CN117932140A - Feature generation method, device and readable storage medium for multimedia resource recommendation - Google Patents

Abstract

The embodiment of the application discloses a feature generation method, equipment and a readable storage medium for multimedia resource recommendation, wherein the method comprises the following steps: acquiring first object behavior data provided by a client data source, and generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data; acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource, and generating at least two second object behavior statistical features of at least two different time periods associated with the second object based on the second object behavior data; and generating object statistical features of the objects based on the first object behavior statistical features associated with the first objects and the second object behavior statistical features associated with the second objects, and determining the object statistical features as input features of the multimedia resource recommendation model. By adopting the method and the device, the recommendation efficiency of the multimedia resources can be improved, and the applicability of the multimedia resource recommendation is enhanced.

Description

Feature generation method, device and readable storage medium for multimedia resource recommendation

Technical Field

The present application relates to the field of internet technologies, and in particular, to a method and an apparatus for generating characteristics of multimedia resource recommendation, and a readable storage medium.

Background

Along with the increasingly powerful functions of terminals such as smart phones and tablet computers, the functions of application programs (or simply applications) applied to the terminals such as smart phones and tablet computers are increasingly rich and various, people can acquire interesting information through the applications, and the variety of the information brings rich and various user experiences to terminal users. Nowadays, applications become more humanized with technological progress, and in order to bring better experience feeling to end users, multimedia resource recommendation of each big application for the end users is more personalized.

In general, multimedia asset recommendation is generally based on user preferences for certain types of multimedia assets, similar types of multimedia asset recommendation, or based on the needs or preferences of a community of users who are interested in competing or having a common experience. However, because the user base is large, the number of multimedia resources interested by the user in the past data can reach billions, so that the accuracy of multimedia resource recommendation for the user is low, and the applicability is poor.

Disclosure of Invention

The embodiment of the application provides a feature generation method, equipment and a computer readable storage medium for multimedia resource recommendation, which can improve the recommendation efficiency of multimedia resources and enhance the applicability of multimedia resource recommendation.

In a first aspect, an embodiment of the present application provides a feature generating method for multimedia resource recommendation, where the method includes:

Acquiring first object behavior data provided by a client data source, and generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data, wherein the first object is each object carried in the first object behavior data, and the first object behavior statistical features of any time period associated with any first object comprise at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource attribute features;

Acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource, and generating at least two second object behavior statistical characteristics of at least two different time periods associated with a second object based on the second object behavior data, wherein the second object is each object carried in the first object behavior data, and at least two object resource behavior attribute characteristics in the second object behavior statistical characteristics of any time period associated with any second object;

Generating object statistical features of each object based on the first object behavior statistical features associated with each first object and the second object behavior statistical features associated with each second object, and determining the object statistical features as input features of a multimedia resource recommendation model, wherein the multimedia resource recommendation model is obtained by training the second object behavior data, and the multimedia resource model is used for outputting multimedia resource recommendation values of each object based on the object statistical features of each object.

In one possible implementation manner, the generating at least two first object behavior statistical features of at least two different time periods associated with the first object based on the first object behavior data includes:

Generating at least two levels of object behavior characteristics of at least two different time periods based on the generation time of each object behavior data carried in the first object behavior data, wherein the one level of object behavior characteristics is used for generating object behavior statistical characteristics of each first object in one time period, and the generation time of the object behavior data contained in the different time periods is not overlapped;

Generating at least two first object behavior statistical features with the object identifiers of the first objects as indexes based on the object identifiers of the first objects carried in the at least two levels of object behavior features, wherein any one first object is associated with at least two first object behavior statistical features of the at least two different time periods with the object identifier of any one first object as an index.

In one possible implementation manner, the generating at least two levels of object behavior features of at least two different time periods based on the generation time of each object behavior data carried in the first object behavior data includes:

Dividing the object behavior data of each first object belonging to the same time period into object behavior data of the same level based on the generation time of each object behavior data carried in the first object behavior data to obtain at least two levels of object behavior data;

generating at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource attribute features of each first object based on the object behavior data of each first object included in each level of object behavior data, and splicing the at least two attribute features associated with each first object to generate each level of object behavior features corresponding to each time period, so as to obtain at least two levels of object behavior features of at least two different time periods.

In one possible implementation manner, the object resource behavior attribute features include one or more of a multimedia resource preference degree of an object or an object preference degree of a multimedia resource; the generating object resource behavior attribute features of the respective first objects based on the object behavior data of the respective first objects included in the respective levels of object behavior data includes:

Acquiring any object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on any multimedia resource, determining a ratio of the any object behavior statistical value to the sum of the object behavior statistical values of all objects in all levels of object behavior data on any multimedia resource as the preference degree of any first object on any multimedia resource so as to obtain the preference degree of any first object on the multimedia resource; or alternatively

Acquiring any object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on any multimedia resource, determining a ratio of the any object behavior statistical value to the sum of the object behavior statistical values of any first object in all levels of object behavior data on all multimedia resources as the preference degree of any first object on any multimedia resource so as to obtain the object preference degree of any multimedia resource.

In one possible implementation manner, the object resource behavior attribute features include expected click rate of the object on the multimedia resource; the generating object resource behavior attribute features of the respective first objects based on the object behavior data of the respective first objects included in the respective levels of object behavior data includes:

Acquiring a first object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on a first multimedia resource, and acquiring a first object behavior operation proportion of all object users in all levels of object behavior data on the first multimedia resource;

acquiring a second object behavior statistical value of any first object in the object behavior data of each level of object, and acquiring a second object behavior operation proportion of all user objects in the object behavior data of each level of object to the second multimedia resource;

And generating expected click quantity of any first object on the multimedia resources based on the first object behavior statistical value, the first object behavior operation proportion, the second object behavior statistical value and the second object behavior operation proportion and the object behavior statistical value of any first object on all the multimedia resources so as to obtain the expected click quantity of each first object on the multimedia resources.

In one possible implementation manner, the generating at least two second object behavior statistics of at least two different time periods associated with each object carried in the first object behavior data based on the second object behavior data includes:

Dividing the second object behavior data into a plurality of P ₁ -level object behavior data based on the generation time of each object behavior data carried in the second object behavior data, wherein the P ₁ -level object behavior data is object behavior data included in the minimum unit time length of the second object behavior data divided according to time intervals;

Generating at least two P ₂ -level object behavior data of at least two different time periods based on accumulation of a plurality of the P ₁ -level object behavior data included in a plurality of minimum unit time periods, wherein one P ₂ -level object behavior data is accumulated by a plurality of the P ₁ -level object behavior data in one target time period, and the target time period is a positive integer multiple of the minimum unit time period;

And generating at least two second object behavior statistical features with the object identifiers of the second objects as indexes based on the object identifiers corresponding to the second objects carried in the P ₂ -level object behavior data, wherein any one second object is associated with at least two second object behavior statistical features in the at least two different time periods with the object identifier of any one second object as an index.

In one possible implementation manner, the generating at least two second object behavior statistical features using the object identifiers of the second objects as indexes based on the object identifiers corresponding to the second objects carried in the P ₂ -level object behavior data includes:

Generating P ₂ -level object behavior characteristics associated with each second object based on object identifiers corresponding to each second object carried in the P ₂ -level object behavior data, wherein the P ₂ -level object behavior characteristics of one second object comprise at least two object resource behavior attribute characteristics of the second object;

Splicing at least two object resource behavior attribute features in P ₂ -level object behavior features associated with the object identifiers of the second objects by taking the object identifiers of the second objects as indexes so as to generate second object behavior statistical features associated with the object identifiers;

The object resource behavior attribute features include at least one object resource behavior associated feature of at least one of click time length, click times, click rate, exposure times, watching time length and interaction time length of the object for the multimedia resource.

In one possible implementation manner, the generating the object statistics of each object based on the first object behavior statistics associated with each first object and the second object behavior statistics associated with each second object includes:

The object identification corresponding to each first object is obtained from the first object behavior statistical characteristics, and the object identification corresponding to each second object is obtained from the second object behavior statistical characteristics;

and splicing the first object behavior statistical features and the second object behavior statistical features by taking the object identifications of the first objects and the object identifications of the second objects as indexes to generate object statistical features corresponding to the object identifications, wherein the first object behavior statistical features and the second object behavior statistical features with the same object identifications are spliced into object statistical features of the same object.

In a second aspect, an embodiment of the present application provides a feature generating device for multimedia resource recommendation, where the device includes:

the acquisition module is used for acquiring first object behavior data provided by a client data source and acquiring the first object behavior data and the second object behavior data aiming at the object behavior;

The first object behavior statistical feature generation module is used for generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data acquired by the acquisition module, wherein the first object is each object carried in the first object behavior data, the first object behavior statistical feature of any time period associated with any first object comprises at least two attribute features of an object attribute feature, a duration attribute feature, a resource content attribute feature, a resource type attribute feature and an object resource attribute feature, the first object behavior data provided by a client data source is acquired, and at least two first object behavior statistical features of at least two different time periods associated with each object carried in the first object behavior data are generated based on the first object behavior data;

the acquisition module is further used for acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource;

The feature generation module is further configured to generate at least two second object behavior statistical features of at least two different time periods associated with a second object based on the second object behavior data acquired by the acquisition module, where the second object is each object carried in the first object behavior data, at least two second object behavior attribute features in the second object behavior statistical features of any time period associated with any second object are acquired from second object behavior attribute features of at least two second object behavior feature data of any time period associated with the second object, and generate at least two second object behavior statistical features of at least two different data periods associated with each object carried in the first object behavior data based on the second object behavior data;

and the feature aggregation module is used for generating object statistical features of all objects based on the first object behavior statistical features associated with all the first objects and the second object behavior statistical features associated with all the second objects, determining the object statistical features as input features of a multimedia resource recommendation model, wherein the multimedia resource recommendation model is obtained by training the second object behavior data, and is used for outputting multimedia resource recommendation values of all the objects based on the object statistical features of all the objects, generating object statistical features of all the objects based on the first object behavior statistical features and the second object behavior statistical features associated with all the objects, and determining the input features of the multimedia resource recommendation model by using the object statistical features.

In a third aspect, an embodiment of the present application provides a computer apparatus, including: a processor, a memory, and a network interface;

The processor is connected to the memory, the network interface for providing data communication functions, the memory for storing program code, the processor for invoking the program code to perform the method as provided in the first aspect and any possible implementation thereof.

In a fourth aspect, embodiments of the present application provide a computer readable storage medium storing a computer program comprising program instructions which, when executed by a processor, perform a method as provided by the first aspect of embodiments of the present application and any one of its possible implementations.

In a fifth aspect, embodiments of the present application provide a computer program product comprising a computer program stored in a computer readable storage medium, the computer program being adapted to be read and executed by a processor to cause a computer device having the processor to perform a method as provided by the first aspect of the embodiments of the present application and any one of its possible implementations.

In the embodiment of the application, first object behavior data provided by a client data source is acquired through a target client, and at least two-stage object behavior characteristics of at least two different time periods are generated based on the generation time of each object behavior data carried in the first object behavior data. The first object is each object carried in the first object behavior data, and the first object behavior statistical feature of any time period associated with any first object includes at least two attribute features of an object attribute feature, a duration attribute feature, a resource content attribute feature, a resource type attribute feature and an object resource behavior attribute feature. And then, acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource through the target client, and generating at least two second object behavior statistical characteristics of at least two different time periods associated with the second object based on the second object behavior data. The second objects are at least two object resource behavior attribute features in second object behavior statistical features of any time period associated with any second object, wherein the second objects are all objects carried in the first object behavior data. Finally, generating object statistical features of the objects based on the first object behavior statistical features associated with the first objects and the second object behavior statistical features associated with the second objects, and determining the object statistical features as input features of the multimedia resource recommendation model. The multimedia resource recommendation model is obtained by training the second object behavior data, is used for outputting the multimedia resource recommendation value of each object based on the object statistical characteristics of each object, can provide a more comprehensive, clear and easily-expanded statistical characteristic system, improves the utilization rate of computing resources, model training resources and storage resources of multimedia resource recommendation, improves the recommendation accuracy and efficiency of the multimedia resources, and enhances the applicability.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a system architecture according to an embodiment of the present application;

FIG. 2 is a flowchart illustrating a method for generating features of media resource recommendation according to an embodiment of the present application;

FIG. 3 is a schematic view of an application scenario of a feature generation method for multimedia resource recommendation according to an embodiment of the present application;

Fig. 4 is a schematic diagram of statistical characteristics of a method for generating characteristics of multimedia resource recommendation according to an embodiment of the present application;

fig. 5 is another application scenario schematic diagram of a feature generation method of multimedia resource recommendation provided in an embodiment of the present application;

fig. 6 is another application scenario schematic diagram of a feature generation method of multimedia resource recommendation provided in an embodiment of the present application;

fig. 7 is another application scenario schematic diagram of a feature generation method of multimedia resource recommendation provided in an embodiment of the present application;

fig. 8 is another application scenario schematic diagram of a feature generation method of multimedia resource recommendation provided in an embodiment of the present application;

fig. 9 is another application scenario schematic diagram of a feature generation method of multimedia resource recommendation provided in an embodiment of the present application;

fig. 10 is a schematic structural diagram of a feature generating device for multimedia resource recommendation according to an embodiment of the present application;

Fig. 11 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

Referring to fig. 1, fig. 1 is a schematic diagram of a system architecture according to an embodiment of the application. As shown in fig. 1, the system architecture may include a service server 100 and a terminal cluster, where the terminal cluster may include: terminal devices 200a, 200b, 200c, … …, 200n, and the like. The service server 100 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server that provides a cloud database, a cloud service, cloud computing, a cloud function, cloud storage, a network service, cloud communication, a middleware service, a domain name service, a security service, a CDN, and basic cloud computing services such as a big data and an artificial intelligence platform. The terminal devices (including terminal device 200a, terminal device 200b, terminal devices 200c, … …, and terminal device 200 n) may be a palm computer, a smart phone, a notebook computer, a desktop computer, a tablet computer, a Mobile Internet Device (MID) INTERNET DEVICE, a wearable device (e.g., a smart watch, a smart bracelet, etc.), a smart computer, a smart car, etc. a smart terminal. The service server 100 may establish communication connection with each terminal device in the terminal cluster, and may also establish communication connection between each terminal device in the terminal cluster. In other words, the service server 100 may establish a communication connection with each of the terminal apparatuses 200a, 200b, 200c, … …, 200n, for example, a communication connection may be established between the terminal apparatus 200a and the service server 100. A communication connection may be established between terminal device 200a and terminal device 200b, and a communication connection may also be established between terminal device 200a and terminal device 200 c. The communication connection is not limited to a connection manner, and may be directly or indirectly connected through a wired communication manner, or may be directly or indirectly connected through a wireless communication manner, and the like, and may be specifically determined according to an actual application scenario, and the present application is not limited herein.

It should be understood that each terminal device in the terminal cluster shown in fig. 1 may be provided with an application client, and when the application client runs in each terminal device, data interaction may be performed between the application client and the service server 100 shown in fig. 1, so that the service server 100 may receive service data from each terminal device, or the service server 100 pushes service data (such as multimedia resources) to each terminal device. The application client may be an application client having a function of displaying data information such as text, image and video, such as a news application, a learning application, a social application, an instant messaging application, a live broadcast application, a short video application, a music application, a shopping application, a novel application, a payment application, etc., and may specifically be determined according to actual application scene requirements, without limitation. The application client may be an independent client, or may be an embedded sub-client integrated in a certain client (such as an instant messaging client, a social client, etc.), which may be specifically determined according to an actual application scenario, and is not limited herein. For convenience of description, taking the target client as an example, each operation object can view, click, collect and share the multimedia resource in the target application through the terminal device in the process of using the target client through the terminal device. It will be appreciated that the above-mentioned multimedia resource may be any kind of multimedia data, and may specifically include, but not limited to, audio, picture, or video, etc., and may specifically be determined according to an actual application scenario, which is not limited herein. The service server 100 may be a server of the multimedia resource recommendation application, and may be a collection of multiple servers including a background server, a data processing server, and the like, which correspond to the application client. The service server 100 may receive object behavior data from a terminal device (for example, obtain first object behavior data provided by a client data source and second object behavior data provided by a training sample data source of multimedia resource recommendation), generate object behavior statistics features associated with each operation object based on the object behavior data of each operation object, determine the object statistics features of each operation object as input features of a multimedia resource recommendation model, and thereby output an object with a higher multimedia resource recommendation value for each operation object, where the multimedia resource model is used to output the multimedia resource recommendation value for each object based on the object statistics features of each object.

It will be appreciated that in the specific embodiments of the present application, related data such as user information is involved, and when the embodiments of the present application are applied to specific products or technologies, user permissions or consents need to be obtained, and the collection, use and processing of related data need to comply with relevant laws and regulations and standards of the relevant countries and regions.

The feature generation method (for convenience of description, may be simply referred to as feature generation method or method) of multimedia resource recommendation provided by the embodiment of the present application is applicable to multimedia resource recommendation based on an application program (such as the above-mentioned target client). It is understood that the terminal device to which the above feature generation method is applied includes, but is not limited to, a smart phone, a computer, a tablet computer, a Personal Digital Assistant (PDA), a mobile internet device (mobile INTERNET DEVICE, MID), a wearable device, and the like. Optionally, the terminal device may also be the smart phone, the computer, the tablet pc, the PDA, the MID, a server corresponding to the wearable device, or the like, which may be specifically determined according to an actual application scenario, and is not limited herein. Correspondingly, the feature generating device (or simply multimedia resource recommending device) for recommending the multimedia resources provided by the embodiment of the application comprises, but is not limited to, a smart phone, a computer, a tablet personal computer, a PDA, an MID, a wearable device and the like. For convenience of description, the multimedia resource recommendation device and/or the terminal device provided by the embodiment of the application will be described by taking a smart phone (or simply called a mobile phone) as an example.

It may be understood that the feature generating method provided by the embodiment of the present application may be performed by the service server 100 shown in fig. 1, or may be performed by a terminal device (e.g., any one of the terminal device 200a, the terminal devices 200b, … …, and the terminal device 200n shown in fig. 1), or may be performed by the terminal device and the service server together, which may be specifically determined according to an actual application scenario, and is not limited herein. For the convenience of subsequent understanding and description, the embodiment of the present application may select one terminal device as a target terminal device in the terminal device cluster shown in fig. 1, for example, use the terminal device 200b as a target terminal device.

It will be appreciated that the statistical class feature based on the object behavior is a very important class feature in the multimedia resource recommendation ranking model. However, the recommendation of the multimedia resources is generally performed based on past behavior data of the operation object, such as the click rate, the viewing time length, the number of times of viewing, and other characteristics of the operation object on a certain type of multimedia resources, or the multimedia resources which may be of interest are recommended to the operation object based on the requirements or preferences of the object group which is interested in or has common experience. However, as the base number of the operation object is large, the number of multimedia resources interested by the operation object in the past data can reach the billion level, so that the accuracy of the application of the multimedia resource recommendation aiming at the operation object is low, and the applicability is poor. The application provides a comprehensive, clear and easily-expanded statistical feature system for a multimedia resource recommendation ordering model, and the statistical feature of the object behavior is divided into five attribute features, namely an object attribute feature, a duration attribute feature, a resource content attribute feature, a resource type attribute feature and an object resource behavior attribute feature, so that multimedia resource recommendation (similar multimedia resource recommendation, author recommendation, preference recommendation and the like) can be performed based on the statistical feature of each object behavior to enhance the recommendation effect of an object application program on the multimedia resource, improve the recommendation accuracy of the multimedia resource and enhance the applicability.

The feature generation method provided by the embodiment of the application can be applied to applications for recommending various types of multimedia resources, wherein the applications for recommending various types of multimedia resources include but are not limited to: the application clients having the function of processing multimedia resources, such as multimedia applications, browser applications, game applications, shopping applications, tools applications, social applications, travel applications, and education applications, are not limited herein. Wherein, the same type of application can comprise multiple applications, and the application is not limited herein. For example, the multimedia applications may include a movie player, a music player, a photography application, a beauty application, an audio entry application, and the like. Such browser-like applications include, but are not limited to, QQ browsers. Such gaming applications include, but are not limited to, queen glory, QQ galloping, and the like. Such shopping applications include, but are not limited to, movie ticketing applications, food booking applications, living necessities purchasing applications, and the like. Such tool class applications include, but are not limited to, file editing, mail, alarm clock, calendar, photo album, settings, compass, and the like. Such social application classes include, but are not limited to, weChat, QQ, and the like. Such travel-like applications include railway 12306, trip travel, and the like. Such educational applications include, but are not limited to, weChat reading, QQ reading, and the like.

Further, referring to fig. 2, fig. 2 is a flowchart illustrating a feature generation method of multimedia resource recommendation according to an embodiment of the present application. For the sake of understanding, the embodiment of the present application is described by taking the terminal device as an example, that is, the terminal device 200b in fig. 2 is described by taking as an example, and the service server may be the service server 100 of the embodiment corresponding to fig. 1. In the feature generation method of multimedia resource recommendation shown in fig. 2, each step of data processing may be performed by the service server 100 in fig. 1 described above, and as shown in fig. 2, the feature generation method of multimedia resource recommendation may include at least the following steps S101 to S103.

S101, acquiring first object behavior data provided by a client data source, and generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data, wherein the first object is each object carried in the first object behavior data.

In some possible embodiments, the operation object (i.e. the end user) performs real-name information registration through the target client loaded on the terminal device 200b, after the real-name information registration is successful, the operation object may obtain, through the target client, an object identifier uniquely corresponding to the operation object, and obtain, through the target client loaded on the terminal device 200b, the first object behavior data provided by the client data source. The first object behavior data provided by the client data source is original statistics data of the whole end generated according to the operation object behavior log. For convenience of description, the embodiment of the present application will be described by taking an application client having a function of processing multimedia resources, which is loaded on the terminal device 200b, as a target client, which is simply referred to as a target client. Referring to fig. 3, fig. 3 is an application scenario schematic diagram of a feature generation method for multimedia resource recommendation according to an embodiment of the present application. In the embodiment of the present application, after the operation object obtains, through the target client, an object identifier uniquely corresponding to the operation object, the object identifier may be logged in to the target client through the target client loaded on the terminal device 200b, so as to implement a recommendation process for the target multimedia resource. The present application refers to an application account number for a target client registered in the target client by an operation object corresponding to the terminal device 200b as an object identifier. The object operation state on its operation interface (e.g., interface 1) can be detected by the terminal device 200 b. When the operation object clicks on an icon of the target client on the operation interface of the terminal device 200b, the terminal device may be triggered to start the operation interface (e.g., interface 2) of the target client. At this time, the terminal device may detect an object operation instruction on its operation interface, and may determine, according to a click position of the object operation instruction, an application triggered to be started by object selection as a target client. At this time, the terminal device may start the operation interface (e.g., interface 2) of the target client. As shown in fig. 3, on the operation interface of the target client, including a plurality of switchable interfaces such as home page, my and extended functions, the operation object may click on different icons on the operation interface of the target client to switch different function interfaces of the target client. For example, it is assumed that when the operation object clicks on the recommended icon of the target client, the terminal device may be triggered to switch the operation interface of the target client to the recommended interface (e.g. interface 3) of the multimedia resource. On the interface 3, a hot ranking list of a plurality of multimedia resources may be displayed, so that corresponding multimedia resources may be recommended to the operation object.

As shown in fig. 3, a recommendation list for recommending a plurality of multimedia resources published on a target client according to a plurality of ranking rules may be displayed on the interface 3, including a ranking list, a surge list, a head-end, and the like. When the target client detects an object operation instruction on a display area of any recommendation list, the target client switches to a recommendation page selected corresponding to the object operation instruction, and the multimedia resource recommendation list on the recommendation list is output and displayed to an operation object. For example, when the operation object clicks on the operation area corresponding to "my", the target client may switch to the personal recommendation list corresponding to "my" for recommending the multimedia resource to the operation object, so as to implement the process of recommending the multimedia resource to the operation object by the terminal device.

In some possible embodiments, after the operation object completes registration on the target client loaded on the terminal device 200b, the operation object may log in to the target client through the object identifier allocated by the target client and uniquely corresponding to the operation object. After the operation object logs in the target client, when the terminal equipment recommends a target multimedia resource to the operation object, first object behavior data provided by a data source of the target client can be obtained first, and at least two first object behavior statistical features of at least two different time periods associated with the first object can be generated based on the first object behavior data. The first object behavior data carries object behavior data of each operation object, where each operation object is successfully registered in the target client, and obtains an object identifier (i.e., an application account number) of the target client. Referring to fig. 4, fig. 4 is a schematic diagram of statistical characteristics of a method for generating characteristics of multimedia resource recommendation according to an embodiment of the present application. The feature generating method provided by the embodiment of the application specifically divides the feature index of the feature generating method into five attribute features, and each attribute feature at least comprises one dimension attribute. The feature index of the feature generating method may be specifically determined according to an actual application scenario, so that the attribute features of the feature generating method may also be determined according to an actual situation, and the dimensions of each attribute feature may also be determined according to an actual application scenario. In the embodiment of the present application, the first object behavior statistical feature generated based on the first object behavior data includes an object attribute feature, a duration attribute feature, a resource content attribute feature, a resource type attribute feature, and an object resource behavior attribute feature, and each attribute feature includes at least one dimension. As described in fig. 4, the object attribute features include an object feature, a content feature, and an object content feature. The time length attribute feature may divide the first object behavior data into a real-time object behavior feature, a short-term object behavior feature, a medium-term object behavior feature and a long-term object behavior feature based on the generation time of each object behavior data carried by the first object behavior data, where the generation times of the object behavior data included in the different time periods do not overlap, and the time lengths of the different time periods may be determined according to an actual application scenario, which is not limited herein. The resource content attribute features comprise a primary class resource content attribute feature, a secondary class resource content attribute feature, an author attribute feature and a tag attribute feature, wherein the primary class resource content range is larger than the secondary class resource content range. Wherein, the resource type attribute features comprise graphic and text multimedia resource content attribute features and video multimedia resource content attribute features. The object resource behavior attribute features include click time length, click times, click rate, exposure times, viewing time length, interaction time length, multimedia resource preference degree of the object, object preference degree of the multimedia resource, expected click quantity of the object on the multimedia resource and the like of the object for the multimedia resource.

Referring to fig. 5, fig. 5 is a schematic diagram of another application scenario of the feature generation method of multimedia resource recommendation according to the embodiment of the present application. After each operation object successfully logs in the target client through loading in the terminal device 200b, the target client may acquire first object behavior data provided by the target client data source, where the first object behavior data is composed of object behavior data associated with each first object. The first object is each operation object carried in the first object behavior data, and the first object behavior data at least carries more than two object behavior data associated with the operation object. Wherein, the at least two or more operation objects complete real-name registration through the target client, and obtain an object identifier uniquely corresponding to each operation object through the target client, where the object identifier may be registered in the target client through the terminal device 200 b. When the operation object successfully logs in the target client, the target client can record object behavior data of the operation object, such as operations of watching, collecting, sharing and the like of the operation object on the multimedia resource. For convenience of description, the above-described first objects (i.e., the operation objects) will be distinguished, and the first objects will be named as operation object 1, operation objects 2, … …, and operation object n, respectively, where n is a positive integer. It may be appreciated that the target multimedia resource recommended on the target client may be any kind of multimedia data, and specifically may include, but is not limited to, audio, picture, video, etc., and may be specifically determined according to an actual application scenario, which is not limited herein. For convenience of description, the embodiments of the present application will describe types of multimedia resources targeting video resources. When the terminal equipment receives the first object behavior data, the object behavior data of each operation object belonging to the same time period is divided into the object behavior data of the same level based on the generation time of each object behavior data carried in the first object behavior data, so as to obtain at least two levels of object behavior data. The division of the first object behavior data based on the generation time of each object behavior data may be divided according to the actual application scenario, where the division is not limited, that is, the first object behavior data may be divided into hour-level object behavior data and day-level object behavior data, that is, several week-level object behavior data, or the first object behavior data may be divided into month-level object behavior data, which may be specifically determined according to the actual application scenario, where the division is not limited. For convenience of description, the first object behavior data is divided into day-level object behavior data and week-level object behavior data. As shown in fig. 5, the above-mentioned day-level object behavior data may be classified into day-level graphic object behavior data and day-level video object behavior data based on the type of the target multimedia resource (i.e., the above-mentioned video resource), and the above-mentioned week-level object behavior data may be classified into week-level graphic object behavior data and week-level video object behavior data. Wherein, the graphics context is the graphics context resource corresponding to the video resource. The generation time of each object behavior data carried by the day-level image-text object behavior data can be used for dividing the day-level image-text object behavior data into 1 day image-text object behavior data, 2 days image-text object behavior data, 4 days image-text object behavior data, 7 days image-text object behavior data and the like. The generation time of each object behavior data carried by the day-level video object behavior data may be used to divide the day-level video object behavior data into 1 day video object behavior data, 2 days video object behavior data, 4 days video object behavior data, and 7 days video object behavior data. It can be understood that the generation times of the object behavior data included in the different time periods are not overlapped, and the division of the different time periods can be determined according to the actual application scenario, which is not limited herein, that is, the day-level object behavior data can be divided into 3-day-level object behavior data, 5-day-level object behavior data, 8-day-level object behavior data, and the like according to the actual application scenario, which is not limited herein. For convenience of description, the week-level graphic object behavior data may be divided into 12-week-level graphic object behavior data based on the generation time of each object behavior data carried by the week-level graphic object behavior data. The week-level object behavior data is object behavior data of a current operation object within the last 12 weeks, and comprises 12 weeks of graphic object behavior data and 12 weeks of video object behavior data. For example, when the current time is xx month xx day, the object behavior data within the 12 weeks is object behavior data of the object operated within 12 weeks before xx month xx day. It can be understood that the above-mentioned division of the object behavior data based on the generation time of each object behavior data may be determined according to the actual application scenario, and is not limited herein, that is, the above-mentioned week object behavior data may be divided into 2-week object behavior data and 4-week object behavior data according to the actual application scenario.

It can be understood that each of the above-mentioned operation objects is a successfully registered operation object in the above-mentioned target client, and obtains an object identifier (i.e., an application account number) for the target client. It should be understood that, since the time for which the operation behavior of the multimedia resource occurs after the real name registration of each operation object is completed on the target client is different, the generation time of the object behavior data of each operation object is also different, the number of operation objects is also different, and the number of operation objects and the generation time of the object behavior data of each operation object may be determined according to the actual application scenario, which is not limited herein. Referring to fig. 5, each of the operation objects carries at least one of the graphic object behavior data and the video object behavior data. For convenience of description, it is assumed that each operation object carries graphic object behavior data and video object behavior data. For example, assuming that the first object behavior data carries object behavior data of 4 different operation objects, the 4 different operation objects are named as operation object 1, operation object 2, operation object 3, and operation object 4, respectively. Assuming that the generation time of the object behavior data of each operation object has been determined, for convenience of description, the operation object 1 carries object behavior data for 14 days, the operation object 2 carries object behavior data for 2 days, the operation object 3 carries object behavior data for 4 days, and the operation object 4 carries object behavior data for 7 days. The target client divides the image-text object behavior data and the video object behavior data of each operation object belonging to the same time period into the object behavior data of the same level based on the generation time of the image-text object behavior data and the video object behavior data of each operation object. Specifically, based on the generation time of the graphic object behavior data carried by the operation object 1, the graphic object behavior data carried by the operation object 2, the graphic object behavior data carried by the operation object 3, and the graphic object behavior data carried by the operation object 4, the graphic object behavior data carried by each of the operation object 1, the operation object 2, the operation object 3, and the operation object 4 is divided into 1-day graphic-level object behavior data to generate the 1-day graphic-level object behavior data. Based on the image-text object behavior data carried by the operation object 1, the image-text object behavior data carried by the operation object 3 and the generation time of the image-text object behavior data carried by the operation object 4, the image-text object behavior data carried by each of the operation object 1, the operation object 3 and the operation object 4 is divided into 2-day image-text object behavior data so as to generate the 2-day image-text object behavior data. Based on the image-text object behavior data carried by the operation object 1 and the generation time of the image-text object behavior data carried by the operation object 4, dividing the image-text object behavior data carried by each of the operation object 1 and the operation object 4 into 4-day image-text object behavior data so as to generate the 4-day image-text object behavior data. Based on the generation time of the graphic object behavior data carried by the operation object 1, dividing the graphic object behavior data carried by the operation object 1 into 7-day graphic object behavior data so as to generate the 7-day graphic object behavior data. Similarly, based on the generation time of the video object behavior data carried by the operation object 1, the video object behavior data carried by the operation object 2, the video object behavior data carried by the operation object 3, and the video object behavior data carried by the operation object 4, the video object behavior data carried by each of the operation object 1, the operation object 2, the operation object 3, and the operation object 4 is divided into 1-day video-level object behavior data to generate the 1-day video-level object behavior data. Based on the video object behavior data carried by the operation object 1, the video object behavior data carried by the operation object 3, and the generation time of the video object behavior data carried by the operation object 4, the video object behavior data carried by each of the operation object 1, the operation object 3, and the operation object 4 is divided into 2-day video-level object behavior data to generate the 2-day video-level object behavior data. Dividing the video object behavior data carried by the operation object 1 and the operation object 4 into 4-day video-level object behavior data based on the video object behavior data carried by the operation object 1 and the generation time of the video object behavior data carried by the operation object 4, so as to generate the 4-day video-level object behavior data; dividing the video object behavior data carried by the operation object 1 into 7-day video-level object behavior data based on the generation time of the video object behavior data carried by the operation object 1, so as to generate the 7-day video-level object behavior data.

For easy understanding, the process of generating the behavior characteristics of each day-level image-text object will be exemplified below by the above-mentioned behavior data of each day-level image-text object. The process of generating the behavior features of each day-level video object based on the behavior data of each day-level video object is similar to that of generating each day-level image-text object, and is not repeated herein. It may be understood that at least two attribute features of the object attribute feature, the duration attribute feature, the resource content attribute feature, the resource type attribute feature, and the object resource behavior attribute feature of each operation object may be generated based on the day-level image-text object behavior data of each operation object included in the day-level image-text object behavior data. The number of dimensions of the object statistical feature generated based on the object behavior data is the product of the number of dimensions of each of the five attribute features. It will be appreciated that the five attribute features may include multiple dimensions, and the total number of dimensions of the statistical feature is the multiplication of the number of dimensions of the five attribute features. Wherein the object statistics feature comprises the five attribute features of which the dimension is one dimension, ultimately, a dimension of 1 (i.e., 1×1) x 1). When the target client generates the behavior data of each day-level image-text object, object attribute features, duration attribute features, resource content attribute features, resource type attribute features, object resource behavior attribute features and the like associated with each operation object are generated based on the day-level image-text object behavior data of each operation object carried by the behavior data of each day-level image-text object. Wherein the object property feature may comprise an object dimension, the duration property feature may comprise a long-term object behavior feature, the resource content property feature may comprise a secondary class resource content property feature, the resource type property feature may comprise a video multimedia resource property feature, and the object resource property feature may comprise one of a multimedia resource preference degree of an object, an object preference degree of a multimedia resource, and a desired click amount of the object on the multimedia resource. When the preference degree of the multimedia resource is generated, acquiring any one day image-text object behavior statistical value of any one of the operation objects in the day image-text object behavior data of any one of the operation objects included in the day image-text object behavior data, and determining the ratio of the any one day image-text object behavior statistical value to the sum of the day image-text object behavior statistical values of all the operation objects in the day image-text object behavior data to any one of the multimedia resources as the preference degree of any one of the operation objects to any one of the multimedia resources so as to obtain the preference degree of the multimedia resource of any one of the operation objects. When the object preference degree of any multimedia resource is generated, any object behavior statistical value of any operation object in the object behavior data of any operation object included in the image-text object behavior data of each day is obtained, and the ratio of any object behavior statistical value to the sum of the object behavior statistical values of any operation object in the image-text object behavior data of each day to all multimedia resources is determined as the preference degree of any operation object to any multimedia resource, so as to obtain the object preference degree of any multimedia resource. When the expected click rate of the operation object to the multimedia resource is generated, acquiring a first day image-text object behavior statistical value of any operation object in the day image-text object behavior data of any operation object included in the day image-text object behavior data of each day image-text object, and acquiring a first day image-text object behavior operation proportion of all operation objects in the day image-text object behavior data of each day to the first multimedia resource; acquiring a second-day image-text object behavior statistical value of any operation object in the day image-text object behavior data of each day image-text object, and acquiring a second-day image-text object behavior operation proportion of all operation objects in the day image-text object behavior data of each day to the second multimedia resource; generating the expected click rate of any operation object on the multimedia resources based on the first-day image-text object behavior statistical value, the first-day image-text object behavior operation proportion, the second-day image-text object behavior statistical value, the second-day image-text object behavior operation proportion and the image-text object behavior statistical values of any operation object on all the multimedia resources.

For example, it is assumed that the operation object 1 displays 10 times at the first location and 20 times at the 3 rd location on the multimedia resource, and the click rate of the multimedia resource at the first location among the all operation objects (i.e., the operation object 1, the operation object 2, the operation object 3, and the operation object 4) is 0.2 and the click rate at the 3 rd location is 0.1. Assuming that the multimedia resource is a group of nine-grid pictures, that is, the operation object 1 clicks 10 times on the picture in the first grid picture and clicks 20 times on the picture in the third grid picture, and the click rates of all the operation objects on the picture in the first grid picture and the picture in the third grid picture are respectively 0.2 and 0.1, the expected click rate of the operation object 1 on the multimedia resource is 4 times, that is, ec_click=10×0.2+20×0.1=4, and if the operation object 1 clicks 6 times on the nine-grid picture in total, the click exceeds the expected click (Click Over Expected Click, COEC) is 1.5 (that is, COEC =6/4=1.5).

For example, it is assumed that the one-dimensional statistical feature is the number of times that each operation object exposes to the target video secondary class content for a long period, that is, a one-dimensional statistical feature is generated based on the first object behavior data of each operation object, where the object attribute feature includes an object, the duration attribute feature includes a long period, the resource content attribute feature includes a secondary class, and the resource type attribute feature includes a video, and when the operation object selects the number of times of exposure, the expected click amount of the object in the object resource behavior data feature on the multimedia resource may be generated. Taking the 1-day image-text level object behavior data as an example, the image-text object behavior data of the operation object 1 carried by the 1-day image-text level object behavior data is used for generating object features, long-term object behavior features, secondary resource content attribute features, video multimedia resource attribute features and expected click rate of objects on multimedia resources associated with the operation object 1, and splicing five attribute features associated with the operation object 1 based on object identifications corresponding to the operation object 1 to generate the 1-day image-text level object behavior features associated with the operation object 1. Specifically, when the object resource behavior data feature of the operation object 1 is generated, a first day-level image-text object behavior statistic value of the operation object 1 on a first multimedia resource in the day-level image-text object behavior data of the operation object 1 included in the day-level image-text object behavior data of each of the day-level image-text object behavior data (i.e., the 1 day-level image-text object behavior data, the 2 day-level image-text object behavior data, the 4 day-level image-text object behavior data, and the 7 day-level image-text object behavior data) is obtained, and an operation proportion of the first day-level image-text object behavior of the first multimedia resource by all of the operation objects (i.e., the operation object 1, the operation object 2, the operation object 3, and the operation object 4) in the 1 day-level image-text object behavior data is obtained; acquiring second-day-level image-text object behavior statistics values of the operation object 1 in the day-level image-text object behavior data of the operation object 1 (namely, the 1-day-level image-text object behavior data, the 2-day-level image-text object behavior data, the 4-day-level image-text object behavior data and the 7-day-level image-text object behavior data), and acquiring second-day-level image-text object behavior operation proportions of all operation objects, namely, the operation object 1, the operation object 2, the operation object 3 and the operation object 4), in the day-level image-text object behavior data of the each day-level image-text object behavior data; generating the expected click rate of the operation object 1 on the multimedia resources on the top-level image-text side based on the first-level image-text object behavior statistics value, the first-level image-text object behavior operation proportion, the second-level image-text object behavior statistics value, the second-level image-text object behavior operation proportion of the operation object 1 and the top-level image-text object behavior statistics values of the operation object 1 on all the multimedia resources. Similarly, the expected click rate of the multimedia resource on the day-level graphics and texts corresponding to the operation object 2, the operation object 3 and the operation object 4 is generated based on the operation. After the target client generates five attribute features associated with the operation object 1 based on the 1-day-level image-text object behavior data of the operation object 1, the five attribute features associated with the operation object 1 are spliced based on the object identification of the operation object 1 to generate the 1-day-level image-text object behavior feature associated with the operation object 1. Similarly, 1-day-level object behavior features associated with the operation object 2, the operation object 3, and the operation object 4, respectively, are generated based on 1-day-level object data of other operation objects (i.e., the operation object 2, the operation object 3, and the operation object 4) carried by the 1-day-level object behavior data.

Similarly, object features, long-term object behavior features, secondary resource content attribute features, video multimedia resource attribute features and expected click quantity of objects on multimedia resources associated with the operation object 1 are generated based on the 2-day-level image-text object behavior data of the operation object 1 carried by the 2-day-level image-text object behavior data, and the five attribute features associated with the operation object 1 are spliced based on object identifications corresponding to the operation object 1 to generate the 2-day-level image-text object behavior features associated with the operation object 1. Similarly, 2-day-image-level object behavior features respectively associated with the operation object 3 and the operation object 4 are generated based on the 2-day-image-level object behavior data. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and expected click quantity of an object on a multimedia resource, which are associated with the operation object 1, based on the image-text object behavior data of the operation object 1 carried by the 4-day image-text object behavior data, and splicing five attribute characteristics associated with the operation object 1 based on the object identification corresponding to the operation object 1 so as to generate the 4-day image-text object behavior characteristics associated with the operation object 1. Similarly, 4-day teletext-level object behavior characteristics associated with the operation object 4 are generated based on the 4-day teletext-level object behavior data. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and expected click quantity of an object on a multimedia resource, which are associated with the operation object 1, based on the image-text object behavior data of the operation object 1 carried by the 7-day image-text object behavior data, and splicing five attribute characteristics associated with the operation object 1 based on the object identification corresponding to the operation object 1 so as to generate the 7-day image-text object behavior characteristics associated with the operation object 1.

In some possible embodiments, when the target client generates the daily graphic object behavior feature associated with each operation object, the daily graphic object behavior feature of each operation object is spliced based on the object identifier corresponding to each operation object, so as to generate the first object graphic behavior statistical feature associated with each operation object. Specifically, when the target client generates a 1-day teletext-level object behavior feature, the 2-day teletext-level object behavior feature, the 4-day teletext-level object behavior feature, and the 7-day teletext-level object behavior feature associated with the operation object 1, the 1-day teletext-level object behavior feature, the 2-day teletext-level object behavior feature, the 4-day teletext-level object behavior feature, and the 7-day teletext-level object behavior feature, based on the object identification corresponding to the operation object 1, are spliced to generate a daily teletext-level object behavior feature associated with the operation object 1. Similarly, the top-level graphic object behavior features associated with the operation object 2, the operation object 3, and the operation object 4, respectively, are generated. Similarly, the implementation process of generating the behavior features of the top-level video object associated with the operation object 1, the operation object 2, the operation object 3, and the operation object 4 based on the operation objects carried by the top-level graphic object behavior data is similar to the top-level graphic object behavior features, which are not repeated herein.

Further, a top-level object behavior statistical feature indexed by the object identification of each operation object is generated based on the top-level image-text object behavior feature associated with each operation object and the object identification of each operation object carried in the top-level video object behavior feature associated with each operation object. Specifically, the top-level graphic object behavior features generated by the target client include a top-level graphic object behavior feature associated with the operation object 1, a top-level graphic object behavior feature associated with the operation object 2, a top-level graphic object behavior feature associated with the operation object 3, and a top-level graphic object behavior feature associated with the operation object 4; the top-level video object behavior features generated by the target client include a top-level video object behavior feature associated with the operation object 1, a top-level video object behavior feature associated with the operation object 2, a top-level video object behavior feature associated with the operation object 3, and a top-level video object behavior feature associated with the operation object 4; and generating a top-level object behavior statistical feature with the object identification of the operation object 1 as an index based on the top-level image-text object behavior feature and the top-level video object behavior feature associated with the operation object 1. Similarly, based on the top-level graphic object behavior feature and the top-level video object behavior feature associated with each of the operation object 2, the operation object 3, and the operation object 4, a top-level object behavior statistical feature indexed by the object identification of each of the operation objects (i.e., the operation object 2, the operation object 3, and the operation object 4) is generated.

On the other hand, referring to fig. 5 again, the first object behavior data is divided into the top-level object behavior data and the bottom-level object behavior data based on the generation time of the object behavior data of each operation object (i.e., the operation object 1, the operation object 2, the operation object 3, and the operation object 4) carried by the first object behavior data, and the bottom-level object behavior data may be divided into the bottom-level graphic object behavior data and the bottom-level video object behavior data based on the type of the target multimedia resource (i.e., the video resource). Wherein, the graphics context is the graphics context resource corresponding to the video resource. It may be appreciated that the division of the first object behavior data based on the generation time of each object behavior data may be divided according to the actual application scenario, which is not limited herein. The week-level image-text object behavior data can be divided into 12-week-level image-text object behavior data based on the generation time of each object behavior data carried by the week-level image-text object behavior data. Similarly, the week-level video object behavior data may be divided into 12-week video object behavior data based on the generation time of each object behavior data carried by the week-level video object behavior data. It can be appreciated that the above-mentioned generation time based on each object behavior data may be determined according to the actual application scenario, and is not limited herein. And generating at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource behavior attribute features of each operation object based on the week object behavior data of each operation object included in the week object behavior data. The total dimension number of the statistical features is the product of the respective dimension numbers of the five attribute features. And when the target client generates the week-level image-text object behavior data and the week-level video object behavior data, generating 12-week image-text object behavior features and 12-week video object behavior features based on the week-level image-text object behavior data and the week-level video object behavior data, wherein the 12-week image-text object behavior features comprise the object attribute features, the duration attribute features, the resource content attribute features, the resource type attribute features and the object resource behavior attribute features. The object attribute features include object dimensions, the duration attribute features include long-term object behavior features, the resource content attribute features include secondary resource content attribute features, the resource type attribute features include video multimedia resource attribute features, and when the operation object selects the exposure times, expected click quantity of the object in the object resource behavior data features on the multimedia resource can be generated. Next, a process of generating 12-week object behavior features based on the above 12-week object behavior data will be described in detail. The process of generating the 12-week video object behavior feature based on the 12-week video object behavior data is similar to the process of generating the 12-week image-text object behavior feature, which is not repeated herein.

As is known in the art, the 12-week-object behavior data carries 12-week-object behavior data of the operation object 1, the operation object 2, the operation object 3, and the operation object 4. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and expected click quantity of objects on multimedia resources associated with the operation object 1 based on the 12-week image-text object behavior data of the operation object 1 carried by the 12-week image-text object behavior data, and splicing five attribute characteristics associated with the operation object 1 based on object identifications corresponding to the operation object 1 to generate 12-week image-text object behavior characteristics associated with the operation object 1. When generating the expected click rate of the object of the operation object 1 on the multimedia resource, acquiring a first 12-week-graphics-object-behavior statistic value of the operation object 1 on the first multimedia resource in the 12-week-graphics-object-behavior data of the operation object 1 included in the 12-week-graphics-object-behavior data, and acquiring a first 12-week-graphics-object-behavior operation proportion of all the operation objects (i.e., the operation object 1, the operation object 2, the operation object 3, and the operation object 4) in the 12-week-graphics-object-behavior data on the first multimedia resource; acquiring a second 12-week graphic object behavior statistic value of the operation object 1 on a second multimedia resource in 12-week graphic object behavior data of the operation object 1 in the 12-week graphic object behavior data, and acquiring a second 12-week graphic object behavior operation proportion of the operation object 1, the operation object 2, the operation object 3 and the operation object 4) on the second multimedia resource, which are all operation objects in the 12-week graphic object behavior data; generating an expected click rate of the operation object 1 on the 12-week-graphics-side multimedia resource based on the first 12-week-graphics-object behavior statistic value, the first 12-week-graphics-object behavior operation proportion, the second 12-week-graphics-object behavior statistic value, the second 12-week-graphics-object behavior operation proportion of the operation object 1, and the 12-week-graphics-object behavior statistic value of the operation object 1 on all multimedia resources. Similarly, the expected click rate of the 12-week-side multimedia resource corresponding to each of the operation object 2, the operation object 3, and the operation object 4 is generated based on the operation. After the target client generates five attribute features associated with the operation object 1 based on the 12-week graphic object behavior data of the operation object 1, the five attribute features associated with the operation object 1 are spliced based on the object identification of the operation object 1 to generate 12-week graphic object behavior features associated with the operation object 1.

Similarly, object features, long-term object behavior features, secondary resource content attribute features, video multimedia resource attribute features and expected click amounts of objects on multimedia resources associated with the respective operation objects are generated based on the operation object 2, the operation object 3 and the 12-week-graphics-object behavior data of the operation object 4 carried by the 12-week-graphics-object behavior data, and the five attribute features associated with the respective operation objects are spliced based on the object identifications corresponding to the respective operation objects to generate 12-week-graphics-level object behavior features associated with the operation object 2, 12-week-graphics-level object behavior features associated with the operation object 3 and 12-week-graphics-level object behavior features associated with the operation object 4. Similarly, the implementation process of generating the 12-week video object behavior feature associated with the operation object 1, the operation object 2, the operation object 3, and the operation object 4 based on the operation objects carried by the 12-week video object behavior data is similar to the 12-week-level graphic object behavior feature, which is not repeated here.

Further, a week-level object behavior statistical feature indexed by the object identification of each operation object is generated based on the 12-week graphic object behavior feature associated with each operation object and the object identification of each operation object carried in the 12-week video object behavior feature associated with each operation object. Specifically, the 12-week graphic object behavior feature generated by the target client includes a 12-week graphic object behavior feature associated with the operation object 1, a 12-week graphic object behavior feature associated with the operation object 2, a 12-week graphic object behavior feature associated with the operation object 3, and a 12-week graphic object behavior feature associated with the operation object 4; the 12-week video object behavior feature generated by the target client includes a 12-week video object behavior feature associated with the operation object 1, a 12-week video object behavior feature associated with the operation object 2, a 12-week video object behavior feature associated with the operation object 3, and a day-level video object behavior feature associated with the operation object 4; based on the 12-week graphic object behavior feature and the 12-week video object behavior feature associated with the operation object 1, a 12-week object behavior statistical feature indexed by the object identifier of the operation object 1 is generated. Similarly, based on the 12-week object behavior feature and the 12-week video object behavior feature associated with each of the operation object 2, the operation object 3, and the operation object 4, 12-week object behavior statistical features indexed by the object identifiers of the respective operation objects (i.e., the operation object 2, the operation object 3, and the operation object 4) are generated.

Further, a first object behavior statistical feature indexed by the object identifier of each operation object is generated based on the day-level object behavior statistical feature and the object identifier of each operation object carried by the 12-week object behavior statistical feature. Specifically, based on the day-level object behavior statistical feature and the 12-week object behavior statistical feature of the operation object 1, a first object behavior statistical feature associated with the operation object 1 with the object identification of the operation object 1 as an index is generated. Similarly, based on the day-level object behavior statistical features and the 12-week object behavior statistical features of the respective operation objects (i.e., operation object 2, operation object 3, and operation object 4), first object behavior statistical features associated with the respective operation objects, that is, first object behavior statistical features associated with the operation object 2, first object behavior statistical features associated with the operation object 3, and first object behavior statistical features associated with the operation object 4, respectively, are generated with the object identification of the respective operation objects as an index.

The feature generation method for multimedia resource recommendation provided by the embodiment of the application can increase the statistical feature by increasing the attribute feature or increasing the dimension of a certain attribute feature, can be specifically determined according to the actual application scene, is not limited herein, can provide a more comprehensive and expanded statistical feature using mode, and has wide application range.

S102, acquiring second object behavior data provided by a training sample data source of multimedia resource recommendation, and generating at least two second object behavior statistical features of at least two different time periods associated with a second object based on the second object behavior data, wherein at least two object resource behavior attribute features in the second object behavior statistical features of any time period associated with any second object.

In some possible embodiments, the operation object performs real-name information registration through the target client loaded on the terminal device 200b, and after the operation object obtains the object identifier uniquely corresponding to the operation object through the target client, obtains the second object behavior data provided by the training sample data source recommended by the multimedia resource through the target client loaded on the terminal device 200 b. The training sample data source recommended by the multimedia resource is a data sample of each terminal user who has completed registration at the target client, or sample data of the multimedia resource issued at the target client, which can be specifically determined according to an actual application scenario, and is not limited herein. Referring to fig. 6, fig. 6 is a schematic diagram of another application scenario of the feature generation method of multimedia resource recommendation according to the embodiment of the present application. When the target client acquires second object behavior data provided by a training sample data source recommended by the multimedia resource, dividing the second object behavior data into a plurality of P ₁ -level object behavior data based on generation time of the second object behavior data carried in the second object behavior data, wherein the P ₁ -level object behavior data is object behavior data included in minimum unit time periods of time interval division of the second object behavior data. The second object is each object carried in the first object behavior data. It may be understood that the second object may include each object carried in the behavior data of the first object, or may be an object in a training sample, where the second object may be determined according to an actual application scenario, and the application is not limited herein. Here, for convenience of description, it is assumed that the second object includes an operation object carried in the first object behavior data, that is, an operation object 1, the operation object 3, and the operation object 4, and an operation object 5, which are the operation objects 1,2,3, and 4 carried in the first object behavior data mentioned above. The dividing manner of the second object behavior data according to the time interval can be determined according to the actual application scenario, and the application is not limited herein, that is, the P ₁ level object behavior data may be hour level object behavior data or two hour level object behavior data. For convenience of description herein, the above P ₁ -level object behavior data is determined as hour-level object behavior data. At least two P ₂ -level object behavior data of at least two different time periods are generated based on the accumulation of the plurality of P ₁ -level object behavior data included in the plurality of minimum unit durations. The P ₂ -level object behavior data is accumulated by a plurality of P ₁ -level object behavior data within a target duration, where the target duration is a positive integer multiple of the minimum unit duration. It will be appreciated that the target duration may be determined according to an actual application scenario, and the present application is not limited herein, for example, the target duration may be 24 hours or one week. Here, for convenience of description, the target period is set to 24 hours. It can be understood that the duration of the P ₂ -level object behavior data can be determined according to the actual application scenario, and the present application is not limited herein, and may be 1-week-level object behavior data or 2-week-level object behavior data. For convenience of description herein, the above-mentioned at least two P ₂ -level object behavior data are determined as the day-level object behavior data and the week-level object behavior data, such as 24-hour day-level object behavior data, 1 week, 2 weeks or 4 weeks of week-level object behavior data. Specifically, the target client divides the second object behavior data into a plurality of hour-level object behavior data based on the generation time of each object behavior data carried in the second object behavior data, where the hour-level object behavior data is object behavior data included in a minimum unit time length divided by time intervals of the second object behavior data. For example, assuming that the current time is xx month xx day, the hour-level object behavior data is object behavior data of each operation object preceding xx month xx day. When the accumulated time length of the target client based on the plurality of the hour-level object behavior data included in the plurality of minimum unit time lengths reaches 24 hours (i.e., 24 points per day), generating day-level object behavior data including the time of xx, month, xx, and day (hereinafter, abbreviated as the day for convenience of description). And generating 1-week-level object behavior data (namely object behavior data of eight days in total) including the current day when the accumulated time length of the day-level object behavior data including the current day of the time xx month xx reaches one week. Based on the day-level object behavior data generated when the cumulative time period of the plurality of the hour-level object behavior data reaches 24 hours (i.e., 24 points per day), 1-week-level object behavior data, 2-week-level object behavior data, and 4-week-level object behavior data are generated, respectively. Wherein the generation time of the 1-week-level object behavior data including the current day, the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data do not overlap. Based on the object identifications corresponding to the respective operation objects (i.e., the operation object 1, the operation object 3, the operation object 4, and the operation object 5) carried in the 1-week-level object behavior data including the current day, the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data, the respective week-level object behavior statistics feature indexed by the object identifications of the respective operation objects is generated. It is understood that at least two attribute features of the object attribute feature, the duration attribute feature, the resource content attribute feature, the resource type attribute feature, and the object resource attribute feature of each operation object are generated based on the week object behavior data of each operation object included in the week object behavior data. The total dimension number of the statistical features is the product of the respective dimension numbers of the five attribute features. It will be appreciated that the five attribute features may include multiple dimensions, and the total number of dimensions of the statistical feature is the multiplication of the number of dimensions of the five attribute features. When the target client generates the week object behavior data, generating object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource behavior attribute features of the operation objects based on the week object behavior data of the operation objects carried by the week object behavior data. The object attribute features include object dimensions, the duration attribute features include long-term object behavior features, the resource content attribute features include secondary resource content attribute features, the resource type attribute features include video multimedia resource attribute features, and the object resource behavior attribute features include at least one object resource behavior association feature of clicking duration, clicking times, clicking rates, exposure times, watching duration and interaction duration for a target multimedia resource. The clicking times for the multimedia resource are the times of the operation object clicking to watch the multimedia resource in a certain period of time. The clicking rate for the multimedia resource is the ratio of the clicking times of the operation object to the target multimedia resource to the exposure times of the target multimedia resource. The viewing time length for the multimedia resource is the total viewing time length of the operation object for the target multimedia resource. The interaction time length for the multimedia resource is the interaction of the operation object to the target multimedia resource, such as praise, comment, share and the like. And when the click time length of the target multimedia resource is larger than a certain threshold value, generating a long click (long CLICK RATE, LCR) of the target multimedia resource. It will be appreciated that the above threshold may be determined according to the actual application scenario, and the present application is not limited herein. Based on the object resource behavior attribute characteristics including at least one object resource behavior-associated characteristic of a click time length, a click number, a click rate, an exposure number, a viewing time length, and an interaction time length for a target multimedia resource, a completion (play complete rate, PCR) for the target multimedia resource may be generated. The completion degree may be a ratio of a viewing time period of the operation object for the target multimedia resource (e.g., video resource) to a total time period of the target multimedia resource, or a ratio of the operation object for the reference picture in the target multimedia resource (e.g., picture resource) in the target multimedia resource, where an actual meaning of the completion degree may be determined according to an actual application scenario, and the application is not limited herein. It may be appreciated that, in the embodiment of the present application, the target client generates at least two object resource behavior attribute features in each level of object behavior features associated with the object identifiers of the respective operation objects based on the object resource behavior attribute features including at least one feature associated with the object resource behavior for the target multimedia resource, including a click time, a click frequency, a click rate, an exposure time, a viewing time, and an interaction time. The behavior attribute features of the at least two objects are the long click and the completion degree of each operation object aiming at the target multimedia resource. It can be understood that each of the operation objects may carry a long click on the target multimedia resource, may carry a completion degree on the target multimedia resource, and may also carry a long click and a completion degree on the target multimedia resource. For convenience of description, it is assumed that each of the above-mentioned operation objects in the embodiment of the present application carries a long click and a completion degree for the above-mentioned target multimedia resource.

For example, it is assumed that when the target client obtains object behavior data of each of the operation objects (i.e., the operation object 1, the operation object 3, the operation object 4, and the operation object 5) carried in second object behavior data provided by a training sample data source for multimedia resource recommendation, the object behavior data carried by each of the operation objects is divided into a plurality of hour-level object behavior data based on generation time of the object behavior data of each of the operation objects carried in the second object behavior data. The plurality of hour-level object behavior data are object behavior data included in a minimum unit time length divided by the second object behavior data according to time intervals. Specifically, the object behavior data of the operation object 1 is divided into a plurality of hour-level object behavior data based on the generation time of the object behavior data of the operation object 1 carried in the second object data. Specifically, assuming that the current time is xx month xx day, the hour-level object behavior data is object behavior data of each operation object preceding xx month xx day. When the target client reaches a target time period based on the time period of the accumulation of the hour-level object behavior data of the operation object 1, it generates day-level object behavior data (hereinafter, simply referred to as day-level object behavior data including the current day) including the current day of the time xx month xx for the operation object 1. Wherein the target time length is a positive integer multiple of the minimum unit time length. Here, for convenience of description, when the target time period is determined to be 24 hours, that is, 24 points per day, day-level object behavior data including the current day for the operation object 1 is generated. The target client may accumulate the day-containing day-level object behavior data based on the day-containing day-level object behavior data for the operation object 1 when the target client generates the day-level object behavior data for the operation object 1, and may generate the day-containing 1-week-level object behavior data (i.e., eight days total object behavior data) for the operation object 1 when the accumulated time length of the day-containing day-level object behavior data reaches one week. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time length associated with the operation object 1 based on the 1-week-class object behavior data including the current day for the operation object 1, and splicing five attribute characteristics associated with the operation object 1 based on the object identification corresponding to the operation object 1 to generate the 1-week-class object behavior characteristics including the current day associated with the operation object 1. Wherein the 1-week-level object behavior feature including the current day associated with the operation object 1 includes at least two object resource behavior attribute features of the operation object 1. Specifically, the long click and the completion degree of the operation object 1 for the target multimedia resource are respectively generated based on the click time length of the operation object 1 for the target multimedia resource.

Similarly, the object behavior data of the operation object 1 is divided into a plurality of hour-level object behavior data based on the generation time of the object behavior data of the operation object 1 carried in the second object data. Specifically, assuming that the current time is xx month xx day, the hour-level object behavior data is object behavior data of each operation object one week before xx month xx day. When the target client reaches a target time period based on the hour-level object behavior data accumulation time period of the operation object 1, generating day-level object behavior data for the operation object 1. Wherein the target time length is a positive integer multiple of the minimum unit time length. Here, for convenience of description, the target time period is determined to be 24 hours, that is, 24 points per day, and day-level object behavior data for the operation object 1 is generated. The target client accumulates the day-level object behavior data based on the day-level object behavior data for the operation object 1 when the target client generates the day-level object behavior data for the operation object 1, and generates 1-week-level object behavior data (i.e., object behavior data for seven days in total) for the operation object 1 when the accumulated time length for the day-level object behavior data reaches one week. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time length associated with the operation object 1 based on the 1-week-level object behavior data for the operation object 1, and splicing five attribute characteristics associated with the operation object 1 based on object identifications corresponding to the operation object 1 to generate 1-week-level object behavior characteristics associated with the operation object 1. Wherein the 1-week-level object behavior feature associated with the generated operation object 1 includes at least two object resource behavior attribute features of the operation object 1. Specifically, the long click and the completion degree of the operation object 1 for the target multimedia resource are respectively generated based on the click time length of the operation object 1 for the target multimedia resource.

Similarly, based on the object behavior data of the operation object 1 carried in the second object data, 2-week-level object behavior data (i.e., object behavior data for seven days in total) for the operation object 1 is generated. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time length associated with the operation object 1 based on the 2-week-level object behavior data for the operation object 1, and splicing five attribute characteristics associated with the operation object 1 based on object identifications corresponding to the operation object 1 to generate 2-week-level object behavior characteristics associated with the operation object 1. Wherein the 2-week-level object behavior feature associated with the generated operation object 1 includes at least two object resource behavior attribute features of the operation object 1. Specifically, the long click and the completion degree of the operation object 1 for the target multimedia resource are respectively generated based on the click time length of the operation object 1 for the target multimedia resource. Based on the object behavior data of the operation object 1 carried in the second object data, 4-week-level object behavior data (i.e., object behavior data for seven days in total) for the operation object 1 is generated. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time length associated with the operation object 1 based on the 4-week-level object behavior data for the operation object 1, and splicing five attribute characteristics associated with the operation object 1 based on object identifications corresponding to the operation object 1 to generate the 4-week-level object behavior characteristics associated with the operation object 1. Wherein the 4-week-level object behavior feature associated with the generated operation object 1 includes at least two object resource behavior attribute features of the operation object 1. Specifically, the long click and the completion degree of the operation object 1 for the target multimedia resource are respectively generated based on the click time length of the operation object 1 for the target multimedia resource. Wherein the generation time of the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data for the operation object 1 including the current day do not overlap.

Similarly, based on the object behavior data of the operation object 3 carried in the second object data, 1 week-level object behavior data, 2 week-level object behavior data, and 4 week-level object behavior data including the current day for the operation object 3 are generated. The generation times of the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data of the operation object 3 including the current day do not overlap. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time associated with the operation object 3 based on the 1-week object behavior data including the current day, the 1-week object behavior data, the 2-week object behavior data and the 4-week object behavior data for the operation object 3, and concatenating five attribute characteristics associated with the operation object 3 based on the object identification corresponding to the operation object 3 to generate the 1-week object behavior characteristics including the current day, the 1-week object behavior characteristics, the 2-week object behavior characteristics and the 4-week object behavior characteristics associated with the operation object 3. Wherein, the object behavior characteristics of each level associated with the generated operation object 3 include at least two object resource behavior attribute characteristics of the operation object 3. Specifically, the long click and the completion degree for the target multimedia resource in each level of object behavior feature are respectively generated based on the click time length for the target multimedia resource in each level of object behavior feature of the operation object 3.

Similarly, based on the object behavior data of the operation object 4 carried in the second object data, 1 week-level object behavior data, 2 week-level object behavior data, and 4 week-level object behavior data including the current day for the operation object 4 are generated. The generation times of the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data of the operation object 4 including the current day do not overlap. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time associated with the operation object 4 based on the 1-week object behavior data including the current day, the 1-week object behavior data, the 2-week object behavior data and the 4-week object behavior data for the operation object 4, and concatenating five attribute characteristics associated with the operation object 4 based on the object identification corresponding to the operation object 4 to generate the 1-week object behavior characteristics including the current day, the 1-week object behavior characteristics, the 2-week object behavior characteristics and the 4-week object behavior characteristics associated with the operation object 4. Wherein, the object behavior characteristics of each level associated with the generated operation object 4 include at least two object resource behavior attribute characteristics of the operation object 4. Specifically, the long click and the completion degree for the target multimedia resource in the object behavior feature of each level are respectively generated based on the click time length for the target multimedia resource in the object behavior feature of each level of the operation object 4.

Similarly, based on the object behavior data of the operation object 5 carried in the second object data, 1 week-level object behavior data, 2 week-level object behavior data, and 4 week-level object behavior data including the current day for the operation object 5 are generated. The generation times of the 1-week-level object behavior data, the 2-week-level object behavior data, and the 4-week-level object behavior data of the operation object 5 including the current day do not overlap. Generating object characteristics, long-term object behavior characteristics, secondary resource content attribute characteristics, video multimedia resource attribute characteristics and click time associated with the operation object 5 based on the 1-week object behavior data including the current day, the 1-week object behavior data, the 2-week object behavior data and the 4-week object behavior data for the operation object 5, and concatenating five attribute characteristics associated with the operation object 5 based on the object identification corresponding to the operation object 5 to generate the 1-week object behavior characteristics including the current day, the 1-week object behavior characteristics, the 2-week object behavior characteristics and the 4-week object behavior characteristics associated with the operation object 5. Wherein, the object behavior characteristics of each level associated with the generated operation object 5 include at least two object resource behavior attribute characteristics of the operation object 5. Specifically, the long click and the completion degree for the target multimedia resource in the object behavior feature of each level are respectively generated based on the click time length for the target multimedia resource in the object behavior feature of each level of the operation object 5.

When the target client generates object behavior characteristics of each level associated with each operation object based on the object behavior data of each operation object carried by the second object behavior data, at least two object resource behavior attribute characteristics of each level of object behavior characteristics associated with the object identifications of each operation object are spliced to generate second object behavior statistical characteristics associated with each operation object identification. Specifically, the object identifier of the operation object 1 is used as an index, and the long click and the completion degree of the target multimedia resource, which are included in the 1-week-level object behavior feature including the current day and associated with the operation object 1, are spliced. Similarly, with the object identifier of the operation object 1 as an index, the long click and the completion degree for the target multimedia resource included in the 1-week-level object behavior feature associated with the operation object 1 are spliced. Similarly, with the object identifier of the operation object 1 as an index, the long click and the completion degree for the target multimedia resource included in the 2-week-level object behavior feature associated with the operation object 1 are spliced. Similarly, with the object identifier of the operation object 1 as an index, the long click and the completion degree for the target multimedia resource included in the 4-week-level object behavior feature associated with the operation object 1 are spliced. And respectively splicing the long click and the completion degree for the target multimedia resource, which are included in the 1 week object behavior feature, the 2 week object behavior feature and the 4 week object behavior feature which are related by the operation object 3 and contain the current day, by taking the object identification of the operation object 3 as an index. And respectively splicing the long click and the completion degree for the target multimedia resource, which are included in the 1 week object behavior feature, the 2 week object behavior feature and the 4 week object behavior feature, which are associated with the operation object 4 and include the current day, by using the object identifier of the operation object 4 as an index. And respectively splicing the long click and the completion degree for the target multimedia resource, which are included in the 1 week object behavior feature, the 2 week object behavior feature and the 4 week object behavior feature which are related by the operation object 5 and contain the current day, by taking the object identification of the operation object 5 as an index.

Referring to fig. 6, based on the object identifiers of the operation objects, object behavior features of each stage associated with the operation objects are spliced according to the object identifiers of the operation objects, so as to generate second object behavior statistical features associated with the object identifiers of the operation objects. Specifically, based on the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature associated with the operation object 1, the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature are concatenated with the object identification of the operation object 1 as an index to generate a second object behavior statistical feature associated with the object identification of the operation object 1. Similarly, based on the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature, which are associated with the operation object 3, the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature are concatenated with the object identification of the operation object 3 as an index to generate a second object behavior statistical feature associated with the object identification of the operation object 3. Similarly, based on the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature, which are associated with the operation object 4, the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature are concatenated with the object identification of the operation object 4 as an index to generate a second object behavior statistical feature associated with the object identification of the operation object 4. Similarly, based on the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature associated with the operation object 5, the current day-containing 1-week-level object behavior feature, the 2-week-level object behavior feature, and the 4-week-level object behavior feature are concatenated with the object identification of the operation object 5 as an index to generate a second object behavior statistical feature associated with the object identification of the operation object 5.

And S103, generating object statistical characteristics of each object based on the first object behavior statistical characteristics related to each first object and the second object behavior statistical characteristics related to each second object, and determining input characteristics of a multimedia resource recommendation model by the object statistical characteristics, wherein the multimedia resource recommendation model is obtained by training the second object behavior data, and the multimedia resource model is used for outputting multimedia resource recommendation values of each object based on the object statistical characteristics of each object.

Referring to fig. 7, fig. 7 is a schematic diagram of another application scenario of the feature generation method of multimedia resource recommendation according to the embodiment of the present application. In some possible embodiments, after acquiring the first object behavior data provided by the target client data source, the target client generates first object behavior statistics associated with the first objects based on the object behavior data of the first objects carried by the first object behavior data. And after acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource, the target client generates second object behavior statistical characteristics associated with the second object based on the object behavior data of the second object carried by the second object behavior data. The target client obtains the object identifiers corresponding to the first objects from the first object behavior statistical characteristics, and obtains the object identifiers corresponding to the second objects from the second object behavior statistical characteristics. Specifically, the target client obtains, from the first object behavior statistical feature, object identifiers corresponding to the operation object 1, the operation object 2, the operation object 3, and the operation object 4, respectively. The target client obtains object identifiers corresponding to the operation object 1, the operation object 3, the operation object 4, and the operation object 5, respectively, from the second object behavior statistical feature. And the target client uses the object identification of each first object and the object identification of each second object as indexes, and splices the first object behavior statistical feature and the second object behavior statistical feature to generate object statistical features corresponding to each object identification. And the first object behavior statistical feature and the second object behavior statistical feature with the same object identification are spliced into the object statistical feature of the same object. It is to be understood that each of the operation objects may have only the first object behavior statistical feature associated with the operation object, may have only the second object behavior statistical feature associated with the operation object, and may have both the first object behavior statistical feature and the second object behavior statistical feature associated with the operation object. Specifically, the target client concatenates the first object behavior statistical feature and the second object behavior statistical feature associated with the operation object 1 with the object identifier corresponding to the operation object 1 as an index, so as to generate an object statistical feature corresponding to the object identifier of the operation object 1. The target client generates an object statistical feature corresponding to the object identifier of the operation object 2 based on the first object behavior statistical feature associated with the operation object 2 with the object identifier corresponding to the operation object 2 as an index. The target client uses the object identifier corresponding to the operation object 3 as an index, and concatenates the first object behavior statistical feature and the second object behavior statistical feature associated with the operation object 3 to generate an object statistical feature corresponding to the object identifier of the operation object 3. The target client uses the object identifier corresponding to the operation object 4 as an index, and concatenates the first object behavior statistical feature and the second object behavior statistical feature associated with the operation object 4 to generate an object statistical feature corresponding to the object identifier of the operation object 4. The target client generates an object statistical feature corresponding to the object identifier of the operation object 5 based on a second object behavior statistical feature associated with the operation object 5 with the object identifier corresponding to the operation object 5 as an index.

When the target client generates an object statistical feature associated with each operation object based on the first object behavior statistical feature and the second object behavior statistical feature of each operation object, the object identification of each operation object is used as an index, the object behavior feature associated with the object identification of each operation object is used as an input feature for determining a multimedia resource recommendation model, the object identification of each operation object is used as an index, and the object behavior feature associated with each operation object is written into a statistical feature storage unit. The feature extraction and online sequencing service can read the object statistical features in the statistical feature storage unit, score through a sequencing model of multimedia resource recommendation, output the multimedia resource recommendation value of each operation object, and recommend target multimedia resources to the operation object with the multimedia resource recommendation value of each operation object larger than a certain threshold. The multimedia resource recommendation model is trained by the second object behavior data, and is used for outputting the multimedia resource recommendation value of each operation object based on the object statistical characteristics of each operation object. Based on each multimedia resource recommendation value, recommending the target multimedia resource to the user, then updating the training sample data of the scene of the user, and continuously using the updated training sample data for model training, thereby improving the expression capacity of the model and enhancing the utilization rate of the data resource. Optionally, the multimedia resource recommended values of the operation objects are output according to the object statistics characteristics of the operation objects, and after the target client obtains the multimedia resource recommended values of the operation objects, the multimedia resources may be ranked based on the multimedia resource recommended values of the operation objects. Further, the target client may order the multimedia resources of each operation object from high to low according to the recommendation value of the multimedia resource of each operation object to obtain a multimedia resource recommendation list for each operation object, and sequentially recommend the top n multimedia resources to each operation object in a manner of from high to low according to the recommendation value of the multimedia resource recommendation list, wherein n is a positive integer.

Optionally, after the target client obtains the multimedia resource recommendation values of the respective operation objects, the respective multimedia resources may be ranked based on the multimedia resource recommendation values of the respective operation objects. The target client may sort the multimedia resources of each operation object according to the recommendation value from high to low based on the multimedia resource recommendation value of each operation object to obtain a multimedia resource recommendation list for each operation object, and extract the first n multimedia resources of the front order based on the multimedia resource recommendation list, where n is a positive integer, so as to improve the diversity of the multimedia resource recommendation list.

In some possible embodiments, based on the implementation provided in steps S101 to S103, the filtered object statistics of 335 dimensions may be generated, including video side client statistics 260 dimensions of the object, and long click completion statistics 75 dimensions within the scene of the training sample. For convenience of description, terms appearing in the following examples will be explained:

pv: exposure times; clk: the number of clicks; longClk: the number of long clicks; crSum: sum of degrees of completion; time: playing time length; ec: the number of clicks desired; allCtr: the total click rate of the current user in a certain time period; allLongCtr: the total long click rate of the current user in a certain time period; allCrAvg: the total average completion of the current user in a certain time period; ALLTIMEPERPV: the time average playing duration of the current user in a certain time period; sumCtr: the total click rate of all users in a certain time period; sumLongCtr: the total long click rate of all users in a certain time period; sumCrAvg: the total average completion of all users in a certain time period; sumTimePerPv: the time average playing time length of all users in a certain time period; pvFea: min (ln (pv+1), 11.0)/11.0, represents the number of exposures and normalization thereof; ctrFea: min (clk/(pv+0.01), 1.0) represents click rate and normalization thereof; longCtrFea: min (longClk/(pv+0.01), 1.0), representing the long click rate and its normalization; crAvgFea: min (crSum/(pv+0.01), 5.0)/5.0, representing the average completion per exposure and normalization; TIMEPERPVFEA: min (time/(pv+0.01), 2000)/2000, represents the average play duration per exposure and its normalization; coecFea: min (clk/(ec+0.01), 10.0)/10.0 represents the feature of the operation object COEO and its normalization.

Inner preference degree:

allCoec: min (clk/(pv. AllCtr +0.01), 10.0)/10.0, represents the click preference degree and normalization of the current operation object to the current content for the operation object;

allLoel: min (longClk/(pv) allLongCtr +0.01), 10.0)/10.0, which represents the long click preference degree of the current operation object on the current content for the operation object, and normalizes;

allRoer: min (crSum/(pv) allCrAvg +0.01), 10.0)/10.0, which represents the degree of completion preference of the current operation object for the current content and its normalization for the operation object;

allToet: min (time/(pv. ALLTIMEPERPV +0.01), 10.0)/10.0 represents the preference degree and normalization of the playing time of the current operation object to the current content for the operation object.

Inter preference degree:

sumCoec: min (clk/(pv× sumCtr +0.01), 10.0)/10.0, which represents the click preference degree of the current operation object on the current content and its normalization among all operation objects;

sumLoel: min (longClk/(pv) sumLongCtr +0.01), 10.0)/10.0, which represents the long click preference degree of the current operation object on the current content and its normalization among all operation objects;

sumRoer: min (crSum/(pv) sumCrAvg +0.01), 10.0)/10.0, which represents the degree of completion preference of the current operation object for the current content and its normalization among all operation objects;

sumToet: min (time/(pv. SumTimePerPv +0.01), 10.0)/10.0 represents the preference degree of the playing time length of the current operation object to the current content and the normalization thereof in all operation objects.

For example, assume that the object uses the target client for 7 months and 15 days, wherein the video statistics are 150 dimensions, as described in detail below:

0 th-5 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object within 1 day of 7 months and 14 days;

6 th-11 th dimension: pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object within 2 days of 7 months 12 days-7 months 14 days;

12 th-17 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object within 4 days of 7 months 8 days-7 months 11 days;

18 th-24 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object in 7 days of 7 months 1 day to 7 months 7 days;

24 th-29 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation subject within the last 12 weeks;

40 th-47 th dimension: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet, in 1 day of 7 months and 14 days, the current operation object is classified according to the first class of the current video;

48 th-45 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first class classification of the current video within 2 days of 7-14 days of the current operation object in the time period of 7-12 months;

46-54 dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first class classification of the current video within 4 days of 7-8-7-11 days of the current operation object;

54 th-61 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first class classification of the current video within 7 days of 7 months 1 day to 7 months 7 days;

Dimension 62-69: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of secondary classification of the current video within 1 day of 7 months and 14 days;

70 th-77 th dimension: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of secondary classification of the current video within 2 days of 7-14 days of the current operation object in the time period of 7-12 months;

78 th to 85 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of secondary classification of the current video within 4 days of 7-11 days of the current operation object in the time period of 7-8 days;

86-94: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of secondary classification of the current video within 7 days of 7 months 1 day to 7 months 7 days;

94 th-101 th dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current video author within 1 day of 7 months and 14 days;

102-109 dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current video author in 2 days of 7 months 12 days-7 months 14 days;

110 th-117 th dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current video author in 4 days of 7 months 8 days-7 months 11 days;

118-125 dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current video author in 7 days of 7 months 1 day-7 months 7 days;

126 th to 144 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet, in the last 12 weeks, the current operation object classifies the current video to which the current video belongs at a first level;

144 th to 141 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet, in the last 12 weeks, the current operation object belongs to the secondary classification of the current video;

142 th to 159 th dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current video author in the last 12 weeks.

Suppose that the operation object uses the target client time to be 7 months and 15 days, wherein the video corresponds to 110 dimensions of the graphic statistics feature, as described in detail below:

0 th-5 th dimensions: graphic text pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object within 1 day of 7 months and 14 days;

6 th-11 th dimension: graphic text pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object within 2 days of 7 months 12 days to 7 months 14 days;

12 th-17 th dimensions: the current operation object has pictures and texts pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet in 4 days from 7 months 8 days to 7 months 11 days;

18 th-24 th dimensions: the current operation object has pictures and texts pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet in 7 days from 7 months 1 day to 7 months 7 days;

24 th-29 th dimensions: the image-text pvFea, ctrFea, timePerPvFea, coecFea, sumCoec, sumToet of the current operation object in the last 12 weeks;

40 th-47 th dimension: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first-level classification of the graphics and texts corresponding to the current video within 1 day of 7 months and 14 days;

48 th-45 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first-level classification of the pictures and texts corresponding to the current video in 2 days of the total of 7 months 12 days to 7 months 14 days of the current operation object;

46-54 dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first-level classification of the pictures and texts corresponding to the current video in 4 days of 7-8-7-11 days of the current operation object;

54 th-61 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the first-level classification of the pictures and texts corresponding to the current video in 7 days from 7 months 1 day to 7 months 7 days;

Dimension 62-69: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet, performing secondary classification on graphics and texts corresponding to the current video within 1 day of 7 months and 14 days;

70 th-77 th dimension: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the image-text secondary classification corresponding to the current video in 2 days of the total of 7 months 12 days to 7 months 14 days of the current operation object;

78 th to 85 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the image-text secondary classification corresponding to the current video in 4 days of 7-11 days of the current operation object in the time period of 7-8 days;

86-94: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the image-text secondary classification corresponding to the current video in 7 days of 7 months 1-7 months 7 days;

94 th-101 th dimensions: pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the current operation object, which is used for carrying out primary classification on the graphics and texts corresponding to the current video in the last 12 weeks;

102-109 dimensions: the current operation object is pvFea, ctrFea, timePerPvFea, coecFea, allCoec, allToet, sumCoec, sumToet of the two-stage classification of the corresponding graphics context of the current video in the last 12 weeks.

Assume that the object uses the target client time to be 7 months and 15 days, wherein the long click, the completion degree and the duration statistics in the scene are 75 dimensions, and the details are as follows:

0 th-4 th dimension: pvFea, longCtrFea, timePerPvFea, crAvgFea of the current operation object within 8 days from 7 months 8 days to 7 months 15 days;

4 th-7 th dimensions: pvFea, longCtrFea, timePerPvFea, crAvgFea of 7 days of 7 months 1 to 7 months 7 of the current operation object;

8 th-14 th dimensions: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of the current operation object, which is the first class classification of the current video within 8 days from 7 months 8 days to 7 months 15 days;

15 th-21 st dimension: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of secondary classification of the current video within 8 days of 7 months 8 days to 7 months 15 days;

22 th-28 th dimensions: the current operation object is pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of the current video author in 8 days from 7 months 8 days to 7 months 15 days;

29 th-45 th dimensions: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of the current operation object, which is to be classified into the first class of the current video within 7 days from 7 months 1 day to 7 months 7 days;

46 th to 42 th dimensions: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of secondary classification of the current video within 7 days from 7 months 1 day to 7 months 7 days;

44 th to 49 th dimensions: the current operation object is pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of the current video author in 7 days from 7 months 1 day to 7 months 7 days;

50 th-54 th dimension: pvFea, longCtrFea, timePerPvFea, crAvgFea of the current operation object within 14 days from 17 days of 6 months to 40 days of 6 months;

54 th-60 th dimension: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of the first class classification of the current video within 14 days of the current operation object from 17 days of 6 months to 40 days of 6 months;

dimensions 61-67: pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer of secondary classification of the current video within 14 days of the current operation object from 17 days of 6 months to 40 days of 6 months;

68 th-74 th dimensions: the current user is pvFea, longCtrFea, timePerPvFea, crAvgFea, allLoel, allToet, allRoer for the current video author within 14 days from day 6, 17 to day 6, 40.

In some possible implementations, after generating the object statistics feature based on the implementation manner provided by the foregoing embodiments, the object statistics feature may be further spliced with the output feature of the last network layer of the multimedia resource recommendation model by a statistics feature convolution layer to be used as the estimated feature for multimedia resource recommendation, and specifically, fig. 8 may be referred to fig. 8, where fig. 8 is another application scenario schematic diagram of the feature generation method for multimedia resource recommendation provided by the embodiment of the present application. The multimedia resource recommendation model can splice the embedded features with the object side content side cross features, the embedded features with the object side features and the embedded features with the content side features, and generate the network layer output features of the multimedia resources through the network layer of the multimedia resource recommendation model. After the target client generates the object statistical features (i.e., 335-dimensional object behavior features) associated with the operation objects, the object statistical features associated with the operation objects pass through a statistical feature convolution layer and are spliced with the network layer output features of the multimedia resource recommendation model to serve as recommendation value estimation features of the multimedia resources, so as to obtain recommendation values of the multimedia resources. The feature generation method of the multimedia resource recommendation provided by the embodiment of the application not only can optimize the multimedia resource recommendation model, but also can improve the accuracy of the object statistical features associated with each operation object.

In some possible implementations, after the object statistics feature is generated based on the implementation provided by the foregoing embodiment, the object statistics feature may also be interacted with other important features, which may be determined according to an actual scene application, and is not limited herein. Specifically, after the target client generates the object statistics feature associated with each operation object, the object statistics feature associated with each operation object is interacted with the multimedia resource identification feature to generate an interaction feature, and the interaction feature, the multimedia resource identification feature, the embedding feature of the object side content side cross feature, the embedding feature of the object side feature and the embedding feature of the content side feature are spliced to serve as network layer input features of the multimedia resource recommendation model. Referring to fig. 9, fig. 9 is a schematic diagram of another application scenario of the feature generation method of multimedia resource recommendation according to the embodiment of the present application. Optionally, the network layers of the multimedia resource recommendation model may include CONCAT layers, EXPERT layers, EXPERT output weighted sum layers, LCR DNN, and PCR DNN. Wherein, the CONCAT layers represent that all the features of the upper layer are connected together to form a 1000+ dimension vector. Wherein, the EXPERT layers represent expert network layers, and each expert network is a multi-layer perceptron. It can be appreciated that the number of the above-mentioned expert network layers can be specifically determined according to the actual scene application, and is not limited herein. For convenience of description, it is assumed that the expert network has 3 expert networks. Wherein, the EXPERT output weighted summation layer performs weighted summation on the outputs of the expert networks of the upper layer. Wherein, the LCR DNN is a multi-layer perceptron of a long click task layer, and the PCR DNN is a multi-layer perceptron of a completion task layer. Specifically, the multimedia resource recommendation model may splice the embedded feature having the object-side content-side cross feature, the embedded feature of the object-side feature, and the embedded feature of the content-side feature, as the embedded feature of the network layer of the multimedia resource recommendation model. The network layer of the multimedia resource recommendation model may splice the embedded feature having the object side content side cross feature, the embedded feature of the object side feature, and the embedded feature of the content side feature through the CONCAT layer, and call EXPERT layer to output the feature value of each expert network for the long click and the feature value of each expert network for the completion degree based on the spliced feature of the previous layer, respectively. The multimedia resource recommendation model calls EXPERT an output weighted summation layer to perform weighted summation on the characteristic value of each expert network for long clicking and the characteristic value of each expert network for completion based on the characteristic value of each expert network for long clicking and the characteristic value of each expert network for completion, so as to generate a characteristic summation value of each expert network for long clicking and a characteristic summation value of each expert network for completion, and inputs the characteristic summation value of each expert network for long clicking and the characteristic summation value of each expert network for completion into LCR DNN and PCR DNN respectively, so as to generate network layer output characteristics of the multimedia resource recommendation model. After the target client generates the object statistical features (i.e., 335-dimensional object behavior features) associated with the operation objects, the object statistical features associated with the operation objects pass through a statistical feature convolution layer and are spliced with the network layer output features of the multimedia resource recommendation model to serve as recommendation value estimation features of the multimedia resources.

Optionally, referring to fig. 9 again, after the target client generates the object statistics feature associated with each operation object, the object statistics feature associated with each operation object is interacted with the multimedia resource identification feature to generate an interaction feature, and the interaction feature, the embedding feature of the multimedia resource identification feature and the object side content side cross feature, the embedding feature of the object side feature, and the embedding feature of the content side feature are spliced to be used as network layer embedding features of the multimedia resource recommendation model. And the multimedia resource recommendation model splices the embedded features through the CONCAT layer, and calls EXPERT layer to respectively output the feature values of each expert network aiming at the long click and the feature values of each expert network aiming at the completion degree based on the splicing features of the previous layer. And calling EXPERT an output weighted summation layer to perform weighted summation on the characteristic values of the long-click expert networks and the characteristic values of the finish expert networks based on the characteristic values of the long-click expert networks and the characteristic values of the finish expert networks, respectively generating characteristic summation values of the long-click expert networks and the finish expert networks, respectively inputting the characteristic summation values of the long-click expert networks and the finish expert networks into LCR DNN and PCR DNN to generate network layer output characteristics of the multimedia resource recommendation model. After the target client generates the object statistical features (i.e., 335-dimensional object behavior features) associated with the operation objects, the object statistical features associated with the operation objects pass through a statistical feature convolution layer and are spliced with the network layer output features of the multimedia resource recommendation model to serve as recommendation value estimation features of the multimedia resources. According to practice, the feature generation method of the multimedia resource recommendation provided by the embodiment of the application can bring a plurality of beneficial effects in a multimedia resource recommendation model after the related policies are online, taking news is taken as an example, the statistics of the related strategies are online, the accumulated total-end average people deep consumption VV is obviously improved by 19.33%, the total-end average people residence time is obviously improved by 1.79%, and the next-day retention rate is obviously improved by 0.16%. By adopting the method and the system, the expression capacity of the model can be greatly improved, so that the target client can recommend multimedia resources which are more interested by the object, and the service index is greatly improved. Meanwhile, the embodiment of the application systematically provides a clear, comprehensive and easily-expanded statistical characteristic system which can be used in a multimedia resource recommendation model, thereby greatly improving the utilization efficiency of computing resources and storage resources and reducing the utilization cost of resources.

The feature generation method of the multimedia resource recommendation provided by the embodiment of the application has the advantages of systematicness, comprehensiveness and expansibility, improves the model expression capability, and provides more personalized service for the operation object. Meanwhile, the feature generation method for multimedia resource recommendation provided by the embodiment of the application has an efficient calculation and storage scheme, and the use efficiency of calculation resources, model training resources and storage resources is greatly improved.

Based on the description of the embodiment of the feature generation method of the multimedia resource recommendation, the embodiment of the application also discloses a feature generation device of the multimedia resource recommendation. The feature generation device of the multimedia resource recommendation can be applied to the feature generation method of the embodiment shown in fig. 1 to 9 for executing the steps in the feature generation method of the multimedia resource recommendation. The feature generating device of the multimedia resource recommendation may be a service server or a terminal device in the embodiments shown in fig. 1 to 9, that is, the feature generating device of the multimedia resource recommendation may be an execution subject of the feature generating method of the multimedia resource recommendation in the embodiments shown in fig. 1 to 9. Referring to fig. 10, fig. 10 is a schematic structural diagram of a feature generating device for multimedia resource recommendation according to an embodiment of the present application. In the embodiment of the application, the device can operate the following modules:

an obtaining module 31, configured to obtain first object behavior data provided by a client data source;

the acquiring module 31 is further configured to acquire second object behavior data provided by a training sample data source recommended by the multimedia resource;

The feature generation module 32 is configured to generate at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data acquired by the acquisition module 31, where the first object is each object carried in the first object behavior data, and the first object behavior statistical feature of any time period associated with any first object includes at least two attribute features of an object attribute feature, a duration attribute feature, a resource content attribute feature, a resource type attribute feature, and an object resource attribute feature;

The feature generation module 32 is further configured to generate at least two second object behavior statistical features of at least two different time periods associated with a second object based on the second object behavior data acquired by the acquisition module, where the second object is each object carried in the first object behavior data, and at least two object resource behavior attribute features in the second object behavior statistical features of any time period associated with any second object;

And a feature aggregation module 33, configured to generate an object statistical feature of each object based on the first object behavior statistical feature associated with each first object and the second object behavior statistical feature associated with each second object, and determine the object statistical feature as an input feature of a multimedia resource recommendation model, where the multimedia resource recommendation model is obtained by training the second object behavior data, and the multimedia resource model is configured to output a multimedia resource recommendation value of each object based on the object statistical feature of each object.

According to the embodiment corresponding to fig. 2, the implementation manner described in steps S101 to S103 in the feature generation method of multimedia resource recommendation shown in fig. 2 may be executed by each module of the apparatus shown in fig. 10. For example, the implementation described in step S101 in the method for generating the multimedia resource recommendation shown in fig. 2 may be performed by the acquiring module 31 and the feature generating module 32 in the apparatus shown in fig. 10, the implementation described in step S102 may be performed by the acquiring module 31 and the feature generating module 32, and the implementation described in step S103 may be performed by the feature aggregation module 33, where the implementation performed in the acquiring module 31, the feature generating module 32, and the feature aggregation module 33 may be referred to the implementation provided in each step in the embodiment corresponding to fig. 2, and will not be repeated herein.

In the embodiment of the application, the feature generating device of the multimedia resource recommendation can receive the first object behavior data provided by the target client data source and the second object behavior data provided by the training sample data source of the multimedia resource recommendation. When the acquisition module receives the first object behavior data, triggering the first object behavior statistical feature generation module to acquire the first object behavior data provided by the client data source, and generating at least two first object behavior statistical features of at least two different time periods associated with each object carried in the first object behavior data based on the first object behavior data. When the acquisition module receives the second object behavior data, triggering the second object behavior statistical feature generation module to acquire the second object behavior data provided by a training sample data source recommended by the multimedia resource, and generating at least two second object behavior statistical features of at least two different data segments associated with each object carried in the first object behavior data based on the second object behavior data. When the target client detects that the first object behavior statistical feature and the second object behavior statistical feature are generated, the triggering aggregation module generates the object statistical feature of each object based on the first object behavior statistical feature and the second object behavior statistical feature associated with each object, and determines the input feature of the multimedia resource recommendation model according to the object statistical feature. The feature generation method of the multimedia resource recommendation provided by the embodiment of the application provides a comprehensive, clear and easily-expanded statistical feature system, and simultaneously greatly improves the utilization efficiency of computing resources, model training resources and storage resources, improves the recommendation accuracy and enhances the applicability.

In the embodiment of the present application, each module in the apparatus shown in the foregoing figures may be combined into one or several other modules separately or all, or some (some) of the modules may be further split into a plurality of modules with smaller functions to form a module, which may achieve the same operation without affecting the implementation of the technical effects of the embodiment of the present application. The above modules are divided based on logic functions, and in practical application, the functions of one module may be implemented by a plurality of modules, or the functions of a plurality of modules may be implemented by one module. In other possible implementations of the present application, the apparatus may also include other modules, where in practical applications, the functions may be implemented with assistance from other modules, and may be implemented by cooperation of multiple modules, which is not limited herein.

Referring to fig. 11, fig. 11 is a schematic structural diagram of a computer device according to an embodiment of the application. As shown in fig. 11, the computer device 1000 may be a terminal device in the embodiment corresponding to fig. 2 to fig. 9. The computer device 1000 may include: processor 1001, network interface 1004, and memory 1005, in addition, the computer device 1000 may further comprise: a user interface 1003, and at least one communication bus 1002. Wherein the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display (Display), a Keyboard (Keyboard), and the optional user interface 1003 may further include a standard wired interface, a wireless interface, among others. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 1005 may also optionally be at least one storage device located remotely from the processor 1001. As shown in fig. 11, an operating system, a network communication module, a user interface module, and a device control application may be included in the memory 1005, which is one type of computer-readable storage medium.

The network interface 1004 in the computer device 1000 may also be connected to the terminal 200b in the embodiment corresponding to fig. 1, and the optional user interface 1003 may also include a Display screen (Display) and a Keyboard (Keyboard). In the computer device 1000 shown in FIG. 11, the network interface 1004 may provide network communication functions; while user interface 1003 is primarily used as an interface to provide input for developers; and the processor 1001 may be configured to invoke the device control application program stored in the memory 1005 to implement the feature generation method of multimedia resource recommendation in the embodiment corresponding to fig. 2.

It should be understood that the computer device 1000 described in the embodiments of the present application may perform the description of the feature generation method of the multimedia resource recommendation in the embodiment corresponding to fig. 2, which is not described herein. In addition, the description of the beneficial effects of the same method is omitted.

In addition, it should be noted that the embodiment of the present application further provides a computer readable storage medium, where a computer program executed by the foregoing apparatus for generating a content recommendation according to the feature of a multimedia resource recommendation is stored, where the computer program includes program instructions, and when the processor executes the program instructions, the description of the method for generating the feature of the multimedia resource recommendation in the foregoing embodiment corresponding to fig. 2 can be executed, and therefore will not be repeated herein. In addition, the description of the beneficial effects of the same method is omitted. For technical details not disclosed in the embodiments of the computer-readable storage medium according to the present application, please refer to the description of the method embodiments of the present application.

In addition, it should be noted that: embodiments of the present application also provide a computer program product, which may include a computer program, which may be stored in a computer readable storage medium. The processor of the computer device reads the computer program from the computer readable storage medium, and the processor may execute the computer program, so that the computer device performs the description of the method for generating the features of the multimedia resource recommendation in the embodiment corresponding to fig. 2 to 9, which will not be described herein. In addition, the description of the beneficial effects of the same method is omitted. For technical details not disclosed in the embodiments of the computer program product according to the present application, reference is made to the description of the method embodiments of the present application.

Those skilled in the art will appreciate that implementing all or part of the above-described methods may be accomplished by way of computer programs, which may be stored on a computer-readable storage medium, and which, when executed, may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random-access Memory (Random Access Memory, RAM), or the like.

The foregoing disclosure is illustrative of the present application and is not to be construed as limiting the scope of the application, which is defined by the appended claims.

Claims (12)

1. A method for generating characteristics of multimedia resource recommendation, the method comprising:

Acquiring first object behavior data provided by a client data source, and generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data, wherein the first object is each object carried in the first object behavior data, and the first object behavior statistical features of any time period associated with any first object comprise at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource attribute features;

acquiring second object behavior data provided by a training sample data source recommended by a multimedia resource, and generating at least two second object behavior statistical features of at least two different time periods associated with a second object based on the second object behavior data, wherein the second object is each object carried in the first object behavior data, and at least two object resource behavior attribute features in the second object behavior statistical features of any time period associated with any second object;

Generating object statistical features of each object based on the first object behavior statistical features related to each first object and the second object behavior statistical features related to each second object, and determining the object statistical features as input features of a multimedia resource recommendation model, wherein the multimedia resource recommendation model is trained by the second object behavior data, and the multimedia resource recommendation model is used for outputting multimedia resource recommendation values of each object based on the object statistical features of each object.

2. The method of claim 1, wherein the generating at least two first object behavior statistics for at least two different time periods associated with a first object based on the first object behavior data comprises:

Generating at least two levels of object behavior characteristics of at least two different time periods based on the generation time of each object behavior data carried in the first object behavior data, wherein the one level of object behavior characteristics is used for generating object behavior statistical characteristics of each first object in one time period, and the generation time of the object behavior data contained in the different time periods is not overlapped;

Generating at least two first object behavior statistical features with the object identifiers of the first objects as indexes based on the object identifiers of the first objects carried in the at least two-stage object behavior features, wherein any one first object is associated with at least two first object behavior statistical features of the at least two different time periods with the object identifier of the any one first object as an index.

3. The method of claim 2, wherein the generating at least two levels of object behavior features for at least two different time periods based on the generation time of each object behavior data carried in the first object behavior data comprises:

Dividing the object behavior data of each first object belonging to the same time period into object behavior data of the same level based on the generation time of each object behavior data carried in the first object behavior data to obtain at least two levels of object behavior data;

generating at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource attribute features of each first object based on the object behavior data of each first object included in each level of object behavior data, and splicing the at least two attribute features associated with each first object to generate each level of object behavior features corresponding to each time period so as to obtain at least two levels of object behavior features of at least two different time periods.

4. A method according to claim 3, wherein the object resource behavioral attribute characteristics include one or more of a multimedia resource preference level of an object or an object preference level of a multimedia resource; the generating the object resource behavior attribute feature of each first object based on the object behavior data of each first object included in the object behavior data of each level includes:

Acquiring any object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on any multimedia resource, determining a ratio of the any object behavior statistical value to the sum of the object behavior statistical values of all objects in all levels of object behavior data on any multimedia resource as the preference degree of any first object on any multimedia resource so as to obtain the preference degree of any first object on the multimedia resource; or alternatively

Acquiring any object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on any multimedia resource, determining a ratio of the any object behavior statistical value to the sum of the object behavior statistical values of any first object in all levels of object behavior data on all multimedia resources as the preference degree of any first object on any multimedia resource so as to obtain the object preference degree of any multimedia resource.

5. A method according to claim 3, wherein the object resource behavioral attribute characteristics include an expected amount of clicks of the object on the multimedia resource; the generating the object resource behavior attribute feature of each first object based on the object behavior data of each first object included in the object behavior data of each level includes:

acquiring a first object behavior statistical value of any first object in object behavior data of any first object included in all levels of object behavior data on a first multimedia resource, and acquiring a first object behavior operation proportion of all objects in all levels of object behavior data on the first multimedia resource;

Acquiring a second object behavior statistical value of any first object in the object behavior data of each level of object, and acquiring a second object behavior operation proportion of all objects in the object behavior data of each level of object to the second multimedia resource;

And generating the expected click quantity of any first object on the multimedia resources based on the first object behavior statistical value, the first object behavior operation proportion, the second object behavior statistical value and the second object behavior operation proportion and the object behavior statistical value of any object on all the multimedia resources so as to obtain the expected click quantity of each first object on the multimedia resources.

6. The method of any of claims 1-5, wherein the generating at least two second object behavior statistics for at least two different time periods associated with respective objects carried in the first object behavior data based on the second object behavior data comprises:

Dividing the second object behavior data into a plurality of P ₁ -level object behavior data based on the generation time of each object behavior data carried in the second object behavior data, wherein the P ₁ -level object behavior data is the object behavior data included in the minimum unit time length of the second object behavior data divided according to time intervals;

Generating at least two P ₂ -level object behavior data of at least two different time periods based on accumulation of a plurality of P ₁ -level object behavior data included in a plurality of minimum unit time periods, wherein one P ₂ -level object behavior data is accumulated by a plurality of P ₁ -level object behavior data in one target time period, and the target time period is a positive integer multiple of the minimum unit time period;

And generating at least two second object behavior statistical features with the object identifiers of the second objects as indexes based on the object identifiers corresponding to the second objects carried in the P ₂ -level object behavior data, wherein any one second object is associated with at least two second object behavior statistical features in at least two different time periods with the object identifier of any one second object as an index.

7. The method of claim 6, wherein generating at least two second object behavior statistics indexed by the object identifier of each second object based on the object identifier corresponding to each second object carried in each P ₂ -level object behavior data comprises:

Generating P ₂ -level object behavior characteristics associated with each second object based on object identifiers corresponding to each second object carried in the P ₂ -level object behavior data, wherein the P ₂ -level object behavior characteristics of one second object comprise at least two object resource behavior attribute characteristics of the second object;

Splicing at least two object resource behavior attribute features in P ₂ -level object behavior features associated with the object identifiers of the second objects by taking the object identifiers of the second objects as indexes so as to generate second object behavior statistical features associated with the object identifiers;

The object resource behavior attribute features comprise at least one object resource behavior associated feature of at least one of click time length, click times, click rate, exposure times, watching time length and interaction time length of an object aiming at the multimedia resource.

8. The method of claim 7, wherein the generating object statistics for each object based on the first object behavior statistics for each of the first object associations and the second object behavior statistics for each of the second object associations comprises:

Obtaining object identifiers corresponding to the first objects from the first object behavior statistical features, and obtaining object identifiers corresponding to the second objects from the second object behavior statistical features;

And splicing the first object behavior statistical features and the second object behavior statistical features by taking the object identifications of the first objects and the object identifications of the second objects as indexes to generate object statistical features corresponding to the object identifications, wherein the first object behavior statistical features and the second object behavior statistical features with the same object identifications are spliced into object statistical features of the same object.

9. A feature generation apparatus for multimedia resource recommendation, comprising:

the acquisition module is used for acquiring first object behavior data provided by the client data source;

The feature generation module is used for generating at least two first object behavior statistical features of at least two different time periods associated with a first object based on the first object behavior data acquired by the acquisition module, wherein the first object is each object carried in the first object behavior data, and the first object behavior statistical features of any time period associated with any first object comprise at least two attribute features of object attribute features, duration attribute features, resource content attribute features, resource type attribute features and object resource behavior attribute features;

the acquisition module is further used for acquiring second object behavior data provided by a training sample data source recommended by the multimedia resource;

The feature generation module is further configured to generate at least two second object behavior statistical features of at least two different time periods associated with a second object based on the second object behavior data acquired by the acquisition module, where the second object is each object carried in the first object behavior data, and at least two object resource behavior attribute features in the second object behavior statistical features of any time period associated with any second object;

And the feature aggregation module is used for generating object statistical features of all objects based on the first object behavior statistical features associated with all the first objects and the second object behavior statistical features associated with all the second objects, and determining the object statistical features as input features of a multimedia resource recommendation model, wherein the multimedia resource recommendation model is obtained by training the second object behavior data, and the multimedia resource model is used for outputting multimedia resource recommendation values of all the objects based on the object statistical features of all the objects.

10. A computer device, comprising: a processor, a memory, and a network interface;

The processor is connected to the memory, the network interface for providing data communication functions, the memory for storing program code, the processor for invoking the program code to perform the method of any of claims 1-8.

11. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program adapted to be loaded by a processor and to perform the method of any of claims 1-8.

12. A computer program product, characterized in that the computer program product comprises a computer program stored in a computer readable storage medium, the computer program being adapted to be read and executed by a processor to cause a computer device having the processor to perform the method of any of claims 1-8.