Multimedia resource recommendation method, device, electronic device and storage medium

Abstract

The present disclosure relates to a multimedia resource recommendation method, device, electronic device and storage medium, and belongs to the field of computer technology. The method includes: obtaining account information to be recommended and recommendation parameters corresponding to the account information, obtaining target parameters of each multimedia resource based on the account information, recommendation parameters and resource information of multiple multimedia resources to be recommended, and determining at least one multimedia resource with the highest target parameter from multiple multimedia resources as a target multimedia resource based on the target parameters of multiple multimedia resources, and recommending the target multimedia resource to a target account. The method provided by the embodiment of the present disclosure introduces a recommendation parameter corresponding to the account information, and recommends the target multimedia resource of interest to the target account based on the account information, recommendation parameters and resource information of multiple multimedia resources, so as to realize a personalized recommendation scheme for the account, thereby ensuring the recommendation effect.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a multimedia resource recommendation method, device, electronic device and storage medium. Background Art

[0002] With the development of Internet technology, there are more and more multimedia resources on the Internet. Each resource display platform stores multiple multimedia resources. Usually, based on the resource information of the multimedia resources and the account information of the account, the multimedia resources are recommended to the account so that the user to whom the account belongs can view the recommended multimedia resources. However, the recommendation effect of this recommendation method is still poor. Summary of the invention

[0003] The present disclosure provides a multimedia resource recommendation method, device, electronic device and storage medium, which improve the recommendation effect.

[0004] According to one aspect of an embodiment of the present disclosure, a multimedia resource recommendation method is provided, the method comprising:

[0005] Obtaining account information to be recommended and recommendation parameters corresponding to the account information;

[0006] Based on the account information, the recommendation parameter, and resource information of the plurality of multimedia resources to be recommended, obtaining a target parameter for each of the multimedia resources, the target parameter indicating the possibility of a terminal corresponding to the target account performing an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information;

[0007] Based on the target parameters of the plurality of multimedia resources, determining at least one multimedia resource having the highest target parameter from the plurality of multimedia resources as the target multimedia resource;

[0008] The target multimedia resource is recommended to the target account.

[0009] The method provided by the embodiment of the present disclosure introduces a recommendation parameter corresponding to the account information, and the recommendation parameter can reflect the recommendation of multimedia resources matching the account information. When recommending multimedia resources for an account, the target parameter of each multimedia resource is determined based on the account information, the recommendation parameter corresponding to the account information, and the resource information of multiple multimedia resources to be recommended. The recommendation of multimedia resources matching the account information is taken into consideration, so that the target parameter matches the target account. The target parameter can indicate the degree of interest of the target account in the multimedia resources, so that the target multimedia resource that the target account is most interested in among the multiple multimedia resources can be recommended to the target account, and a personalized recommendation scheme for the account can be implemented, thereby ensuring the recommendation effect.

[0010] In some embodiments, the acquiring the target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended includes:

[0011] For each of the multimedia resources to be recommended, the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource are fused to obtain a first fused feature;

[0012] Based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, respectively obtaining a first parameter and a second parameter of the multimedia resource, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates the possibility that a terminal corresponding to the target account performs a conversion operation after performing the click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0013] The first parameter and the second parameter are merged to obtain a target parameter of the multimedia resource.

[0014] Since the recommendation situations of the multimedia resources indicated by the first parameter and the second parameter are different and can represent different information, a multi-tasking approach is adopted to first obtain the first parameter and the second parameter to enrich the obtained information, and then the first parameter and the second parameter are fused to obtain the target parameter, thereby realizing a method of obtaining the target parameter by using multiple parameters, thereby improving the accuracy of the target parameter.

[0015] In some embodiments, the obtaining of the first parameter and the second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature, respectively, includes:

[0016] Performing feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature;

[0017] Based on the recommended parameter feature, the first fusion feature and the weight feature, the first parameter and the second parameter are obtained respectively.

[0018] The weight feature is obtained through the recommendation parameter feature. Based on the recommendation parameter feature, the first fusion feature and the weight feature, the first parameter and the second parameter are obtained respectively, and the obtained information is enriched to ensure the accuracy of the obtained first parameter and the second parameter, so that the effect of personalized recommendation can be achieved for the target account in the future.

[0019] In some embodiments, the method further comprises:

[0020] Based on the account information and the resource information of the multimedia resource, acquiring an association feature, the association feature indicating an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicating an interaction between the terminal corresponding to the target account and other multimedia resources, the other multimedia resources being multimedia resources recommended by the local end and other than the multimedia resources;

[0021] The step of fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature includes:

[0022] The account feature, the multimedia resource feature and the associated feature are fused to obtain the first fused feature.

[0023] Since the association feature not only indicates the interaction between the target account and the multimedia resource, but also indicates the interaction between the target account and other multimedia resources, the association feature is introduced to enrich the information contained in the first fusion feature to ensure the accuracy of the target parameters subsequently obtained based on the first fusion feature, thereby ensuring the recommendation effect.

[0024] In some embodiments, the acquiring the target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended includes:

[0025] The recommendation model is called to obtain a target parameter of each multimedia resource based on the account information, the recommendation parameter, and resource information of the plurality of multimedia resources to be recommended.

[0026] The target parameters of each multimedia resource are obtained through the recommendation model to ensure the accuracy of the target parameters and thus ensure the subsequent recommendation effect.

[0027] In some embodiments, the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the calling of the recommendation model, based on the account information, the recommendation parameters and the resource information of the plurality of multimedia resources to be recommended, obtains the target parameter of each of the multimedia resources, including:

[0028] For each of the multimedia resources to be recommended, calling the feature extraction sub-model, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature;

[0029] Calling the click prediction sub-model, and acquiring a first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0030] Calling the conversion prediction sub-model, and acquiring a second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature, the second parameter indicating the possibility that, when the multimedia resource is recommended to the target account, a terminal corresponding to the target account performs the click operation on the multimedia resource and then performs a conversion operation;

[0031] The fusion sub-model is called to fuse the first parameter with the second parameter to obtain the target parameter of the multimedia resource.

[0032] The multiple sub-models included in the recommendation model are used to implement a method of obtaining target parameters by multi-tasking, that is, first obtaining the first parameter and the second parameter to enrich the obtained information, and then fusing the first parameter with the second parameter to obtain the target parameter, thereby realizing a method of obtaining the target parameter by multiple parameters, thereby improving the accuracy of the target parameter.

[0033] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the method further includes:

[0034] Calling the weight acquisition sub-model to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature;

[0035] The calling the click prediction sub-model to obtain the first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature includes:

[0036] Calling the click prediction sub-model, fusing the first fused feature with the weight feature to obtain a second fused feature;

[0037] The click prediction sub-model is called to fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

[0038] The weight feature is obtained based on the weight acquisition sub-model to ensure the accuracy of the weight feature, and based on the click prediction sub-model, the first fusion feature, the weight feature and the recommendation parameter feature are fused, so that the first parameter is integrated with the characteristics of the recommendation parameter matching the account information to ensure that the first parameter matches the account information, thereby ensuring the accuracy of the first parameter, so that the effect of personalized recommendation for the target account can be achieved later.

[0039] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; calling the weight acquisition submodel, performing feature conversion on the recommended parameter features, and obtaining the weight features of the first fusion features, includes:

[0040] Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter feature to obtain the i-th weight feature;

[0041] The calling the click prediction sub-model to fuse the first fused feature with the weight feature to obtain a second fused feature includes:

[0042] Based on the first first parameter acquisition layer, performing dimensionality reduction processing on the first fusion feature to obtain a first first dimensionality reduction feature;

[0043] Based on the first first fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain a first third fusion feature;

[0044] For the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, perform dimensionality reduction processing on the j-1th third fusion feature to obtain the jth first dimensionality reduction feature; based on the jth first fusion layer, fuse the jth first dimensionality reduction feature with the jth weight feature to obtain the jth third fusion feature; repeat the above steps until the second fusion feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

[0045] In the disclosed embodiment, the weight acquisition submodel includes multiple weight acquisition layers, and different weight acquisition layers are used to acquire weights representing features of different dimensions, and the click prediction submodel includes the same number of first parameter acquisition layers and first fusion layers. Based on the first parameter acquisition layer and the first fusion layer included in the click prediction submodel, multiple weight features are fused with the first fusion feature to enhance the features of the first fusion feature in each dimension, so that the obtained second fusion feature matches the characteristics of the recommended parameters matching the account information, thereby ensuring the accuracy of the second fusion feature.

[0046] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the method further includes:

[0047] Calling the weight acquisition sub-model to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature;

[0048] The calling the conversion prediction sub-model to obtain the second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature includes:

[0049] Calling the conversion prediction sub-model, fusing the first fused feature with the weight feature to obtain a fourth fused feature;

[0050] The conversion prediction sub-model is called to fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

[0051] The weight feature is obtained based on the weight acquisition sub-model to ensure the accuracy of the weight feature, and based on the click prediction sub-model, the first fusion feature, the weight feature and the recommendation parameter feature are fused, so that the second parameter obtained incorporates the characteristics of the recommendation parameter matching the account information, to ensure that the obtained second parameter matches the account information, thereby ensuring the accuracy of the second parameter, so that the effect of personalized recommendation for the target account can be achieved later.

[0052] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; calling the weight acquisition submodel, performing feature conversion on the recommended parameter features, and obtaining the weight features of the first fusion features include:

[0053] Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter feature to obtain the i-th weight feature;

[0054] The calling of the conversion prediction sub-model to fuse the first fused feature with the weight feature to obtain a fourth fused feature includes:

[0055] Based on the first second parameter acquisition layer, performing dimensionality reduction processing on the first fusion feature to obtain a first second dimensionality reduction feature;

[0056] Based on the first second fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain a first fifth fusion feature;

[0057] For the jth second parameter acquisition layer and the jth second fusion layer, based on the jth second parameter acquisition layer, perform dimensionality reduction processing on the j-1th fifth fusion feature to obtain the jth second reduced dimensionality feature; based on the jth second fusion layer, fuse the jth second reduced dimensionality feature with the jth weight feature to obtain the jth fifth fusion feature; repeat the above steps until the fourth fusion feature is obtained based on the nth second fusion layer, where j is an integer greater than 1 and not greater than n.

[0058] In the disclosed embodiment, the weight acquisition submodel includes a plurality of weight acquisition layers, and different weight acquisition layers are used to acquire weights representing features of different dimensions, and the conversion prediction submodel includes the same number of second parameter acquisition layers and second fusion layers. Based on the second parameter acquisition layer and the second fusion layer included in the conversion prediction submodel, the plurality of weight features are fused with the first fusion feature to enhance the features of the first fusion feature in each dimension, so that the obtained fourth fusion feature matches the characteristics of the recommended parameters matching the account information, thereby ensuring the accuracy of the fourth fusion feature.

[0059] In some embodiments, the method further comprises:

[0060] Calling the feature extraction sub-model to obtain, based on the account information and the resource information of the multimedia resource, an associated feature, wherein the associated feature indicates an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicates an interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local end and other than the multimedia resources;

[0061] The calling of the feature extraction sub-model to fuse the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature includes:

[0062] The feature extraction sub-model is called to fuse the account feature, the multimedia resource feature and the associated feature to obtain the first fused feature.

[0063] Since the association features not only indicate the interaction between the target account and the multimedia resource, but also indicate the interaction between the target account and other multimedia resources, the association features are obtained based on the feature extraction sub-model, and the account features, multimedia resource features and association features are fused to enrich the information contained in the first fused features to ensure the accuracy of the target parameters subsequently obtained based on the first fused features, thereby ensuring the recommendation effect.

[0064] In some embodiments, the method further comprises:

[0065] Acquire sample data and a sample tag corresponding to the sample data, wherein the sample data includes sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, wherein the sample tag indicates whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0066] Calling the recommendation model, based on the sample data, obtaining a prediction label of a sample multimedia resource corresponding to the sample data, the prediction label indicating the possibility of a terminal corresponding to the sample account performing an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0067] The recommendation model is trained based on the predicted labels of the sample data and the differences between the sample labels.

[0068] In the method provided by the embodiment of the present disclosure, sample data includes sample account information, sample resource information of sample multimedia resources, and sample recommendation parameters corresponding to the sample account information. The recommendation model is trained based on the sample data and the corresponding sample labels to improve the accuracy of the recommendation model so as to subsequently implement personalized recommendations for the account based on the recommendation model.

[0069] In some embodiments, the sample label includes a first sample label, which indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the prediction label includes a first prediction label, which indicates the possibility of the terminal corresponding to the sample account performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; calling the recommendation model to obtain the prediction label of the sample multimedia resource corresponding to the sample data based on the sample data includes:

[0070] Calling the feature extraction sub-model to fuse the sample account feature of the sample account information and the sample multimedia resource feature of the sample resource information to obtain a sample fusion feature;

[0071] Calling the click prediction sub-model, based on the recommendation parameter feature of the sample recommendation parameter and the sample fusion feature, obtaining a second prediction label of the sample multimedia resource, the second prediction label indicating the possibility that a terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0072] Calling the conversion prediction sub-model, and acquiring a third prediction tag of the multimedia resource based on the sample recommendation parameter feature and the sample fusion feature, the third prediction tag indicating the possibility of a terminal corresponding to the sample account performing the click operation and then performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0073] The fusion sub-model is called to fuse the second prediction label with the third prediction label to obtain the first prediction label.

[0074] Since the second prediction label and the third prediction label can represent different information, a multi-task approach is adopted to first obtain the second prediction label and the third prediction label to enrich the obtained information. Then, the second prediction label and the third prediction label are fused to obtain the first prediction label to ensure the accuracy of the obtained first prediction label.

[0075] In some embodiments, the sample label further includes a second sample label, and the second sample label indicates whether a terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the training of the recommendation model based on the difference between the predicted label of the sample data and the sample label includes:

[0076] The recommendation model is trained based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

[0077] The disclosed embodiment adopts a multi-task approach to train the recommendation model, that is, obtaining multiple tasks of the first prediction label, the second prediction label and the third prediction label. Using multi-task to train the recommendation model can alleviate the problem of poor accuracy of the recommendation model caused by sparse sample data, thereby improving the accuracy of the recommendation model.

[0078] In some embodiments, the recommendation parameter is determined based on an operation performed by at least one account logged into a server on a multimedia resource recommended by the server, the server being the server logged into by the target account;

[0079] The recommending the target multimedia resource to the target account includes:

[0080] The target multimedia resource is sent to the server, and the server recommends the target multimedia resource to the target account.

[0081] The disclosed embodiment provides a method for providing recommendation services for multiple servers, using the recommendation parameters of the server logged in by the target account as the recommendation parameters corresponding to the account information. The recommendation parameters can reflect the recommendation of multimedia resources matching the account information, so that the target parameters determined for each multimedia resource match the target account and the server, thereby ensuring the accuracy of the determined target multimedia resources, realizing a personalized recommendation solution for the account in the server, and also ensuring the recommendation effect.

[0082] According to another aspect of the embodiments of the present disclosure, a multimedia resource recommendation device is provided, the device comprising:

[0083] An acquisition unit, configured to acquire account information to be recommended and recommendation parameters corresponding to the account information;

[0084] The acquisition unit is further configured to acquire a target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended, wherein the target parameter indicates the possibility of a terminal corresponding to the target account performing an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information;

[0085] A selection unit is configured to perform, based on target parameters of the plurality of multimedia resources, determining at least one multimedia resource with the highest target parameter from the plurality of multimedia resources as a target multimedia resource;

[0086] The recommendation unit is configured to recommend the target multimedia resource to the target account.

[0087] In some embodiments, the acquisition unit includes:

[0088] A fusion subunit is configured to perform, for each multimedia resource to be recommended, fusion of the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature;

[0089] an acquisition subunit, configured to execute a recommendation parameter feature based on the recommendation parameter and the first fusion feature, and respectively acquire a first parameter and a second parameter of the multimedia resource, wherein the first parameter indicates a possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates a possibility that a terminal corresponding to the target account performs a conversion operation after performing the click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0090] The fusion subunit is further configured to fuse the first parameter with the second parameter to obtain a target parameter of the multimedia resource.

[0091] In some embodiments, the acquisition subunit is configured to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature; and obtain the first parameter and the second parameter respectively based on the recommended parameter feature, the first fusion feature and the weight feature.

[0092] In some embodiments, the acquisition unit is further configured to acquire an associated feature based on the account information and the resource information of the multimedia resource, wherein the associated feature indicates an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicates an interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources;

[0093] The fusion subunit is configured to fuse the account feature, the multimedia resource feature and the associated feature to obtain the first fusion feature.

[0094] In some embodiments, the acquisition unit is configured to execute the call recommendation model, and acquire the target parameter of each of the multimedia resources based on the account information, the recommendation parameters, and the resource information of the multiple multimedia resources to be recommended.

[0095] In some embodiments, the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the acquisition unit includes:

[0096] A fusion subunit is configured to execute, for each multimedia resource to be recommended, calling the feature extraction submodel, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature;

[0097] an acquisition subunit configured to execute calling the click prediction submodel, and acquire a first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, wherein the first parameter indicates a possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0098] The acquisition sub-unit is further configured to execute calling the conversion prediction sub-model, and based on the recommendation parameter feature and the first fusion feature, acquire a second parameter of the multimedia resource, wherein the second parameter indicates the possibility that, when the multimedia resource is recommended to the target account, the terminal corresponding to the target account performs the click operation on the multimedia resource and then performs the conversion operation;

[0099] The fusion sub-unit is further configured to execute the call of the fusion sub-model, fuse the first parameter with the second parameter, and obtain the target parameter of the multimedia resource.

[0100] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the apparatus further includes:

[0101] a conversion unit configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommendation parameter feature, and obtain a weight feature of the first fusion feature;

[0102] The acquisition sub-unit is configured to execute calling the click prediction sub-model, fuse the first fusion feature with the weight feature to obtain a second fusion feature; call the click prediction sub-model, fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

[0103] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; the conversion unit is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature;

[0104] The acquisition subunit is configured to perform dimensionality reduction processing on the first fused feature based on the first first parameter acquisition layer to obtain the first first reduced dimension feature; based on the first first fusion layer, fuse the first reduced dimension feature with the first weight feature to obtain the first third fused feature; for the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, perform dimensionality reduction processing on the j-1th third fused feature to obtain the jth first reduced dimension feature; based on the jth first fusion layer, fuse the jth first reduced dimension feature with the jth weight feature to obtain the jth third fused feature, and repeat the above steps until the second fused feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

[0105] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the apparatus further includes:

[0106] a conversion unit configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommendation parameter feature, and obtain a weight feature of the first fusion feature;

[0107] The acquisition sub-unit is configured to execute calling the conversion prediction sub-model, fuse the first fusion feature with the weight feature to obtain a fourth fusion feature; call the conversion prediction sub-model, fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

[0108] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; the conversion unit is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature;

[0109] The acquisition subunit is configured to perform dimensionality reduction processing on the first fused feature based on the first second parameter acquisition layer to obtain the first second reduced dimension feature; based on the first second fusion layer, fuse the first reduced dimension feature with the first weight feature to obtain the first fifth fused feature; for the jth second parameter acquisition layer and the jth second fusion layer, based on the jth second parameter acquisition layer, perform dimensionality reduction processing on the j-1th fifth fused feature to obtain the jth second reduced dimension feature; based on the jth second fusion layer, fuse the jth second reduced dimension feature with the jth weight feature to obtain the jth fifth fused feature, and repeat the above steps until the fourth fused feature is obtained based on the nth second fusion layer, where j is an integer greater than 1 and not greater than n.

[0110] In some embodiments, the acquisition unit is further configured to execute the call of the feature extraction sub-model, and acquire the associated features based on the account information and the resource information of the multimedia resources, wherein the associated features indicate the interaction between the terminal corresponding to the target account and the multimedia resources, and also indicate the interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources;

[0111] The fusion subunit is configured to execute and call the feature extraction submodel, fuse the account feature, the multimedia resource feature and the associated feature, and obtain the first fusion feature.

[0112] In some embodiments, the apparatus further comprises:

[0113] The acquisition unit is further configured to acquire sample data and a sample tag corresponding to the sample data, wherein the sample data includes sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, and the sample tag indicates whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0114] The acquisition unit is further configured to execute calling the recommendation model, and based on the sample data, acquire a prediction tag of a sample multimedia resource corresponding to the sample data, wherein the prediction tag indicates a possibility that a terminal corresponding to the sample account will perform an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0115] The training unit is configured to train the recommendation model based on the predicted labels of the sample data and the differences between the sample labels.

[0116] In some embodiments, the sample label includes a first sample label, which indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the prediction label includes a first prediction label, which indicates the possibility of the terminal corresponding to the sample account performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the acquisition unit is configured to execute the call of the feature extraction sub-model, fuse the sample account feature of the sample account information and the sample multimedia resource feature of the sample resource information to obtain a sample fusion feature; call the click prediction sub-model The test sub-model obtains a second prediction label of the sample multimedia resource based on the recommendation parameter characteristics of the sample recommendation parameters and the sample fusion characteristics, wherein the second prediction label indicates the possibility that the terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the conversion prediction sub-model is called to obtain a third prediction label of the multimedia resource based on the sample recommendation parameter characteristics and the sample fusion characteristics, wherein the third prediction label indicates the possibility that the terminal corresponding to the sample account will perform a conversion operation after performing the click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the fusion sub-model is called to fuse the second prediction label with the third prediction label to obtain the first prediction label.

[0117] In some embodiments, the sample label also includes a second sample label, which indicates whether the terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the training unit is configured to perform training on the recommendation model based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

[0118] In some embodiments, the recommendation parameter is determined based on an operation performed by at least one account logged into a server on a multimedia resource recommended by the server, the server being the server logged into by the target account;

[0119] The recommendation unit is configured to send the target multimedia resource to the server, and the server recommends the target multimedia resource to the target account.

[0120] According to another aspect of the embodiments of the present disclosure, an electronic device is provided, the electronic device comprising:

[0121] processor;

[0122] a memory for storing instructions executable by the processor;

[0123] The processor is configured to execute the instructions to implement the multimedia resource recommendation method described above.

[0124] According to another aspect of the embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device can execute the multimedia resource recommendation method described in the above aspect.

[0125] According to another aspect of the embodiments of the present disclosure, a computer program product is provided. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute the multimedia resource recommendation method described in the above aspects.

[0126] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0128] Figure 1 It is a schematic diagram showing an implementation environment according to an exemplary embodiment.

[0129] Figure 2 The figure is a flowchart of a multimedia resource recommendation method according to an exemplary embodiment.

[0130] Figure 3 The figure is a flowchart of another multimedia resource recommendation method according to an exemplary embodiment.

[0131] Figure 4 The figure is a schematic diagram showing the structure of a recommendation model according to an exemplary embodiment.

[0132] Figure 5 The present invention is a flowchart showing an advertisement recommendation according to an exemplary embodiment.

[0133] Figure 6 The figure is a flowchart of a recommendation model training method according to an exemplary embodiment.

[0134] Figure 7The figure is a flowchart of a multimedia resource recommendation method according to an exemplary embodiment.

[0135] Figure 8 The figure is a block diagram of a multimedia resource recommendation device according to an exemplary embodiment.

[0136] Fig. 9 The figure is a block diagram of another multimedia resource recommendation device according to an exemplary embodiment.

[0137] Fig.10 It is a block diagram of a terminal according to an exemplary embodiment.

[0138] Fig.11 It is a block diagram of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0139] In order to enable ordinary persons 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 in conjunction with the accompanying drawings.

[0140] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned figure descriptions are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0141] The terms "at least one", "a plurality", "each", and "any" used in the present disclosure include one, two or more than two, a plurality includes two or more than two, each refers to each of the corresponding plurality, and any refers to any one of the plurality. For example, the plurality of accounts includes three accounts, and each refers to each of the three accounts, and any refers to any one of the three accounts, which can be the first account, or the second account, or the third account.

[0142] It should be noted that the user information involved in this disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0143] The multimedia resource recommendation method provided in the embodiments of the present disclosure is executed by an electronic device. In some embodiments, the electronic device is a terminal, for example, the terminal is a mobile phone, a tablet computer, a computer, and other types of terminals. In some embodiments, the electronic device is a server, for example, the server is a server, or a server cluster composed of several servers, or a cloud computing service center.

[0144] In some embodiments, the electronic device is provided as a central server. Figure 1 It is a schematic diagram of an implementation environment provided according to an exemplary embodiment. The implementation environment includes: a central server 101 and at least one platform server 102. The central server 101 and each platform server 102 can interact through a network connection.

[0145] The central server 101 can provide multimedia resource recommendation services for each platform server 102. For each platform server 102, the central server 101 can determine the target multimedia resources for the account based on the recommendation parameters of the platform server 102, the account information of the account logged into the platform server 102, and the resource information of multiple multimedia resources to be recommended in the central server 101, and send the target multimedia resources to the platform server 102 through the network connection between the platform server 102, and the platform server 102 recommends the target multimedia resources to the account.

[0146] In some embodiments, for any platform server 102, when a user logs in to the platform server 102 based on an account, the platform server 102 sends a resource acquisition request to the central server 101. The resource acquisition request carries the account information of the account and the recommendation parameters of the platform server 102. After receiving the resource acquisition request, the central server 101 determines the target multimedia resource to be recommended for the account, sends the target multimedia resource to the platform server 102, and the platform server 102 recommends the target multimedia resource to the account.

[0147] Figure 2 is a flowchart of a multimedia resource recommendation method according to an exemplary embodiment. Figure 2 The method is executed by an electronic device and includes the following steps.

[0148] 201. The electronic device obtains account information to be recommended and recommendation parameters corresponding to the account information.

[0149] The account information is used to describe the target account, and the recommendation parameters corresponding to the account information are used to indicate the recommendation of multimedia resources matching the account information.

[0150] 202. The electronic device obtains a target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended.

[0151] The target parameter indicates the possibility that the terminal corresponding to the target account performs an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account. The operation performed by the terminal corresponding to the target account on the multimedia resource is a click operation, a conversion operation, etc. For example, the target parameter is an exposure conversion rate, which indicates the possibility that the terminal corresponding to the target account performs a conversion operation on the multimedia resource after the multimedia resource is recommended to the target account.

[0152] In the disclosed embodiment, based on the account information of the target account, the recommendation parameters and the resource information of each multimedia resource, it is possible to determine the possibility that the terminal corresponding to the target account performs an operation on the multimedia resource when each multimedia resource is recommended to the target account. The terminal corresponding to the target account performs an operation on the multimedia resource, indicating that the user corresponding to the target account is interested in the multimedia resource. Therefore, the target parameter can reflect the degree of interest of the user corresponding to the target account in the multimedia resource.

[0153] Since the target parameters of each multimedia resource incorporate the recommendation parameters corresponding to the account information and the account information of the target account, the target parameters can be matched with the target account to ensure that multimedia resources can be recommended for the target account later.

[0154] 203. The electronic device determines, based on target parameters of the plurality of multimedia resources, at least one multimedia resource having a highest target parameter from the plurality of multimedia resources as a target multimedia resource.

[0155] Since the target parameter of each multimedia resource represents the possibility of the target account performing an operation on each multimedia resource, that is, the target parameter can indicate the degree of interest of the target account in the multimedia resource, at least one multimedia resource with the highest target parameter is selected as the target multimedia resource, that is, at least one target multimedia resource that the target account is most interested in is selected, so that the target multimedia resource can be subsequently recommended to the target account, thereby ensuring the subsequent recommendation effect.

[0156] 204. The electronic device recommends the target multimedia resource to the target account.

[0157] Since the target multimedia resource is a multimedia resource that the target account is interested in, the target multimedia resource that the target account is interested in is recommended to the target account so that the user corresponding to the target account can view the target multimedia resource, thereby ensuring the recommendation effect.

[0158] The method provided by the embodiment of the present disclosure introduces a recommendation parameter corresponding to the account information, and the recommendation parameter can reflect the recommendation of multimedia resources matching the account information. When recommending multimedia resources for an account, the target parameter of each multimedia resource is determined based on the account information, the recommendation parameter corresponding to the account information, and the resource information of multiple multimedia resources to be recommended. The recommendation of multimedia resources matching the account information is taken into consideration, so that the target parameter matches the target account. The target parameter can indicate the degree of interest of the target account in the multimedia resources, so that the target multimedia resource that the target account is most interested in among the multiple multimedia resources can be recommended to the target account, and a personalized recommendation scheme for the account can be implemented, thereby ensuring the recommendation effect.

[0159] In some embodiments, based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended, obtaining the target parameter of each multimedia resource includes:

[0160] For each multimedia resource to be recommended, the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource are fused to obtain a first fused feature;

[0161] Based on the recommendation parameter feature and the first fusion feature of the recommendation parameter, respectively obtaining a first parameter and a second parameter of the multimedia resource, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates the possibility that a terminal corresponding to the target account performs a conversion operation after performing a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0162] The first parameter is merged with the second parameter to obtain a target parameter of the multimedia resource.

[0163] In some embodiments, based on the recommendation parameter feature and the first fusion feature of the recommendation parameter, respectively obtaining the first parameter and the second parameter of the multimedia resource includes:

[0164] Perform feature conversion on the recommended parameter feature to obtain the weight feature of the first fusion feature;

[0165] Based on the recommended parameter feature, the first fusion feature and the weight feature, the first parameter and the second parameter are obtained respectively.

[0166] In some embodiments, the method further comprises:

[0167] Based on the account information and the resource information of the multimedia resource, an association feature is obtained, where the association feature indicates the interaction between the terminal corresponding to the target account and the multimedia resource, and also indicates the interaction between the terminal corresponding to the target account and other multimedia resources, where the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources;

[0168] The account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource are fused to obtain a first fused feature, including:

[0169] The account feature, multimedia resource feature and associated feature are fused to obtain a first fused feature.

[0170] In some embodiments, based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended, obtaining the target parameter of each multimedia resource includes:

[0171] The recommendation model is called to obtain target parameters of each multimedia resource based on the account information, the recommendation parameters, and the resource information of the plurality of multimedia resources to be recommended.

[0172] In some embodiments, the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model, and a fusion sub-model; calling the recommendation model, based on the account information, the recommendation parameters, and the resource information of the multiple multimedia resources to be recommended, obtaining the target parameters of each multimedia resource, including:

[0173] For each multimedia resource to be recommended, calling the feature extraction sub-model, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature;

[0174] Calling the click prediction sub-model, based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, obtaining a first parameter of the multimedia resource, the first parameter indicating the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0175] Calling the conversion prediction sub-model, based on the recommendation parameter feature and the first fusion feature, obtaining a second parameter of the multimedia resource, the second parameter indicating the possibility that, when the multimedia resource is recommended to the target account, a terminal corresponding to the target account performs a click operation on the multimedia resource and then performs a conversion operation;

[0176] The fusion sub-model is called to fuse the first parameter with the second parameter to obtain the target parameter of the multimedia resource.

[0177] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the method further includes:

[0178] Call the weight acquisition sub-model to perform feature conversion on the recommended parameter features to obtain the weight features of the first fusion feature;

[0179] The click prediction sub-model is called to obtain the first parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature of the recommendation parameter, including:

[0180] Call the click prediction sub-model to fuse the first fused feature with the weight feature to obtain the second fused feature;

[0181] The click prediction sub-model is called to fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

[0182] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; calling the weight acquisition submodel, performing feature conversion on the recommended parameter feature, and obtaining the weight feature of the first fusion feature, including:

[0183] Based on the i-th weight acquisition layer, feature conversion is performed on the recommended parameter features to obtain the i-th weight feature;

[0184] Call the click prediction sub-model to fuse the first fusion feature with the weight feature to obtain the second fusion feature, including:

[0185] Based on the first first parameter acquisition layer, the first fusion feature is subjected to dimensionality reduction processing to obtain the first first dimensionality reduction feature;

[0186] Based on the first first fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain the first third fusion feature;

[0187] For the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, perform dimensionality reduction processing on the j-1th third fusion feature to obtain the jth first dimensionality reduction feature, based on the jth first fusion layer, fuse the jth first dimensionality reduction feature with the jth weight feature to obtain the jth third fusion feature, repeat the above steps until the second fusion feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

[0188] In some embodiments, the recommendation model further includes a weight acquisition sub-model, and the method further includes:

[0189] Call the weight acquisition sub-model to perform feature conversion on the recommended parameter features to obtain the weight features of the first fusion feature;

[0190] The conversion prediction sub-model is called to obtain the second parameter of the multimedia resource based on the recommended parameter feature and the first fusion feature, including:

[0191] Call the conversion prediction sub-model to fuse the first fusion feature with the weight feature to obtain the fourth fusion feature;

[0192] The conversion prediction sub-model is called to fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

[0193] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; calling the weight acquisition submodel, performing feature conversion on the recommended parameter feature, and obtaining the weight feature of the first fusion feature, including:

[0194] Based on the i-th weight acquisition layer, feature conversion is performed on the recommended parameter features to obtain the i-th weight feature;

[0195] The conversion prediction sub-model is called to fuse the first fusion feature with the weight feature to obtain the fourth fusion feature, including:

[0196] Based on the first second parameter acquisition layer, the first fusion feature is subjected to dimensionality reduction processing to obtain the first second dimensionality reduction feature;

[0197] Based on the first second fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain the first fifth fusion feature;

[0198] For the j-th second parameter acquisition layer and the j-th second fusion layer, based on the j-th second parameter acquisition layer, perform dimensionality reduction processing on the j-1-th fifth fusion feature to obtain the j-th second reduced dimensionality feature, based on the j-th second fusion layer, fuse the j-th second reduced dimensionality feature with the j-th weight feature to obtain the j-th fifth fusion feature, repeat the above steps until the fourth fusion feature is obtained based on the n-th second fusion layer, where j is an integer greater than 1 and not greater than n.

[0199] In some embodiments, the method further comprises:

[0200] Calling a feature extraction sub-model to obtain associated features based on the account information and the resource information of the multimedia resources, where the associated features indicate the interaction between the terminal corresponding to the target account and the multimedia resources, and also indicate the interaction between the terminal corresponding to the target account and other multimedia resources, where the other multimedia resources are multimedia resources recommended by the local end and other than the multimedia resources;

[0201] The feature extraction sub-model is called to fuse the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature, including:

[0202] The feature extraction sub-model is called to fuse the account features, multimedia resource features, and associated features to obtain the first fused features.

[0203] In some embodiments, the method further comprises:

[0204] Acquire sample data and a sample tag corresponding to the sample data, the sample data including sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, the sample tag indicating whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0205] Calling the recommendation model, based on the sample data, obtaining a prediction label of the sample multimedia resource corresponding to the sample data, the prediction label indicating the possibility of a terminal corresponding to the sample account performing an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0206] The recommendation model is trained based on the predicted labels of the sample data and the differences between the sample labels.

[0207] In some embodiments, the sample tag includes a first sample tag, the first sample tag indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account, the prediction tag includes a first prediction tag, the first prediction tag indicates the possibility that the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; calling the recommendation model, based on the sample data, obtaining the prediction tag of the sample multimedia resource corresponding to the sample data, includes:

[0208] Calling the feature extraction sub-model to fuse the sample account features of the sample account information and the sample multimedia resource features of the sample resource information to obtain a sample fusion feature;

[0209] Calling the click prediction sub-model, based on the recommendation parameter features of the sample recommendation parameters and the sample fusion features, obtaining a second prediction label of the sample multimedia resource, the second prediction label indicating the possibility that a terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0210] Calling the conversion prediction sub-model, based on the sample recommendation parameter feature and the sample fusion feature, obtaining a third prediction tag of the multimedia resource, the third prediction tag indicating the possibility that when the sample multimedia resource is recommended to the sample account, a terminal corresponding to the sample account performs a click operation on the sample multimedia resource and then performs a conversion operation;

[0211] The fusion sub-model is called to fuse the second prediction label with the third prediction label to obtain the first prediction label.

[0212] In some embodiments, the sample label further includes a second sample label, and the second sample label indicates whether a terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; training the recommendation model based on the difference between the predicted label of the sample data and the sample label includes:

[0213] The recommendation model is trained based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

[0214] In some embodiments, the recommendation parameter is determined based on an operation performed by at least one account logged into a server on a multimedia resource recommended by the server, the server being the server logged into by the target account;

[0215] Recommend target multimedia resources to target accounts, including:

[0216] The target multimedia resource is sent to the server, and the server recommends the target multimedia resource to the target account.

[0217] In the above Figure 2 On the basis of the shown embodiment, the recommendation model can also be called to obtain the target parameters of each multimedia resource. Taking the recommendation model including a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model, a fusion sub-model and a weight acquisition sub-model as an example, the multimedia resource recommendation process is detailed in the following embodiment.

[0218] Figure 3 is a flowchart of a multimedia resource recommendation method according to an exemplary embodiment. Figure 3 The method is executed by a central server and includes the following steps.

[0219] 301. The central server obtains account information to be recommended and recommendation parameters of the server, where the account information refers to the target account of the resource to be recommended, the target account is the account logged into the server, and the recommendation parameters are determined based on the operations performed by at least one account logged into the server on the multimedia resources recommended by the server.

[0220] In the disclosed embodiment, the server is provided with recommendation service by the central server, and the target account referred to by the target information is the account logged into the server, that is, the target account is the account registered in the server. In some embodiments, the server is any server among the multiple servers provided with recommendation service by the central server. Each server provides service for a resource display application, and each resource display application can display multimedia resources.

[0221] The account information is used to describe the target account for logging into the server. In some embodiments, the account information includes the occupation, age, gender, interests, and hobbies of the user corresponding to the target account.

[0222] In the disclosed embodiment, since the target account referred to by the account information is the account logged in to the server, the recommended parameter of the server is also the recommended parameter corresponding to the account information. The recommended parameter can reflect the situation that the account logged in to the server performs an operation on the multimedia resource recommended by the server, that is, it can reflect the situation that the server recommends multimedia resources for the account logged in to the server. In some embodiments, the recommended parameter includes the exposure click rate or click conversion rate of the multimedia resource. The exposure click rate indicates the situation that at least one account logged in to the server performs a click operation on the multimedia resource recommended by the server, and the click conversion rate indicates the situation that at least one account logged in to the server performs a click operation on the multimedia resource recommended by the server and then performs a conversion operation. In some embodiments, the recommended parameter includes the exposure click rate or conversion rate of multimedia resources belonging to different types. For example, the server can recommend multimedia resources for each account logged in to the server, and count the operations performed by each account on the recommended multimedia resources, such as the click operation performed by each account on the multimedia resource or the conversion operation performed after the click operation, etc. By counting the operations performed by each account on the recommended multimedia resources, the recommended parameter of the server can be obtained. In some embodiments, the recommended parameter also includes the associated parameters of the server. The associated parameter is a parameter related to the server. For example, the server provides a server for the target application, and the associated information includes a click rate or a conversion rate corresponding to the application type to which the target application belongs.

[0223] In some embodiments, 301 includes: the server sends a resource acquisition request to the central server, the resource acquisition request carries the account information and the recommended parameters of the server, and the central server receives the resource acquisition request.

[0224] In the disclosed embodiment, the central server provides a recommendation service for the server. When receiving a resource acquisition request sent by the server, it indicates that multimedia resources need to be recommended for a target account logged in to the server, and can subsequently determine the target multimedia resources to be recommended for the target account.

[0225] For example, after the terminal logs in to the server based on the target account, the server sends a resource acquisition request to the central server so as to subsequently recommend target multimedia resources to the target account.

[0226] For another example, the terminal sends a login request to the server based on the target account, and the login request carries the target account. The server receives the login request, verifies whether the target account is an account registered in the server, and if the verification is successful, sends a resource acquisition request to the central server and sends a verification success notification to the terminal. The terminal receives the verification success notification and displays a successful login interface.

[0227] In some embodiments, 301 includes: the server sends a resource acquisition request to the central server, the resource acquisition request carries the account information and the initial recommended parameters of the server, obtains the associated parameters of the server based on the initial recommended parameters of the server, and merges the initial recommended parameters and the associated parameters to obtain the recommended parameters of the server.

[0228] By obtaining the associated parameters related to the server and the information contained in the recommended parameters, the accuracy of the recommended parameters is ensured to ensure the subsequent recommendation effect.

[0229] It should be noted that the embodiment of the present disclosure uses the recommendation parameters of the server as the recommendation parameters corresponding to the account information, while in another embodiment, the recommendation parameters corresponding to the account information can also be other recommendation parameters. For example, other recommendation parameters are used to indicate that other account information similar to the account information performs operations on the recommended multimedia resources. In this case, there is no need to execute 301, and other methods can be adopted to obtain the account information to be recommended and the recommendation parameters corresponding to the account information.

[0230] 302. For each multimedia resource to be recommended, the central server calls the feature extraction sub-model in the recommendation model, fuses the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource, and obtains a first fused feature.

[0231] Among them, the account feature is used to characterize the account information of the target account, the multimedia resource feature is used to characterize the resource information of the multimedia resource, and the resource information of the multimedia resource is used to describe the corresponding multimedia resource. In some embodiments, the resource information includes the introduction information of the multimedia resource or the resource recommendation parameter of the multimedia resource, and the resource recommendation parameter is used to describe the recommendation of the multimedia resource, and the resource recommendation parameter is determined based on the operation performed by at least one account on the multimedia resource. For example, the resource recommendation parameter includes the exposure click rate or click-through rate of the multimedia resource. For example, after the multimedia resource is recommended to at least one account, the exposure click rate of the multimedia resource can be determined based on whether the at least one account performs a click operation on the multimedia resource; after the at least one account performs a click operation on the multimedia resource, the click-through rate of the multimedia resource can be determined based on whether the at least one account performs a conversion operation on the multimedia resource.

[0232] The first fusion feature is obtained by fusing the account feature and the multimedia resource feature of the multimedia resource, that is, the first fusion feature can represent the feature of the target account and the feature of the multimedia resource.

[0233] In the disclosed embodiment, the central server stores a plurality of multimedia resources to be recommended, for example, the multimedia resources are advertisements, videos, images, etc. Based on the interaction between the central server and the server, the multimedia resources stored in the central server can be recommended to the target account logged in to the server, thereby achieving the effect that the central server provides recommendation services for other servers.

[0234] In some embodiments, 302 includes: calling the feature extraction sub-model, performing feature extraction on the account information and the resource information respectively, obtaining the account feature and the multimedia resource feature, and splicing the account feature with the multimedia resource feature to obtain the first fusion feature.

[0235] By splicing the account feature with the multimedia resource feature, the spliced ​​first fusion feature includes both the account feature and the multimedia resource feature, which facilitates the subsequent determination of the target parameters of the multimedia resource based on the first fusion feature.

[0236] In some embodiments, the first fused feature is obtained by fusing account features, multimedia resource features and associated features, that is, the process of obtaining the first fused feature includes: calling the feature extraction sub-model, obtaining associated features based on account information and resource information of multimedia resources, calling the feature extraction sub-model, fusing account features, multimedia resource features and associated features, and obtaining the first fused feature.

[0237] Among them, the association feature indicates the interaction between the terminal corresponding to the target account and the multimedia resource, and also indicates the interaction between the terminal corresponding to the target account and other multimedia resources, and the other multimedia resources are multimedia resources recommended by this terminal and other than the multimedia resource. Since the association feature not only indicates the interaction between the target account and the multimedia resource, but also indicates the interaction between the target account and other multimedia resources, the information contained in the first fusion feature is enriched by introducing the association feature to ensure the accuracy of the target parameters subsequently obtained based on the first fusion feature, thereby ensuring the recommendation effect.

[0238] For example, the association feature can not only indicate whether the terminal corresponding to the target account has performed an operation on the multimedia resource, but also indicate the situation in which the terminal corresponding to the target account has performed operations on other multimedia resources recommended by the central server for the target account. For example, the association feature can reflect the exposure click-through rate or click-through conversion rate corresponding to the target account. For example, the central server has recommended 10 multimedia resources to the target account, and the terminal corresponding to the target account has performed click operations on 6 multimedia resources, that is, the exposure click-through rate corresponding to the target account is 0.6. Among these 6 multimedia resources, the terminal corresponding to the target account has performed conversion operations on 3 of them, that is, the click-through rate corresponding to the target account is 0.5.

[0239] In some embodiments, the process of obtaining the first fusion feature includes: obtaining related information based on the account information and the resource information of the multimedia resource, calling the feature extraction sub-model, performing feature extraction on the account information, the resource information of the multimedia resource and the related information respectively to obtain account features, multimedia resource features and related features, and splicing the account features, the multimedia resource features and the related features to obtain the first fusion feature.

[0240] The associated information is used to describe the interaction between the terminal corresponding to the target account and the multimedia resource, and can also describe the interaction between the terminal corresponding to the target account and other multimedia resources. For example, the associated information includes information describing whether the target account has performed a click operation or a conversion operation on the multimedia resource, or the click rate or conversion rate of the target account on other multimedia resources recommended in the past, or the click rate and conversion rate of the target account on multimedia resources of different types, or the click rate or conversion rate of multimedia resources on which the target account has performed operations, etc.

[0241] 303. The central server calls the weight acquisition sub-model in the recommendation model, performs feature conversion on the recommendation parameter feature of the recommendation parameter, and obtains the weight feature of the first fusion feature.

[0242] In the disclosed embodiment, the recommendation model includes a weight acquisition sub-model, which is used to acquire a weight feature of the first fusion feature, and the weight feature can indicate the influence of the first fusion feature on the target parameter of the multimedia resource recommended by the server in the server. The recommendation parameter feature is used to characterize the recommendation parameter, which is used to reflect the situation that the server recommends multimedia resources for the account logged in to the server.

[0243] In some embodiments, 303 includes: the central server calls the feature extraction sub-model in the recommendation model, performs feature extraction on the recommendation parameter to obtain the recommendation parameter feature, calls the weight acquisition sub-model, performs feature conversion on the recommendation parameter feature to obtain the weight feature.

[0244] The recommended parameters are input into the feature extraction sub-model, and the feature extraction sub-model outputs the recommended parameter features, and the recommended parameter features are input into the weight acquisition sub-model to obtain the weight features.

[0245] In some embodiments, the recommendation model also includes another feature extraction sub-model. For ease of distinction, the feature extraction sub-model used to obtain the first fusion feature is called the first feature extraction sub-model, and the other feature extraction sub-model is called the second feature extraction sub-model. Then 303 includes: calling the second feature extraction sub-model, performing feature extraction on the recommendation parameter to obtain the recommendation parameter feature, calling the weight acquisition sub-model, performing feature conversion on the recommendation parameter feature to obtain the weight feature.

[0246] In the embodiment of the present disclosure, the recommendation model includes a first feature extraction sub-model and a second feature extraction sub-model, the first feature extraction sub-model is used to obtain the first fusion feature, and the second feature extraction sub-model is used to obtain the recommended parameter feature of the recommended parameter.

[0247] 304. The central server calls the click prediction sub-model in the recommendation model, fuses the first fused feature with the weight feature, and obtains a second fused feature.

[0248] Among them, the click prediction sub-model is used to obtain the first parameter, and the first parameter indicates the possibility that the terminal corresponding to the target account will perform a click operation on the multimedia resource when the multimedia resource is recommended to the target account. In the embodiment of the present disclosure, the first parameter also indicates the possibility that the terminal corresponding to the target account will perform a click operation on the multimedia resource when the server recommends the multimedia resource to the target account. The click operation is a viewing operation performed by the terminal corresponding to the target account on the multimedia resource. After performing the click operation on the multimedia resource, the terminal can display the multimedia resource for the user corresponding to the target account to view. For example, the first parameter is the exposure click-through rate, and the click prediction model is used to obtain the exposure click-through rate. For example, the first parameter is PCTR (Predict Click-Through Rate).

[0249] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n. The above process of acquiring the second fusion feature based on the weight acquisition submodel and the click prediction submodel includes the following steps 1 to 4.

[0250] Step 1: Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter features to obtain the i-th weight feature.

[0251] In the disclosed embodiments, the input of each weight acquisition layer is the same, which is the recommended parameter feature, and different weight acquisition layers output different weight features. In some embodiments, the feature sizes of the weight features output by different weight acquisition layers are different. For example, from the 1st weight feature to the nth weight feature, the feature size of the weight feature gradually decreases.

[0252] In some embodiments, the i-th weight feature satisfies the following relationship:

[0253] P i =tanh(Embedding(context)W i )

[0254] Among them, P i represents the i-th weight feature, tanh(·) represents the hyperbolic tangent function, Embedding(context) represents the recommended parameter feature, W i Represents the transformation matrix of the i-th weight acquisition layer.

[0255] Step 2: Based on the first first parameter acquisition layer, perform dimensionality reduction processing on the first fusion feature to obtain the first first dimensionality reduction feature.

[0256] In the disclosed embodiment, the input of the first first parameter acquisition layer of the click prediction sub-model is the first fused feature, and the feature size of the first first dimensionality reduction feature is smaller than the feature size of the first fused feature. For example, the first parameter acquisition layer is a convolution layer, and a small-sized feature, namely the first first dimensionality reduction feature, is obtained by convolution processing the first fused feature.

[0257] Step 3: Based on the first fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain the first third fusion feature.

[0258] In some embodiments, the feature size of the third fused feature output by the first first fusion layer, the feature size of the first dimension reduction feature, and the feature size of the first weight feature are the same. Then step 3 includes: based on the first first fusion layer, determining the product of the first dimension reduction feature and the first weight feature as the first third fused feature.

[0259] For example, the first dimensionality reduction feature includes multiple eigenvalues, and the first weight feature also includes multiple eigenvalues. Based on the first first fusion layer, the product of the eigenvalues ​​at the same position in the first dimensionality reduction feature and the first weight feature is determined, and the multiple products obtained constitute the first third fusion feature.

[0260] Step 4: For the j-th first parameter acquisition layer and the j-th first fusion layer, based on the j-th first parameter acquisition layer, perform dimensionality reduction processing on the j-1-th third fusion feature to obtain the j-th first dimensionality reduction feature; based on the j-th first fusion layer, fuse the j-th first dimensionality reduction feature with the j-th weight feature to obtain the j-th third fusion feature; repeat the above steps until the second fusion feature is obtained based on the n-th first fusion layer, where j is an integer greater than 1 and not greater than n.

[0261] In the disclosed embodiment, the input of the j-th first parameter acquisition layer is the j-1th third fusion feature output by the j-1th first fusion layer. Each first parameter acquisition layer performs dimensionality reduction processing on the input feature to obtain a first dimensionality reduction feature. Similarly to the above steps 2 and 3, the j-th third fusion feature is obtained based on the j-th first parameter acquisition layer and the j-th first fusion layer. The above steps are repeated to obtain the second fusion feature input by the n-th fusion layer.

[0262] In the disclosed embodiment, the weight acquisition submodel includes multiple weight acquisition layers, different weight acquisition layers are used to acquire weights representing features of different dimensions, and the click prediction submodel includes the same number of first parameter acquisition layers and first fusion layers. Based on the first parameter acquisition layer and the first fusion layer included in the click prediction submodel, multiple weight features are fused with the first fusion feature to enhance the features of the first fusion feature in each dimension, so that the obtained second fusion feature matches the characteristics of the server, thereby ensuring the accuracy of the second fusion feature.

[0263] 305. The central server calls the click prediction sub-model in the recommendation model, and fuses the second fusion feature with the recommendation parameter feature to obtain the first parameter.

[0264] In the disclosed embodiment, the weight feature is obtained through the recommended parameter feature of the server, and the weight feature is equivalent to the weight corresponding to the server. The first fusion feature is first fused with the weight feature, and then the fused second weight feature is fused with the recommended parameter feature, so that the first parameter incorporates the characteristics of the server itself to ensure that the obtained first parameter matches the server, thereby ensuring the accuracy of the first parameter, so that the effect of personalized recommendation for the target account can be achieved in the server later.

[0265] In some embodiments, the click prediction sub-model includes a parameter output layer, and 305 includes: based on the parameter output layer, fusing the second fusion feature with the recommended parameter feature to obtain the first parameter.

[0266] In some embodiments, 305 includes: the central server calls the click prediction sub-model, splices the second fusion feature with the recommendation parameter feature to obtain a spliced ​​feature, and performs feature conversion on the spliced ​​feature to obtain the first parameter.

[0267] The second fused feature is usually concatenated with the recommended parameter feature, and the concatenated feature is subjected to feature transformation to enrich the features contained in the concatenated feature, so as to capture the correlation between the second fused feature and the recommended parameter feature corresponding to the server, and ensure the accuracy of the first parameter.

[0268] In some embodiments, the second fusion feature, the recommended parameter feature and the first parameter satisfy the following relationship:

[0269] y=w T ·Concatenate[x, Embedding(context)]+b

[0270] Wherein, y represents the first parameter, w represents the transformation matrix in the click prediction sub-model, T represents the transposition of the matrix, Concatenate[·] represents the concatenation of features; x represents the second fused feature, Embedding(context) represents the recommended parameter feature, and b represents the transformation parameter in the click prediction sub-model.

[0271] It should be noted that, in the embodiment of the present disclosure, the first fusion feature is first fused with the weight feature, and then the second fusion feature is fused with the recommendation parameter feature to obtain the first parameter. In another embodiment, there is no need to execute steps 304-305, and other methods can be adopted to call the click prediction sub-model to obtain the first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature.

[0272] 306. The central server calls the conversion prediction sub-model to fuse the first fused feature with the weight feature to obtain a fourth fused feature.

[0273] Among them, the conversion prediction sub-model is used to obtain the second parameter, and the second parameter indicates the possibility that when the multimedia resource is recommended to the target account, the terminal corresponding to the target account performs a click operation on the multimedia resource and then performs a conversion operation. In the embodiment of the present disclosure, the second parameter also indicates the possibility that when the server recommends the multimedia resource to the target account, the terminal corresponding to the target account performs a click operation on the multimedia resource and then performs a conversion operation. For example, the second parameter is the click-through conversion rate, and the conversion prediction model is used to obtain the click-through conversion rate, that is, predicting the possibility that after the multimedia resource is recommended to the target account, the terminal corresponding to the target account performs a click operation on the multimedia resource and then performs a conversion operation. For example, the conversion operation is a registration operation, an order operation, etc. For example, after the terminal corresponding to the target account performs a click operation on the multimedia resource, the terminal corresponding to the target account performs a registration operation, an order operation, or an activation operation on the multimedia resource. For example, the second parameter is PCVR (Predict Conversion Rate).

[0274] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n. The above process of acquiring the fourth fusion feature based on the weight acquisition submodel and the conversion prediction submodel includes the following steps 5 to 8.

[0275] Step 5: Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter features to obtain the i-th weight feature.

[0276] Step 6: Based on the first second parameter acquisition layer, perform dimensionality reduction processing on the first fusion feature to obtain the first second dimensionality reduction feature.

[0277] Step 7: Based on the first second fusion layer, the first dimension reduction feature is fused with the first weight feature to obtain the first fifth fusion feature.

[0278] Step 8. For the j-th second parameter acquisition layer and the j-th second fusion layer, based on the j-th second parameter acquisition layer, perform dimensionality reduction processing on the j-1-th fifth fusion feature to obtain the j-th second reduced dimensionality feature. Based on the j-th second fusion layer, fuse the j-th second reduced dimensionality feature with the j-th weight feature to obtain the j-th fifth fusion feature. Repeat the above steps until the fourth fusion feature is obtained based on the n-th second fusion layer, where j is an integer greater than 1 and not greater than n.

[0279] It should be noted that the above steps 5-8 are similar to the above steps 1-4 and will not be repeated here.

[0280] In the disclosed embodiment, the weight acquisition submodel includes a plurality of weight acquisition layers, and different weight acquisition layers are used to acquire weights representing features of different dimensions, and the conversion prediction submodel includes the same number of second parameter acquisition layers and second fusion layers. Based on the second parameter acquisition layer and the second fusion layer included in the conversion prediction submodel, the plurality of weight features are fused with the first fusion feature to enhance the features of the first fusion feature in each dimension, so that the obtained fourth fusion feature matches the characteristics of the server, thereby ensuring the accuracy of the fourth fusion feature.

[0281] 307. The central server calls the conversion prediction sub-model to fuse the fourth fusion feature with the recommendation parameter feature to obtain the second parameter.

[0282] In the disclosed embodiment, the weight feature is obtained through the recommended parameter feature of the server. The first fusion feature is first fused with the weight feature, and then the fourth weight feature obtained by fusion is fused with the recommended parameter feature, so that the first parameter incorporates the characteristics of the server itself to ensure that the obtained second parameter matches the server, thereby ensuring the accuracy of the second parameter, so that the effect of personalized recommendation for the target account can be achieved in the server later.

[0283] In some embodiments, the conversion prediction sub-model includes a parameter output layer, and 305 includes: based on the parameter output layer, fusing the fourth fusion feature with the recommended parameter feature to obtain a second parameter.

[0284] In some embodiments, 307 includes: the central server calls the conversion prediction sub-model, splices the fourth fusion feature with the recommended parameter feature to obtain a spliced ​​feature, and performs feature conversion on the spliced ​​feature to obtain the second parameter.

[0285] It should be noted that the embodiment of the present disclosure first fuses the first fusion feature with the weight feature, and then fuses the fourth fusion feature with the recommended parameter feature to obtain the second parameter. In another embodiment, there is no need to execute steps 306-307, and other methods can be adopted to call the conversion prediction sub-model to obtain the second parameter of the multimedia resource based on the recommended parameter feature and the first fusion feature.

[0286] 308. The central server calls the fusion sub-model in the recommendation model, fuses the first parameter with the second parameter, and obtains the target parameter of the multimedia resource.

[0287] Among them, the target parameter indicates the possibility that the terminal corresponding to the target account performs an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information. In the embodiment of the present disclosure, the target parameter also indicates the possibility that the terminal corresponding to the target account performs an operation on the multimedia resource when the server recommends the corresponding multimedia resource to the target account. For example, the target parameter is the exposure conversion rate. For example, the target parameter is PCTCVR (PredictClick-Through Conversion Rate).

[0288] Since the first parameter indicates the possibility that the terminal corresponding to the target account performs a click operation on the multimedia resource when the server recommends the multimedia resource to the target account, and the second parameter indicates the possibility that the terminal corresponding to the target account performs the click operation and then performs a conversion operation on the multimedia resource when the server recommends the multimedia resource to the target account, the first parameter is merged with the second parameter to obtain the possibility that the terminal corresponding to the target account performs a conversion operation on the multimedia resource when the server recommends the corresponding multimedia resource to the target account. Since the recommendation situations of the multimedia resources indicated by the first parameter and the second parameter are different and can represent different information, a multi-tasking method is adopted to first obtain the first parameter and the second parameter to enrich the obtained information, and then the first parameter and the second parameter are merged to obtain the target parameter, realizing the method of obtaining the target parameter by multiple parameters, thereby improving the accuracy of the target parameter.

[0289] In some embodiments, 308 includes: calling the fusion sub-model to determine the product of the first parameter and the second parameter as the target parameter of the multimedia resource.

[0290] In the embodiment of the present disclosure, only the acquisition of target parameters of a multimedia resource is taken as an example for explanation. According to the above steps 302 - 308 , the target parameters of each multimedia resource to be recommended can be acquired.

[0291] It should be noted that the embodiment of the present disclosure takes the recommendation model including a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model, a fusion sub-model and a weight acquisition sub-model as an example, and obtains the target parameters based on the multiple sub-models included in the recommendation, and the target parameters indicate the possibility of the terminal corresponding to the target account performing a conversion operation on the multimedia resource when the server recommends the corresponding multimedia resource to the target account. In another embodiment, there is no need to execute steps 302-308, and other methods can be adopted to obtain the target parameters of each multimedia resource based on the account information, the recommendation parameters and the resource information of the multiple multimedia resources to be recommended, and the target parameters can also indicate the possibility of the terminal corresponding to the target account performing other operations on the multimedia resource when the server recommends the corresponding multimedia resource to the target account.

[0292] 309. The central server determines at least one multimedia resource with the highest target parameter from the multiple multimedia resources as a selected target multimedia resource based on the target parameters of the multiple multimedia resources.

[0293] The target parameter of the target multimedia resource is greater than the target parameters of the remaining multimedia resources in the multiple multimedia resources. Since the target parameter can reflect the interest of the user corresponding to the target account in the multimedia resource, selecting at least one target multimedia resource with the largest target parameter from the multiple multimedia resources indicates that the user corresponding to the target account is most interested in the target multimedia resource.

[0294] In some embodiments, step 309 includes: the central server selects a target number of target multimedia resources with the highest target parameters from the multiple multimedia resources based on the target parameters of the multiple multimedia resources. The target number is an arbitrary number, for example, the target number is 3 or 5. For example, the target parameters of the multiple multimedia resources are arranged in descending order, and the multimedia resources with the target number that are ranked first are selected from the multiple multimedia resources as the target multimedia resources.

[0295] 310. The central server sends the target multimedia resource to the server, and the server recommends the target multimedia resource to the target account.

[0296] The central server sends the target multimedia resource to the server, so that the server sends the target multimedia resource to the target account.

[0297] In some embodiments, step 310 includes: the central server sends the target multimedia resource to the server, the server receives the target multimedia resource, recommends the target multimedia resource to the terminal corresponding to the target account, the terminal receives the target multimedia resource, and displays the target multimedia resource. The target multimedia resource is recommended to the terminal corresponding to the target account by the server, and the terminal displays the target multimedia resource so that the user can view the target multimedia resource.

[0298] In some embodiments, after step 310, the method further includes: the server obtains the recommendation result of the target account, sends the recommendation result to the central server, the central server receives the recommendation result, and stores the recommendation result in correspondence with the target account and the server identifier.

[0299] The recommendation result indicates the actual situation of the terminal corresponding to the target account performing operations on the target multimedia resource after the server recommends the target multimedia resource to the target account, that is, whether the terminal corresponding to the target account performs operations on the target multimedia resource, and if the operations are performed, what kind of operations are performed, such as click operations or conversion operations. Therefore, the recommendation results can reflect the recommendation effect of the multimedia resource. The central server receives the recommendation results and stores them in correspondence with the target account and server identifier, so as to optimize the recommendation model based on the stored recommendation results later.

[0300] It should be noted that the embodiment of the present disclosure is described by taking the example of a central server providing recommendation services for multiple servers. In another embodiment, the account referred to by the account information is the account for logging into the central server. In this case, there is no need to execute steps 301 and 310. The central server obtains the account information and the recommendation parameters of the central server, and determines the target multimedia resources that the target account referred to by the account information is interested in according to the above steps 302-309. Thereafter, the central server recommends the target multimedia resources to the target account.

[0301] The method provided by the embodiment of the present disclosure provides a way to provide recommendation services for multiple servers, introduces a recommendation parameter corresponding to account information, and the recommendation parameter can reflect the recommendation of multimedia resources matching the account information. When recommending multimedia resources for an account logged in to any server, based on the account information, the recommendation parameter of the server and the resource information of multiple multimedia resources to be recommended, the target parameter of each multimedia resource is determined so that the target parameter matches the target account and the server. The target parameter can indicate the degree of interest of the target account in the multimedia resource, so that the target multimedia resource that the target account is most interested in among the multiple multimedia resources can be recommended to the target account, and a personalized recommendation scheme can be implemented for the account in the server, thereby ensuring the recommendation effect.

[0302] Furthermore, in the disclosed embodiment, the target parameters of each multimedia resource are acquired through a recommendation model to ensure the accuracy of the target parameters, and the multimedia resources are subsequently recommended based on the target parameters, thereby ensuring the recommendation effect.

[0303] In addition, the multiple sub-models included in the recommendation model are used to implement a method of obtaining the target parameters by multi-tasking, that is, first obtaining the first parameter and the second parameter to enrich the obtained information, and then fusing the first parameter with the second parameter to obtain the target parameter, thereby realizing a method of obtaining the target parameter by multiple parameters, thereby improving the accuracy of the target parameter.

[0304] Based on the above Figure 3 The embodiment shown is a schematic diagram of a structure of a recommendation model according to an exemplary embodiment, see Figure 4 The recommendation model includes a first feature extraction sub-model, a second feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model, a weight acquisition sub-model and a fusion sub-model.

[0305] The first feature extraction submodel is used to extract features from account information, resource information of multimedia resources and related information respectively, and fuse the extracted features to obtain the first fused features. The second feature extraction submodel is used to extract features from the recommended parameters of the server to obtain recommended parameter features.

[0306] The click prediction sub-model includes n first parameter acquisition layers, n first fusion layers and parameter output layers. The conversion prediction sub-model includes n second parameter acquisition layers, n second fusion layers and parameter output layers. The weight acquisition sub-model includes n weight acquisition layers, each weight acquisition layer is used to acquire a weight feature.

[0307] For the click prediction sub-model, the first parameter acquisition layer inputs the first fusion feature, and the output first first dimensionality reduction feature is input into the first first fusion layer. The first first fusion layer fuses the input first first dimensionality reduction feature and the first weight feature to obtain the first third fusion feature. The first third fusion feature is input into the second first parameter acquisition layer. After that, based on the remaining first parameter acquisition layers, the first fusion layer, and the remaining weight features, in the above manner, until the second fusion feature output by the nth first fusion layer, based on the parameter output layer, the second fusion feature is fused with the recommended parameter feature to obtain the first parameter.

[0308] For the conversion prediction sub-model, similar to the above-mentioned click prediction sub-model, after the nth second fusion layer outputs the fourth fusion feature, the fourth fusion feature is fused with the recommended parameter feature based on the parameter output layer in the click prediction sub-model to obtain the second parameter.

[0309] The fusion sub-model is used to fuse the first parameter output by the click prediction sub-model with the second parameter output by the conversion prediction sub-model to obtain the target parameter.

[0310] Based on the above Figure 3 The embodiment shown is a flowchart of an advertisement recommendation according to an exemplary embodiment, see Figure 5 , the process includes the following steps.

[0311] 501. The terminal sends a login request to a server corresponding to the target application based on the target application, and the login request carries the target account.

[0312] The server provides services for the target application.

[0313] 502. The server receives the login request and verifies whether the target account is an account registered in the server. If the verification is successful, a resource acquisition request is sent to the central server. The resource recommendation request carries the account information of the target account that logs in to the server and the recommended parameters of the server, and a verification success notification is sent to the terminal.

[0314] In the disclosed embodiment, a central server stores multiple advertisements to be recommended. The central server can provide advertisement recommendation services for multiple servers. Each server provides services for an application in which advertisements can be displayed. Different servers correspond to different applications. For example, among the applications corresponding to multiple servers, one application is a navigation application, one application is a weather application, and another application is a video sharing application.

[0315] 503. The terminal receives a notification of successful verification and displays a login success interface based on the target application.

[0316] 504. The central server receives the resource recommendation request sent by the server, and obtains target parameters of each advertisement according to the above steps 302-308 for multiple advertisements to be recommended.

[0317] 505. The central server determines at least one advertisement with the highest target parameter from the multiple advertisements as a target advertisement based on the target parameters of the multiple advertisements, and sends the target advertisement to the server.

[0318] 506. The server receives the target advertisement sent by the central server, and sends a recommendation of the target advertisement to the terminal corresponding to the target account.

[0319] 507. The terminal receives the target advertisement sent by the server, and displays the target advertisement based on the target application.

[0320] The recommendation model provided in the embodiments of the present disclosure has a better recommendation effect than the recommendation model in the related art, can expand the revenue brought by the recommended advertisements, and can reduce the number of recommended advertisements while ensuring the revenue, thereby saving advertisement spaces for placing advertisements, so that more advertisements can be placed, thereby increasing the revenue obtained.

[0321] The method provided by the embodiment of the present disclosure can be applied in the scenario of alliance, where the central server provides recommendation services for servers corresponding to multiple applications, and based on the central server, advertisements can be placed in applications corresponding to different servers. Since different servers have different recommendation parameters, when recommending advertisements for accounts logged in to different servers, the advertisements recommended in different servers may be different, which can avoid the problem of uneven quality of advertisement traffic in multiple servers, resulting in poor advertisement recommendation effects, thereby achieving personalized recommendations for each server.

[0322] In the above Figure 3 On the basis of the illustrated embodiment, before obtaining the target parameters based on the recommendation model, the recommendation model needs to be trained. Taking the recommendation model including a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model as an example, the training process is detailed in the following embodiment.

[0323] Figure 6 is a flowchart of a recommendation model training method according to an exemplary embodiment. Figure 6 , the method comprises the following steps.

[0324] 601. The central server obtains sample data and sample labels corresponding to the sample data.

[0325] The sample data includes sample account information, sample resource information of sample multimedia resources, and sample recommendation parameters corresponding to the sample account information. The sample tag indicates whether the terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account. In the embodiment of the present disclosure, the sample account information refers to the sample account logged in to the sample server, and the recommendation parameters of the sample server are used as the sample recommendation parameters corresponding to the sample account information. The sample recommendation parameters are determined based on the operation performed by at least one account logged in to the sample server on the multimedia resource recommended by the sample server.

[0326] In the disclosed embodiment, the sample tag includes a first sample tag and a second sample tag, the first sample tag indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account, and the second sample tag indicates whether the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account. That is, the first sample tag indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account, and the second sample tag indicates whether the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account.

[0327] In some embodiments, 601 includes: the central server obtains multiple sample data and a sample label corresponding to each sample data.

[0328] The sample servers corresponding to the multiple sample data may be different. For example, the sample servers corresponding to the multiple sample data include sample server 1 and sample server 2. The recommendation model is subsequently trained according to the multiple sample data, and the recommendation model can be called to provide resource recommendation services for the sample server 1 and the sample server 2.

[0329] 602. The central server calls the feature extraction sub-model to fuse the sample account features of the sample account information and the sample multimedia resource features of the sample resource information to obtain sample fusion features.

[0330] This step is the same as the above step 302 and will not be described again.

[0331] 603. The central server calls the click prediction sub-model to obtain a second prediction label of the sample multimedia resource based on the recommendation parameter features of the sample recommendation parameters and the sample fusion features.

[0332] Among them, the second prediction tag indicates the possibility that the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account. In the embodiment of the present disclosure, the second prediction tag also indicates the possibility that the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account. For example, the second prediction tag is the predicted exposure click rate. This step is similar to the above steps 305-305 and will not be repeated here.

[0333] 604. The central server calls the conversion prediction sub-model to obtain a third prediction label of the multimedia resource based on the sample recommendation parameter features and the sample fusion features.

[0334] Among them, the third prediction tag indicates the possibility that when the sample multimedia resource is recommended to the sample account, the terminal corresponding to the sample account performs a click operation on the sample multimedia resource and then performs a conversion operation. In the embodiment of the present disclosure, the third prediction tag also indicates the possibility that when the sample server recommends the sample multimedia resource to the sample account, the terminal corresponding to the sample account performs a click operation on the sample multimedia resource and then performs a conversion operation. For example, the third prediction tag is the predicted click-through conversion rate. This step is the same as the above steps 306-307 and will not be repeated here.

[0335] 605. The central server calls the fusion sub-model to fuse the second prediction label with the third prediction label to obtain the first prediction label.

[0336] Among them, the first prediction tag indicates the possibility that the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account. In the embodiment of the present disclosure, the first prediction tag indicates the possibility that the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account. For example, the first prediction tag is a predicted exposure conversion rate. This step is similar to the above step 308 and will not be repeated here.

[0337] It should be noted that the embodiment of the present disclosure is explained by taking the prediction tag including the first prediction tag and the second prediction tag as an example. In another embodiment, there is no need to execute steps 602-605, and other methods can be adopted to call the recommendation model to obtain the prediction tag of the sample multimedia resource corresponding to the sample data based on the sample data. The prediction tag indicates the possibility of the terminal corresponding to the sample account performing an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account.

[0338] 606. The central server trains the recommendation model based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

[0339] In an embodiment of the present disclosure, the sample tag corresponding to the sample data further includes a second sample tag, which indicates whether the terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account.

[0340] Since the first prediction label indicates the possibility that the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account, and the first sample label indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account, the difference between the first prediction label and the first sample label can reflect the accuracy of the recommendation model. In addition, since the second prediction label indicates the possibility that the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account, and the second sample label indicates whether the terminal corresponding to the sample account performs a click operation on the sample multimedia resource when the sample server recommends the sample multimedia resource to the sample account, the difference between the second prediction label and the second sample label can also reflect the accuracy of the recommendation model. The recommendation model is adjusted based on these two differences to improve the accuracy of the recommendation model.

[0341] In some embodiments, 606 includes: determining a first loss value based on a difference between a first predicted label and a first sample label of the sample data, determining a second loss value based on a difference between a second predicted label and a second sample label of the sample data, and training the recommendation model based on the sum of the first loss value and the second loss value.

[0342] For example, the first prediction tag is the predicted exposure conversion rate, the first sample tag is the exposure conversion rate, the second prediction tag is the predicted exposure click rate, and the second sample tag is the exposure click rate. The exposure conversion rate is 0 or 1, 0 means that the sample account has not performed a conversion operation on the sample multimedia resource, and 1 means that the sample account has performed a conversion operation on the sample multimedia resource; the exposure click rate is 0 or 1, 0 means that the sample account has not performed a click operation on the sample multimedia resource, and 1 means that the sample account has performed a click operation on the sample multimedia resource. The first loss value can be calculated based on the predicted exposure conversion rate and the exposure conversion rate, and the second loss value can be determined based on the predicted exposure click rate and the exposure click rate.

[0343] It should be noted that the embodiment of the present disclosure is described using one piece of sample data as an example. In another embodiment, multiple pieces of sample data and sample labels corresponding to each piece of sample data are obtained, and the recommendation model is iteratively trained multiple times according to the above steps 601-606.

[0344] It should be noted that the embodiment of the present disclosure takes the example of sample labels including a first sample label and a second sample label, and trains the recommendation model based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label. In another embodiment, there is no need to execute step 606, and other methods can be adopted to train the recommendation model based on the difference between the predicted label of the sample data and the sample label.

[0345] The method provided by the embodiment of the present disclosure, the sample data includes sample account information, sample resource information of sample multimedia resources and sample recommendation parameters of the sample server, and the recommendation model is trained based on the sample data and the corresponding sample labels to improve the accuracy of the recommendation model, so as to facilitate the subsequent call of the recommendation model and realize personalized recommendation for the server based on the recommendation parameters of the server.

[0346] Moreover, in the process of training the recommendation model, multiple sample data and sample labels corresponding to each sample data are obtained, and the multiple sample data correspond to multiple sample servers, that is, the recommendation model trained according to the multiple sample data can provide recommendation services for the multiple sample servers. Subsequently, based on the recommendation model, multimedia resources are recommended to the accounts that log in to the multiple sample servers. This can avoid the problem of poor recommendation effects due to different multimedia resource recommendation situations in multiple servers, and can also realize personalized recommendations for multiple servers.

[0347] Furthermore, since the second prediction label and the third prediction label can represent different information, a multi-task approach is adopted to first obtain the second prediction label and the third prediction label to enrich the obtained information, and then the second prediction label and the third prediction label are fused to obtain the first prediction label to ensure the accuracy of the obtained first prediction label.

[0348] Furthermore, the disclosed embodiment adopts a multi-task approach to train the recommendation model, that is, multiple tasks of obtaining the first prediction label, the second prediction label, and the third prediction label. Using multiple tasks to train the recommendation model can alleviate the problem of poor accuracy of the recommendation model due to sparse sample data, thereby improving the accuracy of the recommendation model. Furthermore, training the recommendation model based on sample data corresponding to multiple sample servers enriches the sample data for training the recommendation model, which can also improve the accuracy of the recommendation model and alleviate the problem of poor accuracy of the recommendation model due to sparse sample data.

[0349] It should be noted that the above Figure 3 The illustrated embodiment takes the recommendation model as an example and obtains the target parameters of multimedia resources based on the recommendation model. In another embodiment, the target parameters can be obtained in other ways without calling the recommendation model. The process of obtaining the target parameters is detailed in the following embodiment.

[0350] Figure 7 is a flowchart of a multimedia resource recommendation method according to an exemplary embodiment. Figure 7 The method is executed by a central server and includes the following steps.

[0351] 701. The central server obtains account information to be recommended and recommendation parameters of the server, where the account information refers to the target account of the resource to be recommended, the target account is the account logged into the server, and the recommendation parameters are determined based on the operations performed by at least one account logged into the server on the multimedia resources recommended by the server.

[0352] This step is the same as the above step 301 and will not be described again here.

[0353] 702. For each multimedia resource to be recommended, the central server fuses the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature.

[0354] In some embodiments, the first fusion feature is obtained by fusing account features, multimedia resource features and associated features, that is, the process of obtaining the first fusion feature includes: obtaining associated features based on account information and resource information of multimedia resources, the associated features indicating the interaction between the terminal corresponding to the target account and the multimedia resources, and also indicating the interaction between the terminal corresponding to the target account and other multimedia resources, other multimedia resources are multimedia resources recommended by this end and other than multimedia resources, the account features, multimedia resource features and associated features are fused to obtain the first fusion feature.

[0355] This step is the same as the above step 302 and will not be described again.

[0356] 703. The central server obtains the first parameter and the second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature of the recommendation parameter, respectively. The first parameter indicates the possibility that the terminal corresponding to the target account will click on the multimedia resource when the server recommends the multimedia resource to the target account, and the second parameter indicates the possibility that the terminal corresponding to the target account will click on the multimedia resource and then perform a conversion operation when the server recommends the multimedia resource to the target account.

[0357] In some embodiments, step 703 includes: performing feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature; and obtaining the first parameter and the second parameter respectively based on the recommended parameter feature, the first fusion feature and the weight feature.

[0358] This step is the same as the above steps 303-307 and will not be repeated here.

[0359] 704. The central server merges the first parameter and the second parameter to obtain a target parameter of the multimedia resource.

[0360] This step is the same as the above step 308 and will not be described again here.

[0361] In the embodiment of the present disclosure, only obtaining the target parameter of a multimedia resource is used as an example for explanation. According to the above steps 702 - 704 , the target parameter of each multimedia resource to be recommended can be obtained.

[0362] It should be noted that the embodiment of the present disclosure obtains the target parameters based on the account characteristics of the account information, the recommendation characteristics of the recommendation parameters and the multimedia resource characteristics of the multimedia resources. In another embodiment, there is no need to execute steps 702-704, and other methods can be adopted to obtain the target parameters of each multimedia resource based on the account information, the recommendation parameters and the resource information of multiple multimedia resources to be recommended.

[0363] 705. The central server determines, based on the target parameters of the multiple multimedia resources, at least one multimedia resource with the highest target parameter from the multiple multimedia resources as the target multimedia resource.

[0364] 706. The central server sends the target multimedia resource to the server, and the server recommends the target multimedia resource to the target account.

[0365] Steps 705-706 are similar to the above steps 309-310 and will not be described again here.

[0366] The method provided by the embodiment of the present disclosure provides a way to provide recommendation services for multiple servers, introduces a recommendation parameter corresponding to account information, and the recommendation parameter can reflect the recommendation of multimedia resources matching the account information. When recommending multimedia resources for an account logged in to any server, based on the account information, the recommendation parameter of the server and the resource information of multiple multimedia resources to be recommended, the target parameter of each multimedia resource is determined so that the target parameter matches the target account and the server. The target parameter can indicate the degree of interest of the target account in the multimedia resource, so that the target multimedia resource that the target account is most interested in among the multiple multimedia resources can be recommended to the target account, and a personalized recommendation scheme can be implemented for the account in the server, thereby ensuring the recommendation effect.

[0367] Figure 8 is a block diagram of a multimedia resource recommendation device according to an exemplary embodiment. Figure 8 , the device comprises:

[0368] An acquisition unit 801 is configured to acquire account information to be recommended and recommendation parameters corresponding to the account information;

[0369] The acquisition unit 801 is further configured to acquire a target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the plurality of multimedia resources to be recommended, the target parameter indicating the possibility of a terminal corresponding to the target account performing an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information;

[0370] The selection unit 802 is configured to execute target parameters based on multiple multimedia resources, and determine at least one multimedia resource with the highest target parameter from the multiple multimedia resources as the target parameter;

[0371] The recommendation unit 803 is configured to recommend the target multimedia resource to the target account.

[0372] In some embodiments, Fig. 9 As shown, the acquisition unit 801 includes:

[0373] The fusion subunit 8011 is configured to perform, for each multimedia resource to be recommended, fusion of the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature;

[0374] The acquisition subunit 8012 is configured to execute the recommendation parameter feature and the first fusion feature based on the recommendation parameter, and respectively acquire the first parameter and the second parameter of the multimedia resource, wherein the first parameter indicates the possibility that the terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates the possibility that the terminal corresponding to the target account performs a conversion operation after performing a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0375] The fusion subunit 8011 is further configured to perform fusion of the first parameter and the second parameter to obtain a target parameter of the multimedia resource.

[0376] In some embodiments, the acquisition subunit 8012 is configured to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature; based on the recommended parameter feature, the first fusion feature and the weight feature, the first parameter and the second parameter are respectively acquired.

[0377] In some embodiments, the acquisition unit 801 is further configured to execute resource information based on the account information and the multimedia resource, and acquire the association feature, the association feature indicating the interaction between the terminal corresponding to the target account and the multimedia resource, and also indicating the interaction between the terminal corresponding to the target account and other multimedia resources, the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources;

[0378] The fusion subunit 8011 is configured to perform fusion of the account feature, the multimedia resource feature and the associated feature to obtain a first fusion feature.

[0379] In some embodiments, the acquisition unit 801 is configured to execute the call recommendation model, and acquire the target parameter of each multimedia resource based on the account information, the recommendation parameter, and the resource information of the multiple multimedia resources to be recommended.

[0380] In some embodiments, the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model, and a fusion sub-model; Fig. 9 As shown, the acquisition unit 801 includes:

[0381] The fusion subunit 8011 is configured to execute, for each multimedia resource to be recommended, calling the feature extraction submodel, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature;

[0382] The acquisition subunit 8012 is configured to execute the calling of the click prediction submodel, and acquire a first parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature of the recommendation parameter, wherein the first parameter indicates the possibility that the terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account;

[0383] The acquisition sub-unit 8012 is further configured to execute the call conversion prediction sub-model, and acquire a second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature, the second parameter indicating the possibility that, when the multimedia resource is recommended to the target account, the terminal corresponding to the target account performs a click operation on the multimedia resource and then performs a conversion operation;

[0384] The fusion sub-unit 8011 is further configured to execute the call of the fusion sub-model, fuse the first parameter with the second parameter, and obtain the target parameter of the multimedia resource.

[0385] In some embodiments, the recommendation model also includes a weight acquisition sub-model, such as Fig. 9 As shown, the device also includes:

[0386] The conversion unit 804 is configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommended parameter feature, and obtain the weight feature of the first fusion feature;

[0387] The acquisition subunit 8012 is configured to execute calling the click prediction sub-model, fuse the first fusion feature with the weight feature to obtain the second fusion feature; call the click prediction sub-model, fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

[0388] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; the conversion unit 804 is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature;

[0389] The acquisition subunit 8012 is configured to perform dimensionality reduction processing on the first fused feature based on the 1st first parameter acquisition layer to obtain the 1st first reduced dimension feature; based on the 1st first fusion layer, the 1st reduced dimension feature is fused with the 1st weight feature to obtain the 1st third fused feature; for the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, the j-1th third fused feature is performed dimensionality reduction processing to obtain the jth first reduced dimension feature, based on the jth first fusion layer, the jth first reduced dimension feature is fused with the jth weight feature to obtain the jth third fused feature, and the above steps are repeated until the second fused feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

[0390] In some embodiments, the recommendation model also includes a weight acquisition sub-model, such as Fig. 9 As shown, the device also includes:

[0391] The conversion unit 804 is configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommended parameter feature, and obtain the weight feature of the first fusion feature;

[0392] The acquisition subunit 8012 is configured to execute calling the conversion prediction sub-model, fuse the first fusion feature with the weight feature to obtain the fourth fusion feature; call the conversion prediction sub-model, fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

[0393] In some embodiments, the weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; the conversion unit 804 is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature;

[0394] The acquisition subunit 8012 is configured to perform dimensionality reduction processing on the first fused feature based on the first second parameter acquisition layer to obtain the first second reduced dimension feature; based on the first second fusion layer, fuse the first reduced dimension feature with the first weight feature to obtain the first fifth fused feature; for the jth second parameter acquisition layer and the jth second fusion layer, based on the jth second parameter acquisition layer, perform dimensionality reduction processing on the j-1th fifth fused feature to obtain the jth second reduced dimension feature, based on the jth second fusion layer, fuse the jth second reduced dimension feature with the jth weight feature to obtain the jth fifth fused feature, and repeat the above steps until the fourth fused feature is obtained based on the nth second fusion layer, where j is an integer greater than 1 and not greater than n.

[0395] In some embodiments, the acquisition unit 801 is further configured to execute the call feature extraction sub-model, and acquire the associated features based on the account information and the resource information of the multimedia resources, wherein the associated features indicate the interaction between the terminal corresponding to the target account and the multimedia resources, and also indicate the interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources;

[0396] The fusion sub-unit 8011 is configured to execute the call feature extraction sub-model, fuse the account features, multimedia resource features and related features, and obtain the first fusion feature.

[0397] In some embodiments, Fig. 9 As shown, the device also includes:

[0398] The acquisition unit 801 is further configured to acquire sample data and a sample tag corresponding to the sample data, wherein the sample data includes sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, and the sample tag indicates whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0399] The acquisition unit 801 is further configured to execute the call recommendation model, and acquire the prediction label of the sample multimedia resource corresponding to the sample data based on the sample data, where the prediction label indicates the possibility of the terminal corresponding to the sample account performing an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account;

[0400] The training unit 805 is configured to train the recommendation model based on the predicted labels of the sample data and the differences between the sample labels.

[0401] In some embodiments, the sample tag includes a first sample tag, which indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account, and the prediction tag includes a first prediction tag, which indicates the possibility of the terminal corresponding to the sample account performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the acquisition unit 801 is configured to execute the call feature extraction sub-model, fuse the sample account feature of the sample account information and the sample multimedia resource feature of the sample resource information, and obtain the sample fusion feature ; Call the click prediction sub-model, and based on the recommendation parameter features of the sample recommendation parameters and the sample fusion features, obtain the second prediction label of the sample multimedia resource, where the second prediction label indicates the possibility that the terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; Call the conversion prediction sub-model, and based on the sample recommendation parameter features and the sample fusion features, obtain the third prediction label of the multimedia resource, where the third prediction label indicates the possibility that the terminal corresponding to the sample account will perform a click operation on the sample multimedia resource and then perform a conversion operation when the sample multimedia resource is recommended to the sample account; Call the fusion sub-model to fuse the second prediction label with the third prediction label to obtain the first prediction label.

[0402] In some embodiments, the sample label also includes a second sample label, which indicates whether the terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the training unit 805 is configured to perform training on the recommendation model based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

[0403] In some embodiments, the recommendation parameter is determined based on an operation performed by at least one account logged into a server on a multimedia resource recommended by the server, the server being the server logged into by the target account;

[0404] The recommendation unit 803 is configured to send the target multimedia resource to the server, and the server recommends the target multimedia resource to the target account.

[0405] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0406] In an exemplary embodiment, an electronic device is also provided, characterized in that the electronic device includes:

[0407] processor;

[0408] a memory for storing processor-executable instructions;

[0409] The processor is configured to execute the steps performed by the electronic device in the above-mentioned multimedia resource recommendation method.

[0410] In some embodiments, the electronic device is provided as a terminal. Fig.10 1 is a block diagram of a terminal 1000 according to an exemplary embodiment. The terminal 1000 may be a portable mobile terminal, such as a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer or a desktop computer. The terminal 1000 may also be referred to as a user device, a portable terminal, a laptop terminal, a desktop terminal or other names.

[0411] The terminal 1000 includes: a processor 1001 and a memory 1002 .

[0412] The processor 1001 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1001 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1001 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1001 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1001 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0413] The memory 1002 may include one or more computer-readable storage media, which may be non-transitory. The memory 1002 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1002 is used to store at least one program code, which is used to be executed by the processor 1001 to implement the multimedia resource recommendation method provided in the method embodiment of the present disclosure.

[0414] In some embodiments, the terminal 1000 may also optionally include: a peripheral device interface 1003 and at least one peripheral device. The processor 1001, the memory 1002 and the peripheral device interface 1003 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 1003 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1004, a display screen 1005, a camera assembly 1006, an audio circuit 1007, a positioning assembly 1008 and a power supply 1009.

[0415] The peripheral device interface 1003 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1001 and the memory 1002. In some embodiments, the processor 1001, the memory 1002, and the peripheral device interface 1003 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1001, the memory 1002, and the peripheral device interface 1003 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0416] The radio frequency circuit 1004 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1004 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1004 converts the electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1004 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The radio frequency circuit 1004 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1004 may also include circuits related to NFC (Near Field Communication), which is not limited in the present disclosure.

[0417] The display screen 1005 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1005 is a touch display screen, the display screen 1005 also has the ability to collect touch signals on the surface or above the surface of the display screen 1005. The touch signal can be input to the processor 1001 as a control signal for processing. At this time, the display screen 1005 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1005 can be one, arranged on the front panel of the terminal 1000; in other embodiments, the display screen 1005 can be at least two, respectively arranged on different surfaces of the terminal 1000 or in a folding design; in other embodiments, the display screen 1005 can be a flexible display screen, arranged on a curved surface or a folding surface of the terminal 1000. Even, the display screen 1005 can also be arranged as a non-rectangular irregular figure, that is, a special-shaped screen. The display screen 1005 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0418] The camera assembly 1006 is used to capture images or videos. Optionally, the camera assembly 1006 includes a front camera and a rear camera. The front camera is arranged on the front panel of the terminal, and the rear camera is arranged on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1006 may also include a flash. The flash can be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0419] The audio circuit 1007 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 1001 for processing, or input them into the radio frequency circuit 1004 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 1000. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 1001 or the radio frequency circuit 1004 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 1007 may also include a headphone jack.

[0420] The positioning component 1008 is used to locate the current geographical location of the terminal 1000 to implement navigation or LBS (Location Based Service). The positioning component 1008 can be a positioning component based on the US GPS (Global Positioning System), China's Beidou system or Russia's Galileo system.

[0421] The power supply 1009 is used to power various components in the terminal 1000. The power supply 1009 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 1009 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0422] In some embodiments, the terminal 1000 further includes one or more sensors 1010 , including but not limited to: an acceleration sensor 1011 , a gyroscope sensor 1012 , a pressure sensor 1013 , a fingerprint sensor 1014 , an optical sensor 1015 , and a proximity sensor 1016 .

[0423] The acceleration sensor 1011 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal 1000. For example, the acceleration sensor 1011 can be used to detect the components of gravity acceleration on the three coordinate axes. The processor 1001 can control the display screen 1005 to display the user interface in a horizontal view or a vertical view according to the gravity acceleration signal collected by the acceleration sensor 1011. The acceleration sensor 1011 can also be used to collect motion data of games or users.

[0424] The gyro sensor 1012 can detect the body direction and rotation angle of the terminal 1000, and the gyro sensor 1012 can cooperate with the acceleration sensor 1011 to collect the user's 3D actions on the terminal 1000. The processor 1001 can implement the following functions based on the data collected by the gyro sensor 1012: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0425] The pressure sensor 1013 can be set on the side frame of the terminal 1000 and / or the lower layer of the display screen 1005. When the pressure sensor 1013 is set on the side frame of the terminal 1000, it can detect the user's holding signal of the terminal 1000, and the processor 1001 performs left and right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor 1013. When the pressure sensor 1013 is set on the lower layer of the display screen 1005, the processor 1001 controls the operability controls on the UI interface according to the user's pressure operation on the display screen 1005. The operability controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0426] The fingerprint sensor 1014 is used to collect the user's fingerprint, and the processor 1001 identifies the user's identity based on the fingerprint collected by the fingerprint sensor 1014, or the fingerprint sensor 1014 identifies the user's identity based on the collected fingerprint. When the user's identity is identified as a trusted identity, the processor 1001 authorizes the user to perform relevant sensitive operations, including unlocking the screen, viewing encrypted information, downloading software, paying, and changing settings. The fingerprint sensor 1014 can be set on the front, back, or side of the terminal 1000. When a physical button or a manufacturer logo is set on the terminal 1000, the fingerprint sensor 1014 can be integrated with the physical button or the manufacturer logo.

[0427] The optical sensor 1015 is used to collect the ambient light intensity. In one embodiment, the processor 1001 can control the display brightness of the display screen 1005 according to the ambient light intensity collected by the optical sensor 1015. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1005 is increased; when the ambient light intensity is low, the display brightness of the display screen 1005 is reduced. In another embodiment, the processor 1001 can also dynamically adjust the shooting parameters of the camera assembly 1006 according to the ambient light intensity collected by the optical sensor 1015.

[0428] The proximity sensor 1016, also called a distance sensor, is disposed on the front panel of the terminal 1000. The proximity sensor 1016 is used to collect the distance between the user and the front of the terminal 1000. In one embodiment, when the proximity sensor 1016 detects that the distance between the user and the front of the terminal 1000 is gradually decreasing, the processor 1001 controls the display screen 1005 to switch from the screen-on state to the screen-off state; when the proximity sensor 1016 detects that the distance between the user and the front of the terminal 1000 is gradually increasing, the processor 1001 controls the display screen 1005 to switch from the screen-off state to the screen-on state.

[0429] Those skilled in the art will understand that Fig.10 The structure shown in the figure does not constitute a limitation on the terminal 1000, and the terminal 1000 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0430] In some embodiments, the electronic device is provided as a server. Fig.11 This is a schematic diagram of the structure of a server according to an exemplary embodiment. The server 1100 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 1101 and one or more memories 1102, wherein the memory 1102 stores at least one program code, and the at least one program code is loaded and executed by the processor 1101 to implement the methods provided by the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for implementing device functions, which will not be repeated here.

[0431] In an exemplary embodiment, a non-transitory computer-readable storage medium is also provided. When instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the steps executed by the electronic device in the above-mentioned multimedia resource recommendation method.

[0432] In an exemplary embodiment, a computer program product is further provided. When instructions in the computer program product are executed by a processor of an electronic device, the electronic device is enabled to execute the steps executed by the electronic device in the above-mentioned multimedia resource recommendation method.

[0433] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0434] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A multimedia resource recommendation method, characterized in that: The method comprises: Acquire the account information to be recommended and the recommendation parameters of the server, wherein the account information is used to describe the target account for logging into the server, and the recommendation parameters can reflect the operation performed by each account for logging into the server on the multimedia resources recommended by the server; For each of the plurality of multimedia resources to be recommended, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature; Performing feature conversion on the recommended parameter features of the recommended parameters to obtain weight features of the first fusion features; Based on the recommendation parameter feature, the first fusion feature, and the weight feature, respectively obtaining a target parameter for each of the multimedia resources, the target parameter indicating the possibility of a terminal corresponding to the target account performing an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information; Based on the target parameters of the plurality of multimedia resources, determining at least one multimedia resource having the highest target parameter from the plurality of multimedia resources as the target multimedia resource; The target multimedia resource is recommended to the target account.

2. The method according to claim 1, characterized in that The acquiring, based on the recommendation parameter feature, the first fusion feature, and the weight feature, the target parameter of each of the multimedia resources comprises: For each of the multimedia resources, based on the recommendation parameter feature of the recommendation parameter, the first fusion feature, and the weight feature, respectively obtain a first parameter and a second parameter of the multimedia resource, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates the possibility that a terminal corresponding to the target account performs a conversion operation after performing the click operation on the multimedia resource when the multimedia resource is recommended to the target account; The first parameter and the second parameter are merged to obtain a target parameter of the multimedia resource.

3. The method according to claim 2, characterized in that The method further comprises: Based on the account information and the resource information of the multimedia resource, acquiring an association feature, the association feature indicating an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicating an interaction between the terminal corresponding to the target account and other multimedia resources, the other multimedia resources being multimedia resources recommended by the local end and other than the multimedia resources; The step of fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature includes: The account feature, the multimedia resource feature and the associated feature are fused to obtain the first fused feature.

4. The method according to claim 1, characterized in that: The obtaining of the target parameter of each multimedia resource comprises: The recommendation model is called to obtain a target parameter of each multimedia resource based on the account information, the recommendation parameter, and resource information of the plurality of multimedia resources to be recommended.

5. The method according to claim 4, characterized in that The recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the calling recommendation model, based on obtaining the target parameters of each of the multimedia resources, includes: For each of the multimedia resources to be recommended, calling the feature extraction sub-model, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature; Calling the click prediction sub-model, and acquiring a first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account; Calling the conversion prediction sub-model, and acquiring a second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature, the second parameter indicating the possibility that, when the multimedia resource is recommended to the target account, a terminal corresponding to the target account performs the click operation on the multimedia resource and then performs a conversion operation; The fusion sub-model is called to fuse the first parameter with the second parameter to obtain the target parameter of the multimedia resource.

6. The method according to claim 5, characterized in that The recommendation model also includes a weight acquisition sub-model, and the method further includes: Calling the weight acquisition sub-model to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature; The calling the click prediction sub-model to obtain the first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature includes: Calling the click prediction sub-model, fusing the first fused feature with the weight feature to obtain a second fused feature; The click prediction sub-model is called to fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

7. The method according to claim 6, characterized in that The weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; The calling of the weight acquisition sub-model to perform feature conversion on the recommendation parameter feature to obtain the weight feature of the first fusion feature includes: Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter feature to obtain the i-th weight feature; The calling the click prediction sub-model to fuse the first fused feature with the weight feature to obtain a second fused feature includes: Based on the first first parameter acquisition layer, performing dimensionality reduction processing on the first fusion feature to obtain a first first dimensionality reduction feature; Based on the first first fusion layer, the first first dimensionality reduction feature is fused with the first weight feature to obtain a first third fusion feature; For the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, perform dimensionality reduction processing on the j-1th third fusion feature to obtain the jth first dimensionality reduction feature; based on the jth first fusion layer, fuse the jth first dimensionality reduction feature with the jth weight feature to obtain the jth third fusion feature; repeat the above steps until the second fusion feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

8. The method according to claim 5, characterized in that The recommendation model also includes a weight acquisition sub-model, and the method further includes: Calling the weight acquisition sub-model to perform feature conversion on the recommended parameter feature to obtain a weight feature of the first fusion feature; The calling the conversion prediction sub-model to obtain the second parameter of the multimedia resource based on the recommendation parameter feature and the first fusion feature includes: Calling the conversion prediction sub-model, fusing the first fused feature with the weight feature to obtain a fourth fused feature; The conversion prediction sub-model is called to fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

9. The method according to claim 8, characterized in that The weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; The calling of the weight acquisition sub-model to perform feature conversion on the recommendation parameter feature to obtain the weight feature of the first fusion feature includes: Based on the i-th weight acquisition layer, perform feature conversion on the recommended parameter feature to obtain the i-th weight feature; The calling of the conversion prediction sub-model to fuse the first fused feature with the weight feature to obtain a fourth fused feature includes: Based on the first second parameter acquisition layer, performing dimensionality reduction processing on the first fusion feature to obtain a first second dimensionality reduction feature; Based on the first second fusion layer, the first second dimensionality reduction feature is fused with the first weight feature to obtain a first fifth fusion feature; For the jth second parameter acquisition layer and the jth second fusion layer, based on the jth second parameter acquisition layer, perform dimensionality reduction processing on the j-1th fifth fusion feature to obtain the jth second reduced dimensionality feature; based on the jth second fusion layer, fuse the jth second reduced dimensionality feature with the jth weight feature to obtain the jth fifth fusion feature; repeat the above steps until the fourth fusion feature is obtained based on the nth second fusion layer, where j is an integer greater than 1 and not greater than n.

10. The method according to claim 5, characterized in that The method further comprises: Calling the feature extraction sub-model to obtain, based on the account information and the resource information of the multimedia resource, an associated feature, wherein the associated feature indicates an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicates an interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local end and other than the multimedia resources; The calling of the feature extraction sub-model to fuse the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fused feature includes: The feature extraction sub-model is called to fuse the account feature, the multimedia resource feature and the associated feature to obtain the first fused feature.

11. The method according to any one of claims 4 to 10, characterized in that: The method further comprises: Acquire sample data and a sample tag corresponding to the sample data, wherein the sample data includes sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, wherein the sample tag indicates whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; Calling the recommendation model, based on the sample data, obtaining a prediction label of a sample multimedia resource corresponding to the sample data, the prediction label indicating the possibility of a terminal corresponding to the sample account performing an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; The recommendation model is trained based on the predicted labels of the sample data and the differences between the sample labels.

12. The method according to claim 11, characterized in that The sample label includes a first sample label, which indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the prediction label includes a first prediction label, which indicates the possibility of the terminal corresponding to the sample account performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; calling the recommendation model to obtain the prediction label of the sample multimedia resource corresponding to the sample data based on the sample data includes: Calling the feature extraction sub-model to fuse the sample account feature of the sample account information and the sample multimedia resource feature of the sample resource information to obtain a sample fusion feature; Calling the click prediction sub-model, based on the recommendation parameter feature of the sample recommendation parameter and the sample fusion feature, obtaining a second prediction label of the sample multimedia resource, the second prediction label indicating the possibility that a terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; Calling the conversion prediction sub-model, and acquiring a third prediction tag of the multimedia resource based on the sample recommendation parameter feature and the sample fusion feature, the third prediction tag indicating the possibility of a terminal corresponding to the sample account performing the click operation and then performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; The fusion sub-model is called to fuse the second prediction label with the third prediction label to obtain the first prediction label.

13. The method according to claim 12, characterized in that The sample label further includes a second sample label, which indicates whether a terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; and training the recommendation model based on the difference between the predicted label of the sample data and the sample label includes: The recommendation model is trained based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

14. The method according to claim 1, characterized in that The recommendation parameter is determined based on an operation performed by at least one account logged into the server on the multimedia resource recommended by the server, the server being the server logged into by the target account; The recommending the target multimedia resource to the target account includes: The target multimedia resource is sent to the server, and the server recommends the target multimedia resource to the target account.

15. A multimedia resource recommendation device, characterized in that: The device comprises: an acquisition unit, configured to acquire account information to be recommended and a recommendation parameter of the server, wherein the account information is used to describe a target account for logging into the server, and the recommendation parameter can reflect the operation performed by each account for logging into the server on the multimedia resource recommended by the server; The acquisition unit is further configured to perform, for each of the plurality of multimedia resources to be recommended, fusion of the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature; perform feature conversion on the recommendation parameter feature of the recommendation parameter to obtain a weight feature of the first fusion feature; based on the recommendation parameter feature, the first fusion feature and the weight feature, respectively obtain a target parameter of each of the multimedia resources, the target parameter indicating the possibility of a terminal corresponding to the target account performing an operation on the multimedia resource when the corresponding multimedia resource is recommended to the target account indicated by the account information; A selection unit is configured to perform, based on target parameters of the plurality of multimedia resources, determining at least one multimedia resource with the highest target parameter from the plurality of multimedia resources as a target multimedia resource; The recommendation unit is configured to recommend the target multimedia resource to the target account.

16. The device according to claim 15, characterized in that The acquisition unit comprises: an acquisition subunit, configured to execute, for each of the multimedia resources, based on the recommendation parameter feature of the recommendation parameter, the first fusion feature, and the weight feature, respectively acquiring a first parameter and a second parameter of the multimedia resource, wherein the first parameter indicates the possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account, and the second parameter indicates the possibility that a terminal corresponding to the target account performs a conversion operation after performing the click operation on the multimedia resource when the multimedia resource is recommended to the target account; The fusion subunit is configured to perform fusion of the first parameter and the second parameter to obtain a target parameter of the multimedia resource.

17. The device according to claim 16, characterized in that The acquisition unit is further configured to acquire, based on the account information and the resource information of the multimedia resource, an association feature, wherein the association feature indicates an interaction between the terminal corresponding to the target account and the multimedia resource, and further indicates an interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local terminal and other than the multimedia resources; The fusion subunit is configured to fuse the account feature, the multimedia resource feature and the associated feature to obtain the first fusion feature.

18. The device according to claim 16, characterized in that The acquisition unit is configured to execute the call recommendation model and acquire the target parameter of each multimedia resource based on the account information, the recommendation parameter and the resource information of the multiple multimedia resources to be recommended.

19. The device according to claim 18, characterized in that The recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the acquisition unit includes: A fusion subunit is configured to execute, for each multimedia resource to be recommended, calling the feature extraction submodel, fusing the account feature of the account information and the multimedia resource feature of the resource information of the multimedia resource to obtain a first fusion feature; an acquisition subunit configured to execute calling the click prediction submodel, and acquire a first parameter of the multimedia resource based on the recommendation parameter feature of the recommendation parameter and the first fusion feature, wherein the first parameter indicates a possibility that a terminal corresponding to the target account performs a click operation on the multimedia resource when the multimedia resource is recommended to the target account; The acquisition sub-unit is further configured to execute calling the conversion prediction sub-model, and based on the recommendation parameter feature and the first fusion feature, acquire a second parameter of the multimedia resource, wherein the second parameter indicates the possibility that, when the multimedia resource is recommended to the target account, the terminal corresponding to the target account performs the click operation on the multimedia resource and then performs the conversion operation; The fusion sub-unit is further configured to execute the call of the fusion sub-model, fuse the first parameter with the second parameter, and obtain the target parameter of the multimedia resource.

20. The device according to claim 19, characterized in that The recommendation model further includes a weight acquisition sub-model, and the device further includes: a conversion unit configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommendation parameter feature, and obtain a weight feature of the first fusion feature; The acquisition sub-unit is configured to execute calling the click prediction sub-model, fuse the first fusion feature with the weight feature to obtain a second fusion feature; call the click prediction sub-model, fuse the second fusion feature with the recommendation parameter feature to obtain the first parameter.

21. The device according to claim 20, characterized in that The weight acquisition submodel includes n weight acquisition layers, the click prediction submodel includes n first parameter acquisition layers and n first fusion layers, the i-th first fusion layer corresponds to the i-th weight acquisition layer and the i-th first parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; The conversion unit is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature; The acquisition subunit is configured to perform dimensionality reduction processing on the first fusion feature based on the first first parameter acquisition layer to obtain a first first dimensionality reduction feature; Based on the first first fusion layer, the first first dimensionality reduction feature is fused with the first weight feature to obtain the first third fused feature; for the jth first parameter acquisition layer and the jth first fusion layer, based on the jth first parameter acquisition layer, the j-1th third fused feature is subjected to dimensionality reduction processing to obtain the jth first dimensionality reduction feature; based on the jth first fusion layer, the jth first dimensionality reduction feature is fused with the jth weight feature to obtain the jth third fused feature; the above steps are repeated until the second fused feature is obtained based on the nth first fusion layer, where j is an integer greater than 1 and not greater than n.

22. The device according to claim 19, characterized in that The recommendation model further includes a weight acquisition sub-model, and the device further includes: a conversion unit configured to execute calling the weight acquisition sub-model, perform feature conversion on the recommendation parameter feature, and obtain a weight feature of the first fusion feature; The acquisition sub-unit is configured to execute calling the conversion prediction sub-model, fuse the first fusion feature with the weight feature to obtain a fourth fusion feature; call the conversion prediction sub-model, fuse the fourth fusion feature with the recommended parameter feature to obtain the second parameter.

23. The device according to claim 22, characterized in that The weight acquisition submodel includes n weight acquisition layers, the conversion prediction submodel includes n second parameter acquisition layers and n second fusion layers, the i-th second fusion layer corresponds to the i-th weight acquisition layer and the i-th second parameter acquisition layer, respectively, n is an integer greater than 1, and i is a positive integer not greater than n; The conversion unit is configured to perform feature conversion on the recommended parameter feature based on the i-th weight acquisition layer to obtain the i-th weight feature; The acquisition subunit is configured to perform dimensionality reduction processing on the first fusion feature based on the first second parameter acquisition layer to obtain a first second dimensionality reduction feature; Based on the first second fusion layer, the first second dimensionality reduction feature is fused with the first weight feature to obtain the first fifth fusion feature; for the jth second parameter acquisition layer and the jth second fusion layer, based on the jth second parameter acquisition layer, the j-1th fifth fusion feature is subjected to dimensionality reduction processing to obtain the jth second dimensionality reduction feature; based on the jth second fusion layer, the jth second dimensionality reduction feature is fused with the jth weight feature to obtain the jth fifth fusion feature; the above steps are repeated until the fourth fusion feature is obtained based on the nth second fusion layer, where j is an integer greater than 1 and not greater than n.

24. The device according to claim 19, characterized in that The acquisition unit is further configured to execute the call of the feature extraction sub-model, and acquire the associated features based on the account information and the resource information of the multimedia resources, wherein the associated features indicate the interaction between the terminal corresponding to the target account and the multimedia resources, and also indicate the interaction between the terminal corresponding to the target account and other multimedia resources, wherein the other multimedia resources are multimedia resources recommended by the local end and other than the multimedia resources; The fusion subunit is configured to execute and call the feature extraction submodel, fuse the account feature, the multimedia resource feature and the associated feature, and obtain the first fusion feature.

25. The device according to any one of claims 18 to 24, characterized in that The device also includes: The acquisition unit is further configured to acquire sample data and a sample tag corresponding to the sample data, wherein the sample data includes sample account information, sample resource information of a sample multimedia resource, and a sample recommendation parameter corresponding to the sample account information, and the sample tag indicates whether a terminal corresponding to the sample account has performed an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; The acquisition unit is further configured to execute calling the recommendation model, and based on the sample data, acquire a prediction tag of a sample multimedia resource corresponding to the sample data, wherein the prediction tag indicates a possibility that a terminal corresponding to the sample account will perform an operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; The training unit is configured to train the recommendation model based on the predicted labels of the sample data and the differences between the sample labels.

26. The device according to claim 25, characterized in that The sample label includes a first sample label, which indicates whether the terminal corresponding to the sample account performs a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the prediction label includes a first prediction label, which indicates the possibility of the terminal corresponding to the sample account performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the recommendation model includes a feature extraction sub-model, a click prediction sub-model, a conversion prediction sub-model and a fusion sub-model; the acquisition unit is configured to execute and call the feature extraction sub-model, fuse the sample account feature of the sample account information and the sample multimedia resource feature of the sample resource information, and obtain a sample fusion feature; Calling the click prediction sub-model, based on the recommendation parameter feature of the sample recommendation parameter and the sample fusion feature, obtaining a second prediction label of the sample multimedia resource, the second prediction label indicating the possibility that a terminal corresponding to the sample account will perform a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; Calling the conversion prediction sub-model, and acquiring a third prediction tag of the multimedia resource based on the sample recommendation parameter feature and the sample fusion feature, the third prediction tag indicating the possibility of a terminal corresponding to the sample account performing the click operation and then performing a conversion operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; The fusion sub-model is called to fuse the second prediction label with the third prediction label to obtain the first prediction label.

27. The device according to claim 26, characterized in that The sample label also includes a second sample label, which indicates whether the terminal corresponding to the sample account has performed a click operation on the sample multimedia resource when the sample multimedia resource is recommended to the sample account; the training unit is configured to train the recommendation model based on the difference between the first predicted label of the sample data and the first sample label, and the difference between the second predicted label of the sample data and the second sample label.

28. The device according to claim 15, characterized in that The recommendation parameter is determined based on an operation performed by at least one account logged into a server on a multimedia resource recommended by the server, the server being the server logged into by the target account; The recommendation unit is configured to send the target multimedia resource to the server, and the server recommends the target multimedia resource to the target account.

29. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the multimedia resource recommendation method as claimed in any one of claims 1 to 14.

30. A non-transitory computer-readable storage medium, characterized in that When the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the multimedia resource recommendation method as described in any one of claims 1 to 14.

31. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the multimedia resource recommendation method according to any one of claims 1 to 14 is implemented.

Images