimedia resources may include multimedia resources with the same resource type. In this case, they can be divided in units of m, and then the resource type corresponding to every m multimedia resources is determined, so that the resource type of these multimedia resources can be determined faster, thereby improving the efficiency of determining the resource type.

[0139] In the third implementation manner, the server divides the second historical time range into multiple third historical time ranges in units of a preset time period; determines multiple third multimedia resources corresponding to each third historical time range in chronological order; for each third historical time range, divides the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets in units of m based on the time sequence; and determines at least one second resource type corresponding to m second multimedia resources for each multimedia resource set to obtain multiple second resource types. Herein, p is an integer greater than zero.

[0140] In this implementation manner, the second historical time range is a time period, and the number of second multimedia resources in this time period may be large. Therefore, this time period can be divided into multiple sub-time periods, that is, the second historical time range is divided into multiple third historical time ranges. For each third historical time range, if the number of second multimedia resources in this third historical time range is also large, it can be divided more finely in units of m.

[0141] In the embodiments of the present disclosure, the preset time period can be set and changed as needed, and no specific limitation is made thereto. For example, the second historical time range is one week, and the preset time period is one day. Another example is that the second historical time range is one month, and the preset time period is one week. See Figure 5 , the server divides the second historical time range in units of days, and for the second multimedia resources within one day, divides them more finely in units of m.

[0142] In the embodiments of the present disclosure, when there are many recently appeared multimedia resources, these multimedia resources can also be divided in units of a preset duration first, and then within the divided time range, they can be more finely divided in units of m. In this way, the resource types of these multimedia resources can be determined more quickly, thereby improving the efficiency of determining the resource types.

[0143] In the embodiments of the present disclosure, the server can also determine multiple first resource types through any of the above implementation manners, and the manner for the server to determine multiple first resource types will not be elaborated here.

[0144] It should be noted that the server can first determine multiple first resource types and then determine multiple second resource types, or first determine multiple second resource types and then determine multiple first resource types, or determine multiple first resource types and multiple second resource types simultaneously. In the embodiments of the present disclosure, no specific limitation is made thereto.

[0145] 303. The server determines a first type sequence composed of multiple first resource types and multiple second resource types based on the time sequence.

[0146] This step can be implemented through the following steps (1) to (2) and includes:

[0147] (1) The server sorts multiple first resource types based on the time sequence of the multiple first resource types to obtain a second type sequence.

[0148] The server can sort multiple first resource types in the time sequence from the earliest to the latest to obtain a second type sequence, which can reflect the order in which multiple first resource types appear. Subsequently, it is convenient to predict the resource types that the user is recently interested in according to the order.

[0149] (2) The server adds the second resource types that are different from the multiple first resource types among the multiple second resource types to the second type sequence based on the time sequence of the multiple second resource types to obtain a first type sequence.

[0150] In a possible implementation manner, if the server determines multiple second resource types in the first implementation manner in step 302, then in step (2), the server directly compares each of the multiple second resource types with the multiple first resource types in the second type sequence based on the time sequence of the multiple second resource types, and then determines the second resource types that are different from the multiple first resource types, adds the second resource types to the second type sequence, and updates the second type sequence to obtain a first type sequence.

[0151] In another possible implementation, if the server determines multiple second resource types in the second implementation manner in step 302, then in step (2), the server directly compares at least one second resource type corresponding to every m second multimedia resources with multiple first resource types k in the second type sequence, and finally obtains the first type sequence. Continue to refer to Figure 4 。

[0152] In another possible implementation, if the server determines multiple second resource types in the third implementation manner in step 302, then in step (2), the server directly compares at least one second resource type corresponding to every m second multimedia resources within each third historical time range with multiple first resource types in the second type sequence, and finally obtains the first type sequence.

[0153] Here, only the case where the server determines multiple second resource types in the third implementation manner in step 302 is taken as an example for illustration. For example, the second historical time range is one week before the current time, the preset duration is one day, and the first historical time range is one month before this one week. Then the server divides the second multimedia resources within one week before the current time into second multimedia resources corresponding to one day, and obtains 7 multimedia resource sets. Then for each multimedia resource set, it determines at least one second resource type corresponding to every m second multimedia resources within the set, compares the at least one second resource type with 300 first resource types in the second type sequence, adds the second resource types different from the multiple first resource types to the second type sequence, then determines at least one second resource type corresponding to the next m second multimedia resources, until it traverses the multimedia resource set, and then determines at least one second resource type corresponding to every m second multimedia resources in the next multimedia resource set according to the above method, and compares it with the resource types in the second type sequence, and so on, until it traverses multiple second resource types and obtains the first type sequence. Continue to refer to Figure 5 。

[0154] In the embodiments of the present disclosure, the resource types in the first type sequence are all different resource types, and the order of arrangement of these resource types can reflect the recent interest changes of the user. Therefore, according to the association relationship between each resource type in the first type sequence and time, the recent interests or preferences of the user are determined, so as to predict the resource types that the user is interested in, and then accurately recommend corresponding multimedia resources to the user.

[0155] 304. The server inputs each resource type in the first type sequence into the resource type prediction model based on the time sequence, and obtains the predicted resource type.

[0156] The resource type prediction model is used to predict the resource types that the target account is interested in. The determination process of the resource type prediction model will be introduced in detail in the next embodiment and will not be introduced here for the time being.

[0157] In this step, the server can directly input each resource type in the first type sequence into the resource type prediction model based on the time sequence to obtain the predicted resource types. The number of the predicted resource types can be one or more, and no specific limitation is made thereto.

[0158] In the embodiments of the present disclosure, the resource type prediction model is used to predict the resource types that the user is interested in, and then multimedia resources corresponding to the resource types that the user is interested in are recommended to the user, so as to improve the recommendation effect of the multimedia resources.

[0159] In this step, the resource type prediction model can also predict the label type of the resource type that the target account is interested in, and the label type is used to represent the degree of interest of the target account in the resource type. Correspondingly, this process is implemented through the following steps (1) to (3), including:

[0160] (1) The server determines the label type of each resource type in the first type sequence.

[0161] For each resource type, if the multimedia resource corresponding to the resource type meets the first preset condition, it is determined that the label type of the resource type is the first label; if the multimedia resource corresponding to the resource type meets the second preset condition, it is determined that the label type of the resource type is the second label. Among them, the first label indicates that the degree of interest of the target account in the resource type is higher than that indicated by the second label for the degree of interest of the target account in the resource type.

[0162] In this step, if the multimedia resource corresponding to the resource type meets the first preset condition, it means that the user has actually browsed the multimedia resource or performed a preset operation on the multimedia resource. For example, the multimedia resource is a short video, and the first preset condition is that the duration of the user watching the short video exceeds the preset duration, or the user has watched the short video, or the user likes the short video, or the user comments on the short video, or the user both likes and comments on the short video, or the user downloads the short video, or the user favorites the short video, or the user shares the short video, or the user follows the short video. If the short video meets any of the above, it means that the user is more interested in the short video of this resource type, and then short videos of this resource type will be recommended to the user later. In order to distinguish from the preset duration in step 302, the preset duration in step 302 is called the first preset duration here, and the preset duration in this step is called the second preset duration.

[0163] If the multimedia resource corresponding to the resource type meets the second preset condition, it indicates that the user has browsed the multimedia resource for a short time or has not browsed the multimedia resource. For example, if the multimedia resource is a short video and the second preset condition is that the duration for which the user watches the short video does not exceed the second preset duration, when the user browses the short video for a short time, it indicates that the user may be interested in short videos of this resource type, but due to external reasons, the user browses the short videos of this resource type for a short time. Therefore, when making subsequent recommendations to the user, short videos of this resource type will also be recommended accordingly.

[0164] In the embodiments of the present disclosure, by determining the conditions met by the multimedia resource, the label type corresponding to the resource type of the multimedia resource is determined. Different label types indicate different degrees of interest of the user in the resource type. In this way, multimedia resources can be recommended to the user according to the degree of interest of the user in each resource type, thereby improving the recommendation effect of the multimedia resources.

[0165] (2) The server determines the correlation between the label type of each resource type and time based on the correlation between each resource type in the first type sequence and time.

[0166] After the server obtains the label type of each resource type, based on the time order of each resource type from first to last, the time order of each label type from first to last is obtained, that is, the correlation between the label type of each resource type and time.

[0167] (3) The server inputs each resource type and its corresponding label type in the first type sequence into the resource type prediction model based on the time order to obtain the predicted resource type and the predicted label type.

[0168] In the embodiments of the present disclosure, the resource type prediction model can not only predict the resource type that the user is interested in, but also predict the degree of interest of the user in the resource type. In this way, multimedia resources can be recommended to the user according to the degree of interest of the user in the resource type.

[0169] Since the user is more interested in the resource type of the first label, therefore, in subsequent recommendations, the multimedia resources corresponding to the resource type of the first label are preferentially recommended, and then the multimedia resources corresponding to the resource type of the second label are recommended. However, multimedia resources of both these label types will be recommended. In this way, the resource types of the recommended multimedia resources can be diversified, broadening the user's interests, thereby improving the user experience.

[0170] In the embodiments of the present disclosure, the label types of each resource type are first determined. Since the label types can represent the degree of interest of the user in the resource type, when recommending multimedia resources to the user based on the label types of the resource type, the multimedia resources corresponding to the resource type that the user is interested in can be accurately recommended, the user's recent preferences can be grasped, and the multimedia resources that the user hopes to see recently can be recommended to the user, so as to avoid the user from getting tired of browsing multimedia resources of the same resource type for a long time, thereby improving the recommendation effect of multimedia resources.

[0171] In the embodiments of the present disclosure, starting from the user's most fine-grained data, the model more accurately restores the true physical experience of the user using the product. At the same time, by expanding the observation window, the time series is extended to the past and the present. After the first type of sequence is generated, the resource type prediction model is combined to make an accurate prediction of the user's future interest changes.

[0172] 305. The server obtains the multimedia resources corresponding to the predicted resource type and recommends the multimedia resources to the target account.

[0173] If in step 304 the server only predicts the resource type through the resource type prediction model, then in this step the server directly obtains the multimedia resources corresponding to the predicted resource type. If there are multiple predicted resource types, the server obtains multiple types of multimedia resources. Among them, the quantity of each type of multimedia resource obtained by the server can be set and changed as needed, and the quantity of each type of multimedia resource can be the same or different.

[0174] If in step 304 the server predicts the resource type and the label type through the resource type prediction model, then in this step the server can obtain the multimedia resources corresponding to the predicted resource type according to the label type. If there are multiple predicted resource types, for the predicted resource type with the label type of the first label, the server can obtain the first preset quantity of multimedia resources corresponding to the predicted resource type, and for the predicted resource type with the label type of the second label, the server can obtain the second preset quantity of multimedia resources corresponding to the predicted resource type. Among them, the first preset quantity is greater than the second preset quantity, that is, compared with the predicted resource type with the second label, the server obtains a larger quantity of multimedia resources corresponding to the predicted resource type with the first label, which is more in line with the change of the user's interest in multimedia resources, and at the same time explores more potential consumption interests of the user and improves the user's product experience.

[0175] After the server obtains the multimedia resources, it sends the multimedia resources to the terminal. The terminal receives the multimedia resources and displays the multimedia resources.

[0176] In a possible implementation manner, when the terminal displays the multimedia resources, it sequentially displays the multimedia resources corresponding to each predicted resource type.

[0177] For example, the terminal displays multiple multimedia resources corresponding to one predicted resource type, and then displays multiple multimedia resources corresponding to another predicted resource type, and so on.

[0178] In another possible implementation, when the terminal displays multimedia resources, it cross-displays multimedia resources corresponding to different predicted resource types.

[0179] For example, the terminal first displays one multimedia resource corresponding to one predicted resource type, then displays one multimedia resource corresponding to another predicted resource type, then displays another multimedia resource corresponding to the one predicted resource type, and then displays another multimedia resource corresponding to the another predicted resource type, and so on for cross-display.

[0180] In the embodiments of the present disclosure, the terminal can also display multimedia resources in other ways, which are not specifically limited herein.

[0181] It should be noted that steps 301 to 305 can also be executed by a terminal. A client of a target application is installed on the terminal. The client is a client for recommending multimedia resources. After determining the predicted resource type, the client can obtain multimedia resources corresponding to the predicted resource type from the server corresponding to the client, and then the client recommends the multimedia resources to the target account.

[0182] The embodiments of the present disclosure provide a method for recommending multimedia resources. Based on the time sequence, the first resource type of the historically recommended multimedia resources and the second resource type of the recently recommended multimedia resources are combined into a first type sequence. Since the resource types in the first type sequence are arranged in time sequence, and the arrangement order of these resource types can reflect the recent change of the user's interest in resource types, therefore, this method can predict the resource types that the user is interested in, and then recommend multimedia resources corresponding to the resource types that the user is interested in, thereby improving the recommendation effect of multimedia resources.

[0183] Figure 6 is a flowchart of a method for training a resource type prediction model shown according to an exemplary embodiment, which is executed by a server, as Figure 6 shown, the method includes:

[0184] 601. The server obtains a plurality of first sample multimedia resources and a plurality of second sample multimedia resources.

[0185] The multiple first-sample multimedia resources are the multimedia resources recommended for the sample account within the first-sample historical time range. The multiple second-sample multimedia resources are the multimedia resources recommended for the sample account within the second-sample historical time range. The starting time of the second-sample historical time range is not earlier than the ending time of the first-sample historical time range.

[0186] In the embodiments of the present disclosure, the sample account includes multiple accounts. When the server obtains the first-sample multimedia resources and the second-sample multimedia resources, it can obtain the first-sample multimedia resources and the second-sample multimedia resources corresponding to each account.

[0187] The server can randomly obtain multiple accounts and use these multiple accounts as the sample account. The server can also select multiple accounts with a relatively large number of browsed multimedia resources as the sample account. If the server randomly obtains multiple accounts, for any account among the multiple accounts, if the number of the first-sample multimedia resources or the second-sample multimedia resources corresponding to the account is less than the third preset quantity, the account is discarded, and then the first-sample multimedia resources and the second-sample multimedia resources corresponding to another account are obtained. Among them, the number of the multiple accounts can be set and changed as needed, and no specific limitation is made thereto.

[0188] Step 601 is similar to step 301 and will not be elaborated here. Moreover, the server in this embodiment and the server that recommends multimedia resources to users can be the same server or different servers, and no specific limitation is made thereto.

[0189] 602. The server determines the multiple first-sample resource types corresponding to the multiple first-sample multimedia resources and the multiple second-sample resource types corresponding to the multiple second-sample multimedia resources.

[0190] When the server determines the first-sample resource types and the second-sample resource types, it can also determine the first-sample resource types and the second-sample resource types corresponding to each account. Among them, the manner in which the server determines the first-sample resource types and the second-sample resource types corresponding to each account is similar to step 302 and will not be elaborated here.

[0191] 603. The server determines a first-sample type sequence composed of the multiple first-sample resource types and the multiple second-sample resource types based on the time sequence.

[0192] For each account among the multiple accounts, the server determines the first-sample type sequence corresponding to each account. Among them, the manner in which the server determines the first-sample type sequence corresponding to each account is similar to step 303 and will not be elaborated here.

[0193] See Figure 7, for each account, the server can construct a first sample resource queue based on multiple first sample multimedia resources, construct a second sample resource queue based on multiple second sample multimedia resources, then determine the first sample resource type of the multiple first sample multimedia resources, construct a second sample type sequence, and determine at least one second sample resource type corresponding to every m second sample multimedia resources in units of m. If the at least one second sample resource type is different from the multiple first sample resource types in the second sample type sequence, add the at least one second sample resource type to the second sample type sequence to obtain a first sample type sequence. Moreover, determine the tag type of the at least one second sample resource type and mark it.

[0194] 604. The server performs model training based on the association relationship between each sample resource type in the first sample type sequence and time to obtain a resource type prediction model.

[0195] This resource type prediction model is used to predict the resource types that the target account is interested in, so as to recommend multimedia resources corresponding to the predicted resource types to the target account.

[0196] This step can be implemented through the following steps (1) to (4), including:

[0197] (1) The server respectively takes s and z as units, and based on the time order of each sample resource type in the first sample type sequence, obtains s sample resource types and z sample resource types from the first sample type sequence.

[0198] For the first sample type sequence corresponding to each account, the server obtains s sample resource types and z sample resource types from this first sample type sequence. The z sample resource types are the sample resource types after the s sample resource types, that is, the s sample resource types are the sample resource types with earlier time order, and the z sample resource types are the sample resource types with later time order.

[0199] Among them, both s and z are integers greater than zero, and the quantities of s and z can be the same or different, and both can be set and changed as needed. For example, s is 300, s is 30, or s is 30 and z is 1.

[0200] (2) The server performs model training for the current iteration based on the s sample resource types and the z sample resource types.

[0201] In a possible implementation manner, the server directly performs model training for the current iteration based on the s sample resource types and the z sample resource types.

[0202] In this implementation method, for each account, the server iteratively trains the initial model with s sample resource types to obtain z predicted sample resource types, and compares these z predicted sample resource types with the known z sample resource types. When the z predicted sample resource types are the same as the known z sample resource types, the model training for the next iteration is carried out. When there are predicted sample resource types among the z predicted sample resource types that are different from the known z sample resource types, the server can adjust the initial model parameters through backpropagation based on the known z sample resource types, and then re - conduct the iterative training until the z predicted sample resource types are the same as the known z sample resource types, and then carry out the model training for the next iteration.

[0203] In another possible implementation method, the server also obtains the sample label types of each sample resource type, and conducts the model training for the current iteration based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types.

[0204] In this implementation method, when the server iteratively trains the initial model with s sample resource types, it not only obtains z predicted sample resource types, but also obtains the label types of these z predicted sample resource types, compares these z predicted sample resource types with the known z sample resource types, and compares the label types of these z predicted sample resource types with the label types of the known z sample resource types.

[0205] When the z predicted sample resource types are the same as the known z sample resource types, and the label types of these z predicted sample resource types are the same as the label types of the known z sample resource types, the model training for the next iteration is carried out. When there are predicted sample resource types among the z predicted sample resource types that are different from the known z sample resource types, or there are label types among the label types of these z predicted sample resource types that are different from the label types of the known z sample resource types, the server can adjust the initial model parameters through backpropagation based on the known z sample resource types and their label types, and then re - conduct the iterative training until the z predicted sample resource types are the same as the known z sample resource types, and the label types of these z predicted sample resource types are the same as the label types of the known z sample resource types, and then carry out the model training for the next iteration.

[0206] (3) The server re - obtains s sample resource types and z sample resource types from the first sample type sequence.

[0207] The s re-acquired sample resource types are the sample resource types obtained by shifting the s sample resource types in the current iteration backward by y sample resource types. The z re-acquired sample resource types are the sample resource types obtained by shifting the z sample resource types in the current iteration backward by y sample resource types. Here, y is an integer greater than zero.

[0208] For example, both s and z are 7. For each first sample type sequence, the server uses the time order of 1 to 7 sample resource types represented by S t to predict the time order of 8 to 14 sample resource types. If y is 1, the server uses the time order of 2 to 8 sample resource types represented by S t+1 to predict the time order of 9 to 15 sample resource types. See Figure 8 . If y is 2, the server uses the time order of 3 to 9 sample resource types represented by S t+1 to predict the time order of 10 to 16 sample resource types.

[0209] (4) The server performs model training for the next iteration based on the re-acquired s sample resource types and the re-acquired z sample resource types until the convergence condition is met, and obtains a resource type prediction model.

[0210] The server performs model training for the next iteration based on the re-acquired s sample resource types and the re-acquired z sample resource types until the predicted z sample resource types are the same as the known z sample resource types. Then, with y as the step size, s sample resource types and z sample resource types are re-acquired from the first sample type sequence. Based on the re-acquired s sample resource types and z sample resource types, model training for iteration is performed again until the number of iteration training reaches the preset number of training times, or the prediction accuracy is high and the degree of change is small, and a resource type prediction model is obtained.

[0211] Alternatively, when the server re-obtains s sample resource types and z sample resource types, it can also obtain the sample label types of the z sample resource types, and then perform the next iteration of model training based on the re-obtained s sample resource types, the re-obtained z sample resource types, and the sample label types of the re-obtained z sample resource types, until the predicted z sample resource types are the same as the known z sample resource types, and the label types of these z sample resource types are the same as the label types of the known z sample resource types. Then, with y as the step size, re-obtain s sample resource types, z sample resource types, and the label types of these z sample resource types from the first sample type sequence, and perform iterative model training again based on the re-obtained s sample resource types, z sample resource types, and the label types of these z sample resource types, until the number of iterative training reaches the preset number of training times, or the prediction accuracy is relatively high and the degree of change is relatively small, to obtain a resource type prediction model.

[0212] In the embodiments of the present disclosure, when training the model, it is based on the sample resource types with earlier time order in the first sample type sequence to predict the sample resource types with later time order, that is, the resource type prediction model is trained based on the correlation between the sample resource types and time. Therefore, the trained resource type prediction model can predict the resource types it is interested in based on the resource types of recently appeared multimedia resources, so as to recommend multimedia resources corresponding to the resource types the user is interested in, thereby improving the recommendation effect of multimedia resources.

[0213] It should be noted that the server can train the initial model through algorithms such as DQN (Deep-Q Network, deep reinforcement learning), collaborative filtering, and KNN (K-Nearest Neighbors). These algorithms are applicable to discrete data with time series characteristics. Here, only the case where the server trains the initial model through the DQN algorithm is taken as an example for illustration. Among them, the DQN algorithm includes two parts, one part is the Eval network (evaluation network), and the other part is the Target network (target network).

[0214] See Figure 9 , here only the case where the server predicts 1 predicted sample resource type through multiple sample resource types is taken as an example for illustration. The server inputs the multiple sample resource types represented by S t into the Eval network, and the initial model predicts a1, a2, a3, a4... a w predicted sample resource types and the probability of each predicted sample resource type. The server selects the 1 predicted sample resource type with the highest probability from the w predicted sample resource types, that is, at Among them, w is an integer greater than zero. Then the server takes a t of the predicted sample resource types and compares them with the known corresponding sample resource types. If the predicted sample resource type is the same as the known sample resource type, a relatively high reward value, that is, r t is given to the initial model. If the predicted sample resource type is different from the known sample resource type and the difference is large, a relatively low reward value, that is, r t is given to the initial model. Then the server inputs the multiple sample resource types represented by S t+1 into the Target network to obtain a t+1 . Then the server inputs the multiple sample resource types represented by S t+2 into the Eval network, and cross-trains the Eval network and the Target network in this way.

[0215] During the training process, the server can determine the prediction error rate based on the prediction accuracy Q1 of the Eval network, the prediction accuracy Q2 of the Target network, and r t and then update the Eval network in reverse. At the same time, the Target network is updated once every c predictions of the sample resource type in units of C. By updating the Eval network and the Target network, the prediction accuracy of the model is gradually improved until the model prediction is accurate or reaches the preset number of iterations, and a resource type prediction model is obtained.

[0216] In the embodiments of the present disclosure, based on the association relationship between each sample resource type and time as the behavior and recommendation strategy in reinforcement learning for real-time interaction, the model can dynamically optimize its own recommendation strategy according to each behavior of the user, so as to achieve a more perfect construction of the resource type prediction model. Therefore, the resource type prediction model trained by the above method can understand and predict the user's interests, and predict the interest migration existing in the user itself, so as to recommend interesting multimedia resources to the user, thereby improving the recommendation effect of multimedia resources.

[0217] See Figure 10 , the server obtains the user's historical multimedia resources and recent multimedia resources, then constructs a first sample type sequence, binds user attributes through the first sample type sequence, trains a model based on the first sample type sequence to obtain a resource type prediction model, and predicts the resource types that the user is interested in according to the resource type prediction model, so as to recommend interesting multimedia resources to the user, thereby improving the recommendation effect of multimedia resources.

[0218] It should be noted that steps 601 to 604 can also be executed by a terminal, on which a client of the target application is installed. The client is a client for recommending multimedia resources. Based on the association relationship between the sample resource type and time, the client can perform model training to obtain a resource type prediction model, and use this resource type prediction model to predict the resource type that the user is interested in.

[0219] In the embodiments of the present disclosure, a model is trained based on sample label types, and the sample label types can represent the degree of interest of users in sample resource types. Therefore, the model trained based on sample label types can predict the degree of interest of users in each resource type, that is, predict which resource type the user is more interested in, so as to recommend multimedia resources corresponding to the resource type that the user is interested in, thereby improving the recommendation effect of multimedia resources.

[0220] As can be seen from the above, the method provided by the embodiments of the present disclosure mines new label types from the original data. Instead of simply collecting multimedia resources, it further determines which resource types are new interests for users, that is, determines label types, such as the first label and the second label, and marks them. These label types can be combined with the calculation method of Reward in reinforcement learning during the model training process to provide richer input information for the resource type prediction model, which can help the model converge faster. The resource type prediction model obtained by this method can not only predict the resource type that the user is interested in, but also predict the label type of this resource type, thereby realizing accurate recommendation.

[0221] In the embodiments of the present disclosure, when training the model, rules are discovered from the user's behavior and preference data. Considering the association relationship between each sample resource type in the first sample type sequence and time, model training is performed. The obtained resource type prediction model can change the recommended multimedia resources in a timely manner when the user's interest changes, and finely grasp the user's recent preferences according to the user's recent interests, so as to accurately recommend the multimedia resources that the user hopes to see recently.

[0222] Figure 11 is a block diagram of a multimedia resource recommendation device shown according to an exemplary embodiment, as Figure 11 shown, the device includes:

[0223] A first acquisition unit 1101, configured to acquire a plurality of first multimedia resources and a plurality of second multimedia resources. The plurality of first multimedia resources are multimedia resources recommended for a target account within a first historical time range, and the plurality of second multimedia resources are multimedia resources recommended for the target account within a second historical time range. The start time of the second historical time range is not earlier than the end time of the first historical time range;

[0224] A first determination unit 1102, configured to determine a plurality of first resource types corresponding to a plurality of first multimedia resources and a plurality of second resource types corresponding to a plurality of second multimedia resources;

[0225] A second determination unit 1103, configured to determine a first type sequence composed of the plurality of first resource types and the plurality of second resource types based on a chronological order;

[0226] A recommendation unit 1104, configured to recommend multimedia resources to a target account based on the association relationship between each resource type in the first type sequence and time.

[0227] In some embodiments, the recommendation unit 1104 is configured to determine a label type of each resource type in the first type sequence, where the label type is used to represent the degree of interest of the target account in the resource type; determine the association relationship between the label type of each resource type and time based on the association relationship between each resource type in the first type sequence and time; recommend multimedia resources to the target account based on the association relationship between the label type of each resource type and time.

[0228] In some embodiments, the recommendation unit 1104 is configured to determine that the label type of a resource type is a first label if the multimedia resource corresponding to the resource type meets a first preset condition; determine that the resource label of the resource type is a second label if the multimedia resource corresponding to the resource type meets a second preset condition, where the first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

[0229] In some embodiments, the recommendation unit 1104 is configured to input each resource type in the first type sequence into a resource type prediction model based on a chronological order to obtain a predicted resource type, where the resource type prediction model is used to predict the resource types that the target account is interested in; obtain the multimedia resources corresponding to the predicted resource types, and recommend the multimedia resources to the target account.

[0230] In some embodiments, the first determination unit 1102 is configured to divide the plurality of second multimedia resources into n multimedia resource sets based on a chronological order in units of m, where each multimedia resource set includes m second multimedia resources, and m is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources to obtain a plurality of second resource types.

[0231] In some embodiments, the first determination unit 1102 is configured to divide the second historical time range into a plurality of third historical time ranges in units of a preset duration; determine a plurality of third multimedia resources corresponding to each third historical time range in chronological order; for each third historical time range, divide the plurality of third multimedia resources corresponding to the third historical time range into p multimedia resource sets in units of m in chronological order, where p is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources, obtaining a plurality of second resource types.

[0232] In some embodiments, the second determination unit 1103 is configured to sort the plurality of first resource types based on the chronological order of the plurality of first resource types, obtaining a second type sequence; add the second resource types different from the plurality of first resource types among the plurality of second resource types to the second type sequence based on the chronological order of the plurality of second resource types, obtaining a first type sequence.

[0233] The embodiments of the present disclosure provide a multimedia resource recommendation device. The device forms a first type sequence based on the chronological order of the first resource types of the historically recommended multimedia resources and the second resource types of the recently recommended multimedia resources. Since the resource types in the first type sequence are arranged in chronological order, and the arrangement order of these resource types can reflect the recent interest change of the user in resource types, therefore, the device can predict the resource types that the user is interested in, and then recommend the multimedia resources corresponding to the resource types that the user is interested in, thereby improving the recommendation effect of the multimedia resources.

[0234] It should be noted that: when the above-mentioned multimedia resource recommendation device provided in the embodiments recommends multimedia, only the division of the above-mentioned each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the server is divided into different functional modules to complete all or part of the functions described above. In addition, the above-mentioned multimedia resource recommendation device provided in the embodiments and the embodiments of the multimedia resource recommendation method belong to the same concept, and the specific implementation process thereof can be seen in the method embodiments, which will not be elaborated here.

[0235] Figure 12 is a block diagram of a multimedia resource recommendation device shown according to an exemplary embodiment, as Figure 12 shown, the device includes:

[0236] A second acquisition unit 1201, configured to acquire a plurality of first sample multimedia resources and a plurality of second sample multimedia resources, where the plurality of first sample multimedia resources are multimedia resources recommended for a sample account within a first sample historical time range, and the plurality of second sample multimedia resources are multimedia resources recommended for the sample account within a second sample historical time range, and the start time of the second sample historical time range is not earlier than the end time of the first sample historical time range;

[0237] A third determination unit 1202, configured to determine a plurality of first sample resource types corresponding to the plurality of first sample multimedia resources and a plurality of second sample resource types corresponding to the plurality of second sample multimedia resources;

[0238] A fourth determination unit 1203, configured to determine, based on the time sequence, a first sample type sequence composed of the plurality of first sample resource types and the plurality of second sample resource types;

[0239] A model training unit 1204, configured to perform model training based on the association relationship between each sample resource type in the first sample type sequence and time, to obtain a resource type prediction model, where the resource type prediction model is used to predict the resource types that a target account is interested in, so as to recommend multimedia resources corresponding to the predicted resource types to the target account.

[0240] In some embodiments, the model training unit 1204 is configured to, respectively in units of s and z, based on the time sequence of each sample resource type in the first sample type sequence, obtain s sample resource types and z sample resource types from the first sample type sequence, where the z sample resource types are the sample resource types after the s sample resource types; perform model training for the current iteration based on the s sample resource types and the z sample resource types; re-obtain s sample resource types and z sample resource types from the first sample type sequence, where the re-obtained s sample resource types are the sample resource types obtained by shifting the s sample resource types in the current iteration backward by y sample resource types, and the re-obtained z sample resource types are the sample resource types obtained by shifting the z sample resource types in the current iteration backward by y sample resource types, and s, z, and y are all integers greater than zero; perform model training for the next iteration based on the re-obtained s sample resource types and the re-obtained z sample resource types until a convergence condition is met, to obtain the resource type prediction model.

[0241] In some embodiments, the model training unit 1204 is configured to obtain the sample label types of the z sample resource types; perform model training for the current iteration based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types.

[0242] An embodiment of the present disclosure provides a multimedia resource recommendation device. The device trains a model based on the association relationship between each resource type and time in the first sample type sequence. The arrangement order of these resource types can reflect the change in the user's historical interest in resource types. Therefore, the model trained by the device can predict the resource types that the user is interested in, so that the server can recommend multimedia resources corresponding to the predicted resource types to the user, thereby improving the recommendation effect of multimedia resources.

[0243] It should be noted that when the multimedia resource recommendation device provided in the above embodiment recommends multimedia, only the division of the above function modules is used for illustration. In actual applications, the above functions can be allocated to different function modules according to needs, that is, the internal structure of the server is divided into different function modules to complete all or part of the functions described above. In addition, the multimedia resource recommendation device provided in the above embodiment and the embodiment of the multimedia resource recommendation method belong to the same concept. For the specific implementation process, please refer to the method embodiment, which will not be repeated here.

[0244] Figure 13 It is a schematic structural diagram of a server provided by an embodiment of the present disclosure. The server 1300 may vary greatly due to configuration or performance differences, and may include one or more central processing units (CPUs) 1301 and one or more memories 1302. Among them, instructions are stored in the memory 1302, and the instructions are loaded and executed by the processor 1301 to implement the multimedia resource recommendation method provided by each of the above method embodiments. Of course, the server 1300 may also have components such as wired or wireless network interfaces, keyboards, and input / output interfaces for input / output. The server 1300 may also include other components for implementing device functions, which will not be elaborated here.

[0245] In an exemplary embodiment, a computer-readable storage medium is also provided. When the instructions in the computer-readable storage medium are executed by the processor of the server, the server can execute the multimedia resource recommendation method in the above embodiment.

[0246] In an exemplary embodiment, a computer program product is also provided. The computer program product includes a computer program, and when the computer program is executed by the processor, it implements the multimedia resource recommendation method in the above embodiment.

[0247] Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.

[0248] It should be understood that the present disclosure is not limited to the exact structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A method for recommending multimedia resources, characterized in that, The method includes: Obtaining a plurality of first multimedia resources and a plurality of second multimedia resources, where the plurality of first multimedia resources are the multimedia resources recommended for the target account within a first historical time range, and the plurality of second multimedia resources are the multimedia resources recommended for the target account within a second historical time range, and the start time of the second historical time range is not earlier than the end time of the first historical time range; Determining a plurality of first resource types corresponding to the plurality of first multimedia resources and a plurality of second resource types corresponding to the plurality of second multimedia resources; Determining a first type sequence composed of the plurality of first resource types and the plurality of second resource types based on the time sequence; Recommending multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time.

2. The method according to claim 1, wherein The recommending multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time includes: Determining the label type of each resource type in the first type sequence, where the label type is used to represent the degree of interest of the target account in the resource type; Determining the association relationship between the label type of each resource type and time based on the association relationship between each resource type in the first type sequence and time; Recommending multimedia resources to the target account based on the association relationship between the label type of each resource type and time.

3. The method according to claim 2, wherein The determining the label type of each resource type in the first type sequence includes: If the multimedia resource corresponding to the resource type meets a first preset condition, determining the label type of the resource type as a first label; If the multimedia resource corresponding to the resource type meets a second preset condition, determining the resource label of the resource type as a second label, where the first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

4. The method according to claim 1, characterized in that, The recommending multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time includes: Inputting each resource type in the first type sequence into a resource type prediction model based on the time sequence to obtain a predicted resource type, where the resource type prediction model is used to predict the resource type that the target account is interested in; Obtaining the multimedia resource corresponding to the predicted resource type and recommending the multimedia resource to the target account.

5. The method according to claim 1, wherein The process of determining a plurality of second resource types corresponding to the plurality of second multimedia resources includes: Taking m as a unit, dividing the plurality of second multimedia resources into n multimedia resource sets based on the time sequence, where each multimedia resource set includes m second multimedia resources, and m is an integer greater than zero; For each multimedia resource set, determining at least one second resource type corresponding to the m second multimedia resources to obtain the plurality of second resource types.

6. The method according to claim 1, wherein The process of determining a plurality of second resource types corresponding to the plurality of second multimedia resources includes: Dividing the second historical time range into a plurality of third historical time ranges in units of a preset duration; Determine multiple third multimedia resources corresponding to each third historical time range in chronological order; For each third historical time range, divide the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets in chronological order, where p is an integer greater than zero; For each multimedia resource set, determine at least one second resource type corresponding to the m third multimedia resources to obtain the multiple second resource types.

7. The method according to claim 1, wherein The determining, in chronological order, of the first type sequence composed of the multiple first resource types and the multiple second resource types includes: Sort the multiple first resource types based on the chronological order of the multiple first resource types to obtain a second type sequence; Add the second resource types different from the multiple first resource types among the multiple second resource types to the second type sequence based on the chronological order of the multiple second resource types to obtain the first type sequence.

8. A multimedia resource recommendation method, characterized in that, The method includes: Obtain multiple first sample multimedia resources and multiple second sample multimedia resources, where the multiple first sample multimedia resources are multimedia resources recommended for a sample account within a first sample historical time range, and the multiple second sample multimedia resources are multimedia resources recommended for the sample account within a second sample historical time range, and the start time of the second sample historical time range is not earlier than the end time of the first sample historical time range; Determine multiple first sample resource types corresponding to the multiple first sample multimedia resources and multiple second sample resource types corresponding to the multiple second sample multimedia resources; Based on chronological order, determine a first sample type sequence composed of the multiple first sample resource types and the multiple second sample resource types; Based on the association relationship between each sample resource type in the first sample type sequence and time, perform model training to obtain a resource type prediction model, where the resource type prediction model is used to predict the resource types that a target account is interested in, so as to recommend multimedia resources corresponding to the predicted resource types to the target account.

9. The method according to claim 8, characterized in that The performing, based on the association relationship between each sample resource type in the first sample type sequence and time, of model training to obtain a resource type prediction model includes: Respectively taking s and z as units, based on the chronological order of each sample resource type in the first sample type sequence, obtain s sample resource types and z sample resource types from the first sample type sequence, where the z sample resource types are sample resource types after the s sample resource types; Based on the s sample resource types and the z sample resource types, perform model training for the current iteration; Re-obtain s sample resource types and z sample resource types from the first sample type sequence. The re-obtained s sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the s sample resource types in the current iteration. The re-obtained z sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the z sample resource types in the current iteration. s, z, and y are all integers greater than zero; Based on the re-obtained s sample resource types and the re-obtained z sample resource types, perform model training for the next iteration until the convergence condition is met to obtain the resource type prediction model.

10. The method according to claim 9, characterized in that, The performing model training for the current iteration based on the s sample resource types and the z sample resource types includes: Obtain the sample label types of the z sample resource types; Based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types, perform model training for the current iteration.

11. A multimedia resource recommendation device, characterized in that, The device includes: A first acquisition unit configured to acquire a plurality of first multimedia resources and a plurality of second multimedia resources. The plurality of first multimedia resources are the multimedia resources recommended for the target account within the first historical time range, and the plurality of second multimedia resources are the multimedia resources recommended for the target account within the second historical time range. The starting time of the second historical time range is not earlier than the ending time of the first historical time range; A first determination unit configured to determine a plurality of first resource types corresponding to the plurality of first multimedia resources and a plurality of second resource types corresponding to the plurality of second multimedia resources; A second determination unit configured to determine, based on the time sequence, a first type sequence composed of the plurality of first resource types and the plurality of second resource types; A recommendation unit configured to recommend multimedia resources to the target account based on the association relationship between each resource type in the first type sequence and time.

12. The device according to claim 11, wherein The recommendation unit is configured to determine the label type of each resource type in the first type sequence, and the label type is used to represent the degree of interest of the target account in the resource type; Based on the association relationship between each resource type in the first type sequence and time, determine the association relationship between the label type of each resource type and time; Based on the association relationship between the label type of each resource type and time, recommend multimedia resources to the target account.

13. The device according to claim 12, wherein The recommendation unit is configured to, if the multimedia resource corresponding to the resource type meets the first preset condition, determine that the label type of the resource type is the first label; if the multimedia resource corresponding to the resource type meets the second preset condition, determine that the resource label of the resource type is the second label. The first label represents a higher degree of interest of the target account in the resource type than the second label represents the degree of interest of the target account in the resource type.

14. The device according to claim 11, wherein The recommendation unit is configured to input each resource type in the first type sequence into a resource type prediction model based on chronological order to obtain a predicted resource type, where the resource type prediction model is used to predict the resource types that the target account is interested in; obtain the multimedia resources corresponding to the predicted resource type, and recommend the multimedia resources to the target account.

15. The device according to claim 11, wherein The first determination unit is configured to divide the multiple second multimedia resources into n multimedia resource sets based on chronological order in units of m, where each multimedia resource set includes m second multimedia resources, and m is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m second multimedia resources to obtain the multiple second resource types.

16. The device according to claim 11, characterized in that, The first determination unit is configured to divide the second historical time range into multiple third historical time ranges in units of a preset duration; determine multiple third multimedia resources corresponding to each third historical time range in chronological order; for each third historical time range, divide the multiple third multimedia resources corresponding to the third historical time range into p multimedia resource sets in units of m, where p is an integer greater than zero; for each multimedia resource set, determine at least one second resource type corresponding to the m third multimedia resources to obtain the multiple second resource types.

17. The device according to claim 11, characterized in that, The second determination unit is configured to sort the multiple first resource types based on their chronological order to obtain a second type sequence; add the second resource types different from the multiple first resource types in the multiple second resource types to the second type sequence based on their chronological order to obtain the first type sequence.

18. A multimedia resource recommendation device, characterized in that, The device includes: A second acquisition unit configured to acquire multiple first sample multimedia resources and multiple second sample multimedia resources, where the multiple first sample multimedia resources are the multimedia resources recommended to a sample account within a first sample historical time range, and the multiple second sample multimedia resources are the multimedia resources recommended to the sample account within a second sample historical time range, and the start time of the second sample historical time range is not earlier than the end time of the first sample historical time range; A third determination unit configured to determine multiple first sample resource types corresponding to the multiple first sample multimedia resources and multiple second sample resource types corresponding to the multiple second sample multimedia resources; A fourth determination unit configured to determine a first sample type sequence composed of the multiple first sample resource types and the multiple second sample resource types based on chronological order; A model training unit configured to perform model training based on the association relationship between each sample resource type in the first sample type sequence and time to obtain a resource type prediction model, where the resource type prediction model is used to predict the resource types that the target account is interested in, so as to recommend the multimedia resources corresponding to the predicted resource types to the target account.

19. The device according to claim 18, characterized in that, The model training unit is configured to obtain s sample resource types and z sample resource types from the first sample type sequence respectively in units of s and z, based on the chronological order of each sample resource type in the first sample type sequence, where the z sample resource types are the sample resource types after the s sample resource types; Based on the s sample resource types and the z sample resource types, perform model training for the current iteration; re-obtain s sample resource types and z sample resource types from the first sample type sequence, where the re-obtained s sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the s sample resource types in the current iteration, and the re-obtained z sample resource types are the sample resource types obtained by shifting y sample resource types backward based on the z sample resource types in the current iteration. s, z, and y are all integers greater than zero; based on the re-obtained s sample resource types and the re-obtained z sample resource types, perform model training for the next iteration until the convergence condition is met to obtain the resource type prediction model.

20. The device according to claim 19, wherein The model training unit is configured to obtain the sample label types of the z sample resource types; based on the s sample resource types, the z sample resource types, and the sample label types of the z sample resource types, perform model training for the current iteration.

21. A server, characterized in that, Comprising: A processor; A memory for storing executable instructions of the processor; Wherein, the processor is configured to execute the instructions to implement the multimedia resource recommendation method according to any one of claims 1 to 7 or 8 to 10.

22. A computer-readable storage medium, characterized by, When the instructions in the computer-readable storage medium are executed by the processor of the server, the server is enabled to execute the multimedia resource recommendation method according to any one of claims 1 to 7 or 8 to 10.

23. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multimedia resource recommendation method according to any one of claims 1 to 7 or 8 to 10.

Images