A multimedia resource recommendation method, device, equipment and storage medium

Abstract

The application discloses a multimedia resource recommendation method, device, equipment and storage medium, comprising: obtaining first behavior data of a target object in a first time period and second behavior data of the target object in a second time period; obtaining N first behavior sequences and M second behavior sequences; determining a target feature vector of the target object to L preset multimedia resources; obtaining a predicted interest degree of the target object to each preset multimedia resource through an interest degree prediction model based on the target feature vector of the target object to the L preset multimedia resources; sequentially arranging the predicted interest degree of the target object to each preset multimedia resource, and determining a target multimedia resource based on the sequential arrangement result. Through the above method, the target feature vector can accurately represent the characteristics of the object in each dimension in different time periods, so that the target multimedia resource conforms to the actual behavior data of the object, thereby improving the accuracy of multimedia resource recommendation.

Description

Technical Field

[0001] This application relates to the field of machine learning technology in the field of artificial intelligence, and in particular to a method, apparatus, device and storage medium for recommending multimedia resources. Background Technology

[0002] With the rapid development of internet technology, multimedia resource recommendation based on the internet has also developed accordingly, and methods for multimedia resource recommendation are widely used in various scenarios such as news, videos, music, audiobooks, and e-commerce. Multimedia resource recommendation refers to recommending products that an audience might like, in order to increase audience dwell time and click-through rates. Currently, multimedia resource recommendation methods need to respond quickly and accurately to the audience's current needs. This typically requires pre-analysis of audience behavior to obtain their interests, then mapping these interests to user profiles, and finally making multimedia resource recommendations based on these user profiles. Therefore, the recommended objects are determined based on user profiles. However, since user profiles obtained through interest mapping may be inaccurate, this reduces the accuracy of multimedia resource recommendations. Therefore, improving the accuracy of multimedia resource recommendations is an urgent problem to be solved. Summary of the Invention

[0003] This application provides a method, apparatus, device, and storage medium for recommending multimedia resources. It obtains a first behavioral sequence and a second behavioral sequence based on behavioral data within different time periods. Since the first behavioral sequence includes features of the same dimension within a first time period, and the second behavioral sequence includes features of different dimensions within a second time period, feature modeling is performed on the first and second behavioral sequences of the target object to obtain a target feature vector for the target object's interest in multiple preset multimedia resources. This target feature vector accurately represents the characteristics of the target object in various dimensions within different time periods. Therefore, the interest level of the target object in the multiple preset multimedia resources output by the interest prediction model is more consistent with the actual behavioral data of the target object, meaning the accuracy of the interest level in the multiple preset multimedia resources is high. Consequently, the target multimedia resources recommended to the target object are also more consistent with the actual behavioral data of the target object, thereby improving the accuracy of multimedia resource recommendation.

[0004] In view of this, the first aspect of this application provides a method for recommending multimedia resources, comprising:

[0005] Acquire the first row data of the target object within a first time period and the second row data within a second time period, wherein the length of the first time period is greater than the length of the second time period, and the first time period includes the second time period.

[0006] Based on the first row of data and the second row of data, N first row sequences and M second row sequences are obtained. The first row sequences include features of the same dimension within the first time period, and the second row sequences include features of different dimensions within the second time period. M and N are both positive integers greater than 1.

[0007] Based on N first-row sequences and M second-row sequences, determine the target feature vectors of the target object for L preset multimedia resources, where L is a positive integer greater than 1;

[0008] Based on the target feature vectors of L preset multimedia resources for the target object, the predicted interest degree of the target object for each preset multimedia resource is obtained through the interest degree prediction model.

[0009] The predicted interest of the target object in each preset multimedia resource is ordered, and the target multimedia resource recommended to the target object is determined from L preset multimedia resources based on the order order. The target multimedia resource includes at least one preset multimedia resource.

[0010] A second aspect of this application provides a multimedia resource recommendation device, comprising:

[0011] The acquisition module is used to acquire the first row of data of the target object within a first time period and the second row of data within a second time period, wherein the length of the first time period is greater than the length of the second time period, and the first time period includes the second time period.

[0012] The acquisition module is also used to obtain N first-row sequences and M second-row sequences based on the first-row data and the second-row data. The first-row sequences include features of the same dimension within the first time period, and the second-row sequences include features of different dimensions within the second time period. M and N are both positive integers greater than 1.

[0013] The determination module is used to determine the target feature vector of the target object for L preset multimedia resources based on N first row sequences and M second row sequences, where L is a positive integer greater than 1;

[0014] The acquisition module is also used to obtain the predicted interest of the target object on each preset multimedia resource based on the target feature vector of the target object on L preset multimedia resources through the interest prediction model.

[0015] The determination module is also used to rank the predicted interest of the target object in each preset multimedia resource in order, and to determine the target multimedia resource to be recommended to the target object from the L preset multimedia resources based on the ranking result, wherein the target multimedia resource includes at least one preset multimedia resource.

[0016] In one possible implementation, the acquisition module is specifically used to obtain N first row sequences based on the first row data;

[0017] Based on the data in the second row, M sequences of the second row are obtained.

[0018] In one possible implementation, the acquisition module is specifically used to perform information extraction processing on the first behavior data to obtain the first access behavior corresponding to N dimensions within the first time period, wherein the first access behavior is generated when the target object accesses the preset multimedia resources.

[0019] Based on the first access behavior corresponding to N dimensions within the first time period, N first behavior sequences are generated.

[0020] In one possible implementation, the acquisition module is specifically used to perform information extraction processing on the second behavior data to obtain M second access behaviors within a second time period, wherein the second access behaviors are generated when the target object accesses a preset multimedia resource.

[0021] Based on M second access behaviors within the second time period, generate M second behavior sequences.

[0022] In one possible implementation, the determining module is specifically used to determine the first feature vector of the target object for L preset multimedia resources based on N first action sequences;

[0023] The second feature vector of the target object for L preset multimedia resources is determined based on M second action sequences;

[0024] The first feature vector and the second feature vector are fused together to obtain the target feature vector.

[0025] In one possible implementation, the multimedia resource recommendation device further includes a processing module;

[0026] The acquisition module is also used to acquire basic attribute information of the target object;

[0027] The processing module is used to perform feature processing on the basic attribute information of the target object to obtain the user profile features of the target object.

[0028] The determination module is specifically used to determine the first feature vector based on N first behavior sequences and the user profile features of the target object;

[0029] The determination module is specifically used to determine the first feature vector based on N first behavior sequences and the user profile features of the target object;

[0030] The determination module is specifically used to fuse the first feature vector, the second feature vector, and the user profile features of the target object to obtain the target feature vector.

[0031] In one possible implementation, the acquisition module is further configured to acquire a multimedia resource feature set, wherein the multimedia resource feature set includes F multimedia resource feature subsets, each multimedia resource feature subset includes L preset multimedia resources with the same dimension of features, and different multimedia resource feature subsets include features with different dimensions, F is a positive integer greater than 1, and F is greater than or equal to N.

[0032] The determination module is specifically used to determine the first feature vector based on N first behavior sequences, user profile features of the target object, and a set of multimedia resource features.

[0033] The determination module is specifically used to determine the second feature vector based on M second behavior sequences, user profile features of the target object, and multimedia resource feature set;

[0034] The acquisition module is specifically used to obtain the predicted interest of the target object in each preset multimedia resource based on the target feature vector and the multimedia resource feature set, through an interest prediction model.

[0035] In one possible implementation, the acquisition module is specifically used to acquire a preset multimedia resource set, wherein the preset multimedia resource set includes L preset multimedia resources, and the L preset multimedia resources include F dimensions of features.

[0036] Based on F dimensions, feature extraction is performed on L preset multimedia resources to obtain a feature subset of multimedia resources corresponding to each dimension;

[0037] A multimedia resource feature set is generated based on the multimedia resource feature subset corresponding to each dimension.

[0038] In one possible implementation, the multimedia resource recommendation device further includes a training module;

[0039] The acquisition module is also used to acquire a first set of data samples, a second set of data samples, a set of preset multimedia resources, and a set of real tags for the target object. The first set of data samples includes multiple first set of data samples, the second set of data samples includes multiple second set of data samples, the first set of data samples and the second set of data samples are data samples from different time periods, the set of preset multimedia resources includes L preset multimedia resources, and the set of real tags includes the real tags corresponding to each preset multimedia resource for the target object.

[0040] The determining module is further configured to determine a first row sample sequence set and a second row sample sequence set based on the first row data sample set and the second row data sample set, wherein the first row sample sequence set includes multiple first row sample sequences, the second row sample sequence set includes multiple second row sample sequences, and the first row sample sequences include features of the same dimension, while the second row sample sequences include features of different dimensions.

[0041] The determination module is also used to determine the target feature vector of the target object for L preset multimedia resources based on the first row of sample sequence set and the second row of sample sequence set;

[0042] The acquisition module is also used to obtain the predicted interest of the target object on each preset multimedia resource based on the target feature vector of the target object on L preset multimedia resources and the preset multimedia resource set, through the interest prediction model to be trained.

[0043] The training module is used to train the interest prediction model to be trained based on the real label set and the predicted interest of the target object for each preset multimedia resource, so as to obtain the interest prediction model.

[0044] In one possible implementation, the training module is specifically used to determine the predicted label corresponding to each preset multimedia resource for the target object based on the predicted interest of the target object in each preset multimedia resource.

[0045] Based on the real tag set and the predicted tag corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model to be trained are updated according to the target loss function to obtain the interest prediction model.

[0046] In one possible implementation, the determining module is specifically used to sequentially arrange the predicted interest of the target object for each preset multimedia resource using the max-heap sort method.

[0047] The order of the results is determined by sorting the predicted interest in each preset multimedia resource from highest to lowest.

[0048] Extract target interest levels from high to low based on the sequentially arranged results;

[0049] The preset multimedia resources corresponding to the target interest level are identified as the target multimedia resources.

[0050] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0051] A fourth aspect of this application provides a computer device, characterized in that it includes: a memory, a transceiver, a processor, and a bus system; wherein the memory is used to store a program; the processor is used to execute the program in the memory to implement the methods described in the above aspects; and the bus system is used to connect the memory and the processor to enable the memory and the processor to communicate.

[0052] A fifth aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods described in the above aspects.

[0053] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0054] This application provides a method for recommending multimedia resources. The method involves acquiring first-behavioral data of a target object within a first time period and second-behavioral data within a second time period. The length of the first time period is greater than the length of the second time period, and the first time period includes the second time period. Based on the first-behavioral data and the second-behavioral data, N first-behavioral sequences and M second-behavioral sequences are obtained. The first-behavioral sequences include features of the same dimension within the first time period, and the second-behavioral sequences include features of different dimensions within the second time period. M and N are both positive integers greater than 1. Based on this, according to the N first-behavioral sequences and M second-behavioral sequences, target feature vectors for the target object on L preset multimedia resources are determined, where L is a positive integer greater than 1. These target feature vectors are used as an interest prediction model. The model outputs the predicted interest level of the target object on each preset multimedia resource. The predicted interest levels of the target object on each preset multimedia resource are then ordered. Based on the ordering result, target multimedia resources recommended to the target object are determined from the L preset multimedia resources. These target multimedia resources include at least one preset multimedia resource. Using the above method, a first behavior sequence and a second behavior sequence are obtained based on behavioral data within different time periods. Since the first behavior sequence includes features of the same dimension within the first time period, and the second behavior sequence includes features of different dimensions within the second time period, feature modeling is performed on the first and second behavior sequences of the target object to obtain the target feature vector of the target object for multiple preset multimedia resources. This target feature vector can accurately represent the features of the target object in various dimensions within different time periods. Therefore, the interest of the target object for multiple preset multimedia resources output by the interest prediction model is more consistent with the actual behavioral data of the target object, that is, the accuracy of the interest in multiple preset multimedia resources is high. Thus, the target multimedia resources recommended to the target object are also more consistent with the actual behavioral data of the target object, thereby improving the accuracy of multimedia resource recommendation. Attached Figure Description

[0055] Figure 1 This is a system schematic diagram of a multimedia resource recommendation method in an embodiment of this application;

[0056] Figure 2 A flowchart illustrating the method for recommending multimedia resources provided in this application embodiment;

[0057] Figure 3 A schematic diagram of an embodiment of the multimedia resource recommendation method provided in this application;

[0058] Figure 4 A schematic diagram of an interface for displaying target multimedia resources provided in an embodiment of this application;

[0059] Figure 5 A schematic diagram of a process for training an interest prediction model to be trained, as provided in an embodiment of this application;

[0060] Figure 6 A schematic diagram of a multimedia resource recommendation device provided in an embodiment of this application;

[0061] Figure 7 This is a block diagram of a portion of the structure of a mobile phone associated with a terminal in an embodiment of this application. Detailed Implementation

[0062] This application provides a method, apparatus, device, and storage medium for recommending multimedia resources. It obtains a first behavioral sequence and a second behavioral sequence based on behavioral data within different time periods. Since the first behavioral sequence includes features of the same dimension within a first time period, and the second behavioral sequence includes features of different dimensions within a second time period, feature modeling is performed on the first and second behavioral sequences of the target object to obtain a target feature vector for the target object's interest in multiple preset multimedia resources. This target feature vector accurately represents the characteristics of the target object in various dimensions within different time periods. Therefore, the interest level of the target object in the multiple preset multimedia resources output by the interest prediction model is more consistent with the actual behavioral data of the target object, meaning the accuracy of the interest level in the multiple preset multimedia resources is high. Consequently, the target multimedia resources recommended to the target object are also more consistent with the actual behavioral data of the target object, thereby improving the accuracy of multimedia resource recommendation.

[0063] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0064] With the rapid development of internet technology, multimedia resource recommendation based on the internet has also developed accordingly, and methods for multimedia resource recommendation are widely used in various scenarios such as news, videos, music, audiobooks, and e-commerce. Multimedia resource recommendation refers to recommending products that an audience might like, in order to increase audience dwell time and click-through rates. Currently, multimedia resource recommendation methods need to respond quickly and accurately to the audience's current needs. This typically requires pre-analyzing audience behavior to obtain their interests, then mapping these interests to user profiles, and finally making multimedia resource recommendations based on these user profiles. Therefore, the recommended audience is determined based on the user profile. However, since user profiles obtained through interest mapping may be inaccurate, this reduces the accuracy of multimedia resource recommendations. Alternatively, some recommendation systems employ a two-stage approach: first, recalling a potential candidate set in the recall stage, and then performing precise ranking and recommendation in the ranking stage. Therefore, it is crucial to accurately characterize the object's interests during the recall stage. The industry standard for recall models is collaborative filtering based on multimedia resources. However, collaborative filtering models can only consider the object's static interests and cannot capture its dynamic interests. Consequently, the characterized object interests cannot accurately reflect the object's dynamic behavior, thus reducing the accuracy of multimedia resource recommendations. Therefore, improving the accuracy of multimedia resource recommendations is a pressing issue. Based on this, to address the aforementioned problems, this application provides a method for multimedia resource recommendation. The obtained target feature vector can accurately characterize the object's features across various dimensions within different time periods. Therefore, the target multimedia resources match the object's actual behavioral data, thereby improving the accuracy of multimedia resource recommendations.

[0065] First, for ease of understanding, some terms or concepts involved in the embodiments of this application will be explained.

[0066] I. Session

[0067] A session refers to a single network access session, which is when an object performs an operation on a multimedia resource through a multimedia resource browsing interface displayed on a terminal device. At this time, the interaction between the terminal device and the server generates a session, which is the access behavior described in this embodiment.

[0068] II. Low-dimensional feature vectors (Embedding)

[0069] Embedding uses a low-dimensional vector to represent an object, which could be a word, a product, a movie, etc. In this embodiment, the identifier (id) class coefficient features are specifically transformed into the feature vector of the embedding.

[0070] III. One-Hot Encoding

[0071] One-Hot encoding is an encoding method used to extend the values ​​of discrete features to Euclidean space. A certain value of the discrete feature corresponds to a certain point in Euclidean space, thus obtaining a sparse vector representation. The resulting vector values ​​are either 0 or 1, with only one 1 and the others being 0.

[0072] IV. Multi-Hot Coding

[0073] Multi-Hot encoding is also a type of encoding method used to encode multiple attributes into a single feature, resulting in a sparse vector representation. The resulting vector values ​​are either 0 or 1, with a vector having multiple 1s and the rest being 0s.

[0074] V. Multimedia Resources

[0075] The multimedia resources described in this embodiment may include, but are not limited to, advertising recommendation pages for virtual goods, product detail pages for virtual goods, and virtual item recommendation pages. It should be understood that multimedia resources will differ in different scenarios in actual applications. For example, if applied to a virtual goods shopping application, the multimedia resource may be an advertising recommendation page for virtual goods or a product detail page for virtual goods. If applied to a terminal game application, the multimedia resource may be a virtual item recommendation page. If applied to a video playback application, the multimedia resource may also be a video recommendation page. And if applied to a music playback application, the multimedia resource may also be a music recommendation page, etc. Therefore, the foregoing examples should not be construed as a specific limitation on multimedia resources.

[0076] VI. Behavioral Sequence

[0077] The behavior sequence is specifically a vector of the object's historical behavior sequence. The behavior sequence can include the multimedia resource ID, multimedia resource category, multimedia resource brand, and multimedia resource industry corresponding to the access behavior that the object has performed.

[0078] The application scenarios of the embodiments of this application are described below. It is understood that this method for recommending multimedia resources can be executed by a terminal device or by a server; please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a system schematic diagram of a multimedia resource recommendation method in an embodiment of this application, such as... Figure 1As shown, the multimedia resource recommendation system includes a server and a terminal device. Based on this, when the multimedia resource recommendation method is deployed on the terminal device, the terminal device first obtains an interest prediction model through model training. Then, when an object accesses any preset multimedia resource, it stores the corresponding behavioral data locally. Therefore, it can obtain the first behavioral data of the target object in a first time period and the second behavioral data in a second time period from the local storage. Using the multimedia resource recommendation method described in this application, it determines the target multimedia resource to recommend to the target object and then displays the target multimedia resource on the terminal device's display interface to complete an accurate multimedia resource recommendation. At this time, the terminal device does not need to be connected to the internet, which better protects the privacy of the object's behavioral data, and the multimedia resource recommendation process is more convenient.

[0079] When the multimedia resource recommendation method is deployed on a server, the server can first obtain an interest prediction model through model training, and then obtain the first behavioral data of the target object in the first time period and the second behavioral data in the second time period in real time. Based on the server's hardware performance, richer behavioral information is provided for the behavioral data. Therefore, by determining the target multimedia resources to be recommended to the target object through the multimedia resource recommendation method described in this application embodiment, the processing efficiency of determining the target multimedia resources can be improved. Then, the obtained target multimedia resources are sent to the terminal device used by the target object, thereby improving the efficiency and real-time performance of multimedia resource recommendation.

[0080] It should be noted that the server involved in this application can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, in-vehicle device, etc., but is not limited to these. The terminal and server can be directly or indirectly connected through wired communication, wireless communication, or removable storage, and this application does not impose any restrictions. The aforementioned wireless communication methods use standard communication technologies and / or protocols. The wireless network is typically the Internet, but can also be any network, including but not limited to Bluetooth, Local Area Network (LAN), Metropolitan Area Network (MAN), Wide Area Network (WAN), mobile, private networks, or any combination of virtual private networks. In some embodiments, customized or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies. Removable storage media can be Universal Serial Bus (USB) flash drives, portable hard drives, or other removable storage media.

[0081] Although Figure 1 Only five terminal devices and one server are shown, but it should be understood that... Figure 1 The examples provided are for understanding this solution only; the actual number of terminal devices and servers should be determined flexibly based on the specific circumstances.

[0082] As can be seen from the foregoing embodiments, the model training process in this application requires the implementation of Artificial Intelligence (AI) technology. Before introducing the multimedia resource recommendation method provided in this application, some basic concepts in the field of artificial intelligence will be introduced first. AI is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology of computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making. Artificial intelligence technology is a comprehensive discipline involving a wide range of fields, including both hardware and software technologies. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, large-scale multimedia resource recommendation technology, operating / interactive systems, and mechatronics. Artificial intelligence software technologies mainly include computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0083] With the advancement of artificial intelligence (AI) technology, research has expanded into various directions. Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of AI and the fundamental way to endow computers with intelligence; its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0084] Secondly, the embodiments of this invention can be applied to various scenarios, including but not limited to cloud technology, advertising recommendations on shopping platforms, and item recommendations in game scenarios. Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. Cloud technology is a general term encompassing network technology, information technology, integration technology, management platform technology, and application technology applied to cloud computing business models. It can form resource pools, available on demand, and is flexible and convenient. Cloud computing technology will become a crucial support. The backend services of technical network systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of the internet industry, each multimedia resource may have its own identification mark in the future, requiring transmission to the backend system for logical processing. Data at different levels will be processed separately, and various industry data will require robust system support, which can only be achieved through cloud computing.

[0085] Cloud computing refers to the delivery and usage model of IT infrastructure, meaning obtaining necessary resources in an on-demand and easily scalable manner through a network. In a broader sense, cloud computing refers to the delivery and usage model of services, meaning obtaining necessary services in an on-demand and easily scalable manner through a network. These services can be IT and software related, internet-related, or other services. Cloud computing is a product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.

[0086] With the development of the internet, real-time data streams, and the diversification of connected devices, driven by demands for search services, social networks, mobile commerce, and open collaboration, cloud computing has rapidly developed. Unlike previous parallel distributed computing, cloud computing will fundamentally revolutionize the entire internet model and enterprise management model. The Internet of Things (IoT) refers to the use of various information sensors, RFID technology, GPS, infrared sensors, laser scanners, and other devices and technologies to collect real-time information on any object or process that needs monitoring, connection, or interaction. This information includes sound, light, heat, electricity, mechanics, chemistry, biology, location, and other necessary data. Through various possible network access methods, it achieves ubiquitous connectivity between things and between things and people, enabling intelligent perception, identification, and management of multimedia resources and processes. The IoT is an information carrier based on the internet and traditional telecommunications networks, enabling all independently addressable ordinary physical objects to form an interconnected network.

[0087] Cloud IoT aims to connect the information sensed and commands received by traditional IoT devices to the Internet, truly achieving networking. It also enables massive data storage and computation through cloud computing technology. Due to the nature of IoT, which involves connecting things to each other and sensing the current operating status of each "object" in real time, a large amount of data is generated in this process. How to aggregate this information and how to sift out useful information from the massive amount of data to support decision-making for future development have become key issues affecting the development of IoT. As a result, IoT cloud based on cloud computing and cloud storage technology has become a powerful support for IoT technology and applications.

[0088] For easier understanding, please refer to Figure 2 , Figure 2 Please refer to the flowchart illustrating the method for recommending multimedia resources provided in the embodiments of this application. Figure 2 The multimedia resource recommendation method consists of two stages: the interest prediction model training stage and the multimedia resource recommendation stage. The functions and processes of each stage will be described below:

[0089] In the A1 training phase of the interest prediction model, firstly, behavioral data samples of the target object are acquired within different time periods to obtain the first behavioral data sample set and the second behavioral data sample set of the target object. Secondly, a preset multimedia resource set is generated based on multiple preset multimedia resources. These preset multimedia resources can be preset recommended advertisements obtained from a preset recommended advertisement library or preset recommended products obtained from a preset recommended product library. Next, the true labels corresponding to each preset multimedia resource are determined by manual annotation. For example, a true label of 1 indicates interest in the preset multimedia resource, while a true label of 0 indicates no interest. The aforementioned manual annotation is performed on each preset multimedia resource, thereby obtaining the true label set.

[0090] Based on this, using the first row of data sample sets and the second row of data sample sets, a first row of sample sequence sets and a second row of sample sequence sets are determined. The first row of sample sequences includes features of the same dimension, while the second row of sample sequences includes features of different dimensions. Therefore, based on the first row of sample sequence sets and the second row of sample sequence sets, the target feature vectors of the target object for L preset multimedia resources are determined. These target feature vectors are then used as input to the interest prediction model to be trained. The model outputs the predicted interest level of the target object for each preset multimedia resource. Further, the predicted interest level of the target object for each preset multimedia resource is compared with a preset interest threshold to determine the predicted label corresponding to each preset multimedia resource. For example, if the preset interest threshold is 60%, then if the predicted interest level of the target object for a preset multimedia resource is greater than 60%, the predicted label for that preset multimedia resource is determined to be 1; conversely, if the predicted interest level is less than 60%, the predicted label is determined to be 0. Finally, based on the real label set and the predicted label corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model to be trained are updated according to the target loss function, thereby obtaining the interest prediction model.

[0091] In the multimedia resource recommendation stage A2, when multimedia resource recommendations need to be made for a target object, the first action data of the target object within a first time period and the second action data within a second time period are first obtained. The length of the first time period is greater than the length of the second time period, and the first time period includes the second time period. Based on this, N first action sequences and M second action sequences are obtained according to the first action data and the second action data. The first action sequences include features of the same dimension within the first time period, and the second action sequences include features of different dimensions within the second time period. Then, based on the N first action sequences and the M second action sequences, the target feature vectors of the target object for L preset multimedia resources are determined, where L is a positive integer greater than 1. Further, the target feature vectors of the target object for the L preset multimedia resources are used as input to the interest prediction model obtained in stage A1. The interest prediction model obtained in stage A1 outputs the predicted interest of the target object for each preset multimedia resource. Then, the predicted interest of the target object for each preset multimedia resource is ordered, and based on the order order, the target multimedia resource recommended to the target object is determined from the L preset multimedia resources. This target multimedia resource includes at least one preset multimedia resource. The target feature vector obtained in stage A2 can accurately represent the characteristics of the target object in various dimensions within different time periods. Therefore, the target object's interest in multiple preset multimedia resources output by the interest prediction model is more consistent with the target object's actual behavioral data. In other words, the accuracy of the interest in multiple preset multimedia resources is high. Thus, the target multimedia resources recommended to the target object are also more consistent with the target object's actual behavioral data, thereby improving the accuracy of multimedia resource recommendations.

[0092] Based on the above introduction, the following section uses a terminal device as an example to describe the method for recommending multimedia resources in this application. Please refer to [link / reference]. Figure 3 , Figure 3 A schematic diagram of an embodiment of the multimedia resource recommendation method provided in this application is shown, such as... Figure 3 As shown, the method includes:

[0093] 101. Obtain the first behavior data of the target object within the first time period, and the second behavior data within the second time period.

[0094] In this embodiment, the terminal device acquires first behavioral data of the target object within a first time period and second behavioral data within a second time period. The length of the first time period is greater than the length of the second time period, and the first time period includes the second time period. Specifically, behavioral data refers to the data corresponding to the interactive operations performed by the target object on preset multimedia resources within a certain time period. This behavioral data includes, but is not limited to, accessing, adding to favorites, sharing, or purchasing preset multimedia resources, etc., and is not specifically limited here. Furthermore, the objects described in this embodiment are all actual users, and will not be described in detail below.

[0095] Furthermore, the aforementioned first time period has a longer time span than the second time period, and the time span of the first time period includes the time span of the second time period. For example, if the first time period is from January 1st to 7th, 2021, then the second time period can be from January 1st to 3rd, 2021. Or, if the first time period is the third quarter of 2021 (i.e., July to September), then the second time period can be July, 2021. The aforementioned preset multimedia resources can include, but are not limited to, preset recommended advertisements or preset recommended products, and these preset multimedia resources can be preset recommended advertisements in a preset recommended advertisement library or preset recommended products in a preset recommended product library. For example, preset multimedia resource A is an advertisement for clothing A, and preset multimedia resource B is an advertisement for clothing B, etc. It should be understood that the foregoing examples are only for understanding this solution and should not be construed as specific limitations of this solution.

[0096] 102. Based on the first row of data and the second row of data, obtain N first row sequences and M second row sequences.

[0097] In this embodiment, the terminal device obtains N first-line sequences and M second-line sequences based on the first-line data and the second-line data. The N first-line sequences correspond to the first-line data and include features of the same dimension within a first time period. The M second-line sequences correspond to the second-line data and include features of different dimensions within a second time period. Both M and N are positive integers greater than 1.

[0098] Specifically, the first behavior sequence refers to the features corresponding to multiple access behaviors under the same dimension, that is, the N first behavior sequences include the first behavior sequences under N dimensions. The second behavior sequence refers to the features corresponding to multiple dimensions under the same access behavior, that is, the M second behavior sequences include the second behavior sequences corresponding to M access behaviors respectively. Each of the aforementioned features can be represented as a one-hot vector. As described in the foregoing embodiments, the access behavior described in this embodiment is specifically a session, that is, the target object performs an operation on the multimedia resource through the multimedia resource browsing interface displayed on the terminal device. At this time, the interaction between the terminal device and the server generates an access behavior. Furthermore, the dimensions described in this embodiment are specifically multiple dimensions corresponding to the preset multimedia resource indicated by the access behavior. For example, the multimedia resource identifier corresponding to the preset multimedia resource, the multimedia resource category corresponding to the preset multimedia resource, the specific industry corresponding to the preset multimedia resource, the brand corresponding to the preset multimedia resource, the price range corresponding to the preset multimedia resource, etc. The specific dimensions need to be divided according to the specific needs of the preset multimedia resource maintenance personnel based on the specific scenario, and are not limited here.

[0099] Based on this, the N first-row sequences and the M second-row sequences are specifically described by formulas (1) and (2).

[0100] Formula (1) is as follows:

[0101] (1)

[0102] in, It refers to a sequence of N first rows. This refers to the sequence of the first row in the first dimension. This refers to the first row sequence of the second dimension. This refers to the first row sequence of the third dimension. Similarly, it can be deduced that... This refers to the sequence of the first row in the Nth dimension.

[0103] Formula (2) is as follows:

[0104] (2)

[0105] in, It refers to a sequence of M second lines. This refers to the sequence of second actions corresponding to the first access action. This refers to the sequence of second actions corresponding to the second access action. This refers to the sequence of the second action corresponding to the third access action. Similarly, it can be deduced that... This refers to the second action sequence corresponding to the Mth access action.

[0106] 103. Based on N first-row sequences and M second-row sequences, determine the target feature vector of the target object for L preset multimedia resources.

[0107] In this embodiment, the terminal device determines the target feature vector of the target object for L preset multimedia resources based on N first line sequences and M second line sequences, where L is a positive integer greater than 1. Specifically, the aforementioned L preset multimedia resources are all preset multimedia resources in a preset multimedia resource library. For example, the preset multimedia resource library includes preset multimedia resource 1, preset multimedia resource 2, and preset multimedia resources 3 to 10, that is, the preset multimedia resource library includes 10 preset multimedia resources. Therefore, the target feature vector specifically includes the feature vectors corresponding to the target object for each of the aforementioned 10 preset multimedia resources.

[0108] 104. Based on the target feature vectors of L preset multimedia resources for the target object, obtain the predicted interest degree of the target object for each preset multimedia resource through the interest degree prediction model.

[0109] In this embodiment, the terminal device uses the target feature vectors of the target object for L preset multimedia resources obtained in step 103 as input to the interest prediction model, and outputs the predicted interest level of the target object for each preset multimedia resource through the interest prediction model. Specifically, the predicted interest level is a percentage, and the predicted interest level represents the target object's interest in the preset multimedia resource, that is, it can represent the probability that the target object will access and purchase the preset multimedia resource in actual applications. For example, if the L preset multimedia resources are preset multimedia resource 1, preset multimedia resource 2, preset multimedia resource 3, and preset multimedia resource 4, then the interest prediction model will output the predicted interest level of the target object for preset multimedia resource 1, the predicted interest level of the target object for preset multimedia resource 2, the predicted interest level of the target object for preset multimedia resource 3, and the predicted interest level of the target object for preset multimedia resource 4. It should be understood that the foregoing examples are only for understanding this solution and should not be construed as a specific limitation of this solution.

[0110] 105. Sort the predicted interest of the target object in each preset multimedia resource in order, and determine the target multimedia resource to recommend to the target object from the L preset multimedia resources based on the sorting results.

[0111] In this embodiment, the terminal device ranks the predicted interest of the target object in each preset multimedia resource in order, and determines the target multimedia resource to recommend to the target object from L preset multimedia resources based on the ranking result. The target multimedia resource includes at least one preset multimedia resource. Specifically, the higher the predicted interest, the higher the target object's interest in the preset multimedia resource, indicating that the target object is more likely to access and purchase the preset multimedia resource in actual application. Conversely, the lower the predicted interest, the lower the target object's interest in the preset multimedia resource, indicating that the target object is less likely to access and purchase the preset multimedia resource in actual application.

[0112] Based on this, the target multimedia resource determined by the terminal device is the multimedia resource with the highest interest among L preset multimedia resources. For example, if the L preset multimedia resources are preset multimedia resource 1, preset multimedia resource 2, preset multimedia resource 3, and preset multimedia resource 4, and the predicted interest level of the target object for preset multimedia resource 1 is 80%, the predicted interest level for preset multimedia resource 2 is 20%, the predicted interest level for preset multimedia resource 3 is 60%, and the predicted interest level for preset multimedia resource 4 is 45%, then the predicted interest level of preset multimedia resource 1 (80%) is the highest. In this case, the terminal device can determine preset multimedia resource 1 as the target multimedia resource recommended to the target object, or, based on specific needs, determine preset multimedia resource 1 and preset multimedia resource 3 as the target multimedia resources recommended to the target object. The specific method for determining the target multimedia resource is not limited here, and the target multimedia resource can be one or more preset multimedia resources.

[0113] Furthermore, after determining the target multimedia resource to recommend to the target audience, the terminal device can also display the target multimedia resource on the terminal device's display interface to complete accurate multimedia resource recommendation. For ease of understanding, Figure 4 This is a schematic diagram of an interface for displaying target multimedia resources provided in an embodiment of this application, such as... Figure 4 As shown, Figure 4 Figure (A) specifically shows the multimedia resource recommendation interface of a certain application on a terminal device. This interface displays target multimedia resources, specifically preset multimedia resources B11 and B12. Secondly, Figure 4 Figure (B) specifically illustrates how a terminal device displays a target multimedia resource as a pop-up window on the display interface of an application, and the target multimedia resource displayed in the pop-up window is specifically the preset multimedia resource B2. It should be understood that... Figure 4The examples provided are for understanding how to display target multimedia resources. The specific display method needs to be flexibly determined based on the specific scenario and application requirements. Figure 4 The embodiments described herein should not be construed as specific limitations of this solution.

[0114] In this embodiment, a method for recommending multimedia resources is provided. Using the method described above, a first behavior sequence and a second behavior sequence are obtained based on behavioral data within different time periods. Since the first behavior sequence includes features of the same dimension within the first time period, and the second behavior sequence includes features of different dimensions within the second time period, feature modeling is performed on the first and second behavior sequences of the target object to obtain a target feature vector for the target object's interest in multiple preset multimedia resources. This target feature vector accurately represents the characteristics of the target object in various dimensions within different time periods. Therefore, the interest level of the target object in the multiple preset multimedia resources output by the interest prediction model is more consistent with the actual behavioral data of the target object, meaning the accuracy of the interest level in the multiple preset multimedia resources is high. Consequently, the target multimedia resources recommended to the target object are also more consistent with the actual behavioral data of the target object, thereby improving the accuracy of multimedia resource recommendation.

[0115] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, N first-row sequences and M second-row sequences are obtained according to the first-row data and the second-row data, specifically including:

[0116] Based on the data in the first row, N sequences of the first row are obtained;

[0117] Based on the data in the second row, M sequences of the second row are obtained.

[0118] In this embodiment, the terminal device divides behavioral data into different behavioral sequences according to different time periods. Specifically, N first behavioral sequences are obtained based on the first behavioral data, and M second behavioral sequences are obtained based on the second behavioral data. The following will describe in detail how the first and second behavioral sequences are obtained.

[0119] Optionally, in the above Figure 3 Based on the corresponding embodiments, in one optional embodiment of the multimedia resource recommendation method provided in this application, N first-row sequences are obtained based on the first-row data, specifically including:

[0120] Information extraction and processing are performed on the first behavior data to obtain the first access behavior corresponding to N dimensions within the first time period. The first access behavior is generated when the target object accesses the preset multimedia resources.

[0121] Based on the first access behavior corresponding to N dimensions within the first time period, N first behavior sequences are generated.

[0122] In this embodiment, the terminal device performs information extraction processing on the first behavior data to obtain the first access behavior corresponding to N dimensions within a first time period. This first access behavior is generated when the target object accesses a preset multimedia resource. Specifically, the dimensions described in this embodiment are multiple dimensions corresponding to the preset multimedia resource indicated by the access behavior, such as the multimedia resource identifier, the multimedia resource category, the specific industry, the brand, and the price range of the preset multimedia resource.

[0123] The target object exhibits numerous access behaviors within a relatively long first time period. To capture the sequence of behaviors within this period, the terminal device needs to characterize the target object's behavior across different dimensions. For example, the target object might frequently browse certain types of advertisements or purchase specific types of goods. Therefore, the terminal device extracts information from the first behavior data within the first time period, obtaining N dimensions, such as multimedia resource identifiers, multimedia resource categories, multimedia resource industries, and multimedia resource advertisers. Then, the first behavior data is segmented based on these N dimensions to obtain the first access behaviors corresponding to each dimension.

[0124] For example, the first set of data includes first access behavior 1, first access behavior 2, and first access behavior 3. First access behavior 1 accesses a multimedia resource with the identifier "id 1" and the category "clothing". First access behavior 2 accesses a multimedia resource with the identifier "id 2", the category "household products", and the advertiser "advertiser 1". First access behavior 3 accesses a multimedia resource with the identifier "id 3" and the advertiser "advertiser 2". Information extraction processing of the first set of data yields three dimensions: multimedia resource identifier, multimedia resource category, and multimedia resource advertiser. The multimedia resource identifier dimension corresponds to first access behavior 1, first access behavior 2, and first access behavior 3. Similarly, regarding the multimedia resource category dimension, since first access behavior 3 does not include multimedia resource category information, it is not among the access behaviors corresponding to the multimedia resource category dimension. Therefore, the multimedia resource category dimension corresponds to first access behavior 1 and first access behavior 2. Similarly, for the dimension of multimedia resource advertisers, since the first access behavior 1 does not include the information of multimedia resource advertisers, the first access behavior 1 is not among the access behaviors corresponding to the dimension of multimedia resource advertisers. That is, the dimension of multimedia resource advertisers can correspond to the first access behavior 2 and the first access behavior 3.

[0125] Based on this, the terminal device generates the aforementioned N first behavior sequences based on the first access behaviors corresponding to N dimensions within the first time period. Further, based on the previous example, since the multimedia resource identifier dimension can correspond to first access behavior 1, first access behavior 2, and first access behavior 3; the multimedia resource category dimension can correspond to first access behavior 1 and first access behavior 2; and the multimedia resource advertiser dimension can correspond to first access behavior 2 and first access behavior 3, the first access behaviors corresponding to the aforementioned three dimensions are encoded to generate three first behavior sequences.

[0126] Therefore, the three first behavior sequences specifically include the first behavior sequence of the multimedia resource identifier dimension, the first behavior sequence of the multimedia resource category dimension, and the first behavior sequence of the multimedia resource advertiser dimension. The first behavior sequence of the multimedia resource identifier dimension includes the features of the multimedia resource identifiers corresponding to the first access behavior 1, first access behavior 2, and first access behavior 3. Specifically, the first behavior sequence of the multimedia resource identifier dimension includes the multimedia resource identifier of the multimedia resource accessed by first access behavior 1 being "id 1", the multimedia resource identifier of the multimedia resource accessed by first access behavior 2 being "id 2", and the multimedia resource identifier of the multimedia resource accessed by first access behavior 3 being "id 3". Similarly, the first behavior sequence of the multimedia resource category dimension includes the features corresponding to the first access behavior 1 and first access behavior 2. Specifically, the first behavior sequence of the multimedia resource category dimension includes the multimedia resource category of the multimedia resource accessed by first access behavior 1 being "clothing", and the multimedia resource category of the multimedia resource accessed by first access behavior 2 being "household items". The first action sequence of the multimedia resource advertiser dimension consists of the features corresponding to the first access action 2 and the first access action 3. That is, the first action sequence of the multimedia resource advertiser dimension includes "Advertiser 1" for the multimedia resource accessed by the first access action 2 and "Advertiser 2" for the multimedia resource accessed by the first access action 3. It should be understood that the foregoing examples are only for understanding this solution and should not be construed as specific limitations of this solution.

[0127] This embodiment provides a method for generating behavior sequences. Using this method, information is extracted from first behavior data within a first time period with a relatively long time span, resulting in first access behaviors corresponding to N dimensions. This allows for the characterization of the target object's behavior within the first time period from different dimensions. The resulting first behavior sequence can describe the actual behavior data of the target object from multiple dimensions, thereby improving the flexibility and applicability of the first behavior sequence. Furthermore, in practical applications, there are usually many first access behaviors within a first time period with a long time span. Therefore, dividing the first access information based on different dimensions can also reduce the amount of data required for subsequent feature processing, thereby improving the efficiency of multimedia resource recommendation.

[0128] Optionally, in the above Figure 3 Based on the corresponding embodiments, in one optional embodiment of the multimedia resource recommendation method provided in this application, M second-row sequences are obtained based on the second-row data, specifically including:

[0129] Information extraction processing is performed on the second behavior data to obtain M second access behaviors within the second time period, wherein the second access behaviors are generated when the target object accesses the preset multimedia resources;

[0130] Based on M second access behaviors within the second time period, generate M second behavior sequences.

[0131] In this embodiment, the terminal device performs information extraction processing on the second behavior data to obtain M second access behaviors within a second time period. These second access behaviors are generated when the target object accesses preset multimedia resources. For example, if the object accesses preset multimedia resource 1, preset multimedia resource 2, and preset multimedia resource 3 once each within the second time period, then three second access behaviors can be obtained within the second time period. Second access behavior 1 is generated when the target object accesses preset multimedia resource 1, second access behavior 2 is generated when the target object accesses preset multimedia resource 2, and second access behavior 3 is generated when the target object accesses preset multimedia resource 3.

[0132] Based on this, the terminal device generates M second behavior sequences based on M second access behaviors within the second time period, where each second behavior sequence specifically represents a feature corresponding to multiple dimensions under the same access behavior. Specifically, the terminal device directly encodes each of the M second access behaviors within the second time period, thereby obtaining the second behavior sequence corresponding to each second access behavior. Again, based on the previous example, the terminal device encodes second access behavior 1, second access behavior 2, and second access behavior 3 respectively, obtaining second behavior sequence 1 corresponding to second access behavior 1, second behavior sequence 2 corresponding to second access behavior 2, and second behavior sequence 3 corresponding to second access behavior 3.

[0133] Furthermore, if the multimedia resource accessed by the second access behavior 1 has a multimedia resource identifier of "id 1" and a multimedia resource category of "clothing", the multimedia resource accessed by the second access behavior 2 has a multimedia resource identifier of "id 2", a multimedia resource category of "household products", and a multimedia resource advertiser of "advertiser 1", and the multimedia resource accessed by the second access behavior 3 has a multimedia resource identifier of "id 3" and a multimedia resource advertiser of "advertiser 2", then, since the second behavior sequence specifically represents the features corresponding to multiple dimensions under the same access behavior, then the second behavior sequence 1 is the multimedia resource identifier and multimedia resource category of the multimedia resource accessed by the second access behavior 1 being "clothing". Similarly, the second behavior sequence 2 is the multimedia resource identifier of the multimedia resource accessed by the second access behavior 2 being "id 2", the multimedia resource category being "household products", and the multimedia resource advertiser being "advertiser 1", and the second behavior sequence 3 is the multimedia resource identifier of the multimedia resource accessed by the second access behavior 3 being "id 3" and the multimedia resource advertiser being "advertiser 2". It should be understood that the above examples are only for understanding this solution and should not be construed as specific limitations of this solution.

[0134] In this embodiment, another method for generating behavior sequences is provided. By using the above method, information is extracted from the second behavior data within a second time period with a short time span, and multiple second access behaviors generated based on accessing preset multimedia resources are obtained. This directly depicts the access behavior of the target object within the second time period. Therefore, the obtained second behavior sequence is more consistent with the actual behavior data of the object in various dimensions within the second time period, thus ensuring the reliability of the second behavior sequence.

[0135] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, the target feature vector of the target object for L preset multimedia resources is determined according to N first row sequences and M second row sequences, specifically including:

[0136] The first feature vector of the target object for L preset multimedia resources is determined based on N first line sequences;

[0137] The second feature vector of the target object for L preset multimedia resources is determined based on M second action sequences;

[0138] The first feature vector and the second feature vector are fused together to obtain the target feature vector.

[0139] In this embodiment, the terminal device specifically determines the first feature vector of the target object to L preset multimedia resources based on N first line sequences, and determines the second feature vector of the target object to L preset multimedia resources based on M second line sequences, thereby fusing the first feature vector and the second feature vector to obtain the target feature vector. Specifically, the terminal device uses formula (3) to fuse the first feature vector and the second feature vector to obtain the target feature vector.

[0140] Formula (3) is as follows:

[0141] (3)

[0142] in, This refers to the target feature vector. This refers to the weight parameters. This refers to the second eigenvector. This refers to the first eigenvector.

[0143] In this embodiment, a method for determining a target feature vector is provided. Using the above method, since the first feature vector is obtained based on a first behavior sequence that can describe the actual behavior data of the target object from multiple dimensions, and the second feature vector is obtained based on a second behavior sequence that is more consistent with the actual behavior data of the object in various dimensions within a second time period, the first feature vector and the second feature vector are fused together. The resulting target feature vector can accurately represent the characteristics of the target object in various dimensions within different time periods. This allows the subsequent interest prediction model to learn more feature information based on the target feature vector. Therefore, the obtained interest is more consistent with the actual behavior data of the target object, that is, the accuracy of the interest of multiple preset multimedia resources is high, further improving the accuracy of multimedia resource recommendation.

[0144] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, the multimedia resource recommendation method further includes:

[0145] Obtain the basic attribute information of the target object;

[0146] The basic attribute information of the target object is characterized to obtain the user profile features of the target object;

[0147] Based on N first-line sequences, the first feature vector of the target object for L preset multimedia resources is determined, specifically including:

[0148] The first feature vector is determined based on N first behavior sequences and the user profile features of the target object;

[0149] The second feature vector of the target object for L preset multimedia resources is determined based on M second action sequences, specifically including:

[0150] The second feature vector is determined based on the M second behavior sequences and the user profile features of the target object;

[0151] The first feature vector and the second feature vector are fused to obtain the target feature vector, which specifically includes:

[0152] The first feature vector, the second feature vector, and the user profile features of the target object are fused together to obtain the target feature vector.

[0153] In this embodiment, the terminal device can also obtain the basic attribute information of the target object. The basic attribute information may include, but is not limited to, the target object's gender, age range, education level, and assets, etc., and is not specifically limited here. Based on this, the terminal device performs feature processing on the basic attribute information of the target object to obtain the user profile features of the target object. Specifically, if the basic attribute information of the target object includes P dimensions of basic information, then the obtained user profile features of the target object also include the profile features corresponding to P dimensions. For ease of understanding, formula (4) describes the user profile features of the target object.

[0154] Formula (4) is as follows:

[0155] (4)

[0156] in, This refers to the user profile characteristics of the target object. This refers to the profile feature corresponding to one dimension, while P refers to the number of dimensions included in the basic attribute information of the target object.

[0157] Based on this, the terminal device can determine the first feature vector of the target object for L preset multimedia resources according to N first behavior sequences and the user profile features of the target object. And it can determine the second feature vector of the target object for L preset multimedia resources according to M second behavior sequences and the user profile features of the target object. Furthermore, the first feature vector, the second feature vector, and the user profile features of the target object are fused to obtain the target feature vector. Specifically, the terminal device uses formula (5) to fuse the first feature vector, the second feature vector, and the user profile features of the target object to obtain the target feature vector.

[0158] Formula (5) is as follows:

[0159] (5)

[0160] in, This refers to the target feature vector. This refers to the weight parameters. , as well as This refers to the model parameters. This refers to the user profile characteristics of the target object. This refers to the second eigenvector. This refers to the first eigenvector.

[0161] It should be understood that in practical applications, the terminal device can specifically determine the feature vector of the target object within different time periods. It can also extract text features from specific pages of preset multimedia resources accessed within different time periods to obtain all contextual features corresponding to different behavioral data. These contextual features can include specific contextual information, the time of page access, and further, basic information about the terminal device itself, such as the specific type and model of the terminal device. This feature is then combined with the aforementioned behavioral sequence to perform a common feature modeling to obtain a more accurate and information-rich feature vector. The specific processing steps are not detailed here.

[0162] In this embodiment, a method for determining feature vectors is provided. By using the above method, the first feature vector, the second feature vector, and the user profile features are fused. This not only enables the target feature vector to accurately represent the features of the target object in various dimensions over different time periods, but also enables the target feature vector to more comprehensively represent the basic feature information of the target object. Therefore, subsequent interest prediction can learn features related to object behavior and user profile features, further improving the accuracy of the output interest, thereby improving the accuracy of multimedia resource recommendation.

[0163] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, the multimedia resource recommendation method further includes:

[0164] Obtain a multimedia resource feature set, wherein the multimedia resource feature set includes F multimedia resource feature subsets, each multimedia resource feature subset includes L preset multimedia resources with the same dimension of features, and different multimedia resource feature subsets include features with different dimensions, F is a positive integer greater than 1, and F is greater than or equal to N.

[0165] Based on N first behavior sequences and the user profile features of the target object, the first feature vector is determined, specifically including:

[0166] Based on N first behavior sequences, user profile features of the target object, and multimedia resource feature set, determine the first feature vector;

[0167] Based on the M second behavior sequences and the user profile features of the target object, the second feature vector is determined, specifically including:

[0168] Based on M second behavior sequences, user profile features of the target object, and multimedia resource feature set, determine the second feature vector;

[0169] Based on the target feature vector and the multimedia resource feature set, the predicted interest of the target object on each preset multimedia resource is obtained through the interest prediction model.

[0170] In this embodiment, the terminal device can also acquire a multimedia resource feature set, which includes F multimedia resource feature subsets, and each subset includes L preset multimedia resources with features of the same dimension. Different multimedia resource feature subsets include features of different dimensions, and F is a positive integer greater than 1, and F is greater than or equal to N. That is, the number of dimensions included in the aforementioned L preset multimedia resources can be the same as or greater than the number of dimensions included in the multiple access behaviors in the behavior data; no specific limitation is made here. Based on this, the terminal device can also use the target feature vector and the multimedia resource feature set as input to an interest prediction model, and obtain the predicted interest level of the target object for each preset multimedia resource through the interest prediction model.

[0171] Furthermore, formula (6) describes the set of multimedia resource features, as follows:

[0172] (6)

[0173] in, It refers to the set of features of multimedia resources. This refers to a subset of multimedia resource features corresponding to one dimension, and F refers to the total number of all dimensions included in the aforementioned L preset multimedia resources.

[0174] For example, let's take a multimedia resource feature set comprising three subsets: Multimedia Resource Feature Subset 1, Multimedia Resource Feature Subset 2, and Multimedia Resource Feature Subset 3. The features included in Multimedia Resource Feature Subset 1, Multimedia Resource Feature Subset 2, and Multimedia Resource Feature Subset 3 have different dimensions. Based on this, Multimedia Resource Feature Subset 1 includes the feature of multimedia resource identifiers for L preset multimedia resources. It should be understood that if any of the L preset multimedia resources does not have a corresponding multimedia resource identifier, then the feature value corresponding to the multimedia resource identifier of that preset multimedia resource is "0". Similarly, Multimedia Resource Feature Subset 2 includes the feature of multimedia resource categories for L preset multimedia resources. Likewise, if any of the L preset multimedia resources does not have a corresponding multimedia resource category, then the feature value corresponding to the multimedia resource category of that preset multimedia resource is "0". The multimedia resource feature subset 3 includes the features of the multimedia resource industry for L preset multimedia resources. As can be seen above, if any of the L preset multimedia resources does not have a corresponding multimedia resource industry, then the feature value corresponding to the multimedia resource industry of that preset multimedia resource is "0". It should be understood that the foregoing examples are only for understanding this solution and should not be construed as specific limitations of this solution.

[0175] Based on this, the terminal device specifically determines the first feature vector based on N first behavior sequences, the user profile features of the target object, and the multimedia resource feature set. For ease of understanding, the terminal device specifically determines the first feature vector through formulas (7) to (10).

[0176] Formula (7) is as follows:

[0177] (7)

[0178] in, This refers to the weight parameters. This refers to the hidden layer output corresponding to the nth dimension in the first row of the sequence. This refers to the user profile features of the target object, and N refers to the number of all dimensions included in the first behavior sequence. The hidden layer output corresponding to one dimension in the aforementioned first behavior sequence is obtained based on N first behavior sequences and the multimedia resource feature set.

[0179] Formula (8) is as follows:

[0180] (8)

[0181] in, This refers to the feature vector corresponding to the nth dimension in the first row of the sequence. This refers to the weight parameters. This refers to the hidden layer output corresponding to the nth dimension in the first row sequence, where N refers to the total number of dimensions included in the first row sequence.

[0182] Based on this, the same hidden layer is used to model each dimension in the first behavioral sequence, learning the object's preference for different dimensions of behavior within the first time period, thereby generating the first feature vector. Specifically, formula (9) is as follows:

[0183] (9)

[0184] in, This refers to the feature vector obtained by concatenating the feature vectors corresponding to multiple dimensions in the first row of the sequence. The first row refers to the feature vector corresponding to the nth dimension in the sequence, and N refers to the total number of dimensions included in the first row sequence.

[0185] Formula (10) is as follows:

[0186] (10)

[0187] in, This refers to the first eigenvector. This refers to the feature vector obtained by concatenating the feature vectors corresponding to multiple dimensions in the first row of the sequence. "b" refers to the model parameters, and "b" refers to the preset parameters.

[0188] Furthermore, the terminal device can also determine the second feature vector based on M second behavior sequences, user profile features of the target object, and multimedia resource feature set. For ease of understanding, the terminal device specifically determines the second feature vector through formulas (11) to (17), and the total length of the steps for modeling the second feature vector is preset to be T.

[0189] Formula (11) is as follows:

[0190] (11)

[0191] in, This refers to the weight parameters output by the intermediate layer at step t. as well as This refers to the model parameters. It refers to the set of features of multimedia resources. This refers to the final hidden layer output at step (t-1). This refers to the preset parameters.

[0192] Formula (12) is as follows:

[0193] (12)

[0194] in, This refers to the forgetting parameters output at step (t-1). as well as This refers to the model parameters. It refers to the set of features of multimedia resources. This refers to the final hidden layer output at step (t-1). This refers to the preset parameters.

[0195] Formula (13) is as follows:

[0196] (13)

[0197] in, This refers to the output weight parameters at step t. as well as This refers to the model parameters. It refers to the set of features of multimedia resources. This refers to the final hidden layer output at step (t-1). This refers to the preset parameters.

[0198] Formula (14) is as follows:

[0199] (14)

[0200] in, This refers to the output of the intermediate layer at step t. This refers to the output of the intermediate layer at step (t-1). This refers to the forgetting parameters output at step (t-1). This refers to the weight parameters output by the intermediate layer at step t. as well as This refers to the model parameters. It refers to the set of features of multimedia resources. This refers to the final hidden layer output at step (t-1). This refers to the preset parameters.

[0201] Formula (15) is as follows:

[0202] (15)

[0203] in, This refers to the final hidden layer output at step t. This refers to the output weight parameters at step t. This refers to the output of the intermediate layer at step t.

[0204] Based on this, each preset multimedia resource accessed by the second access behavior in the second behavior sequence corresponds to a final hidden layer output. Then, an attention layer is used to learn the target object's preference for the preset multimedia resources accessed by different second access behaviors. Specifically, formula (16) is as follows:

[0205] (16)

[0206] in, This refers to the degree of preference a target object has for a pre-defined multimedia resource accessed by a second access action. This refers to the final hidden layer output at step t. It refers to the set of features of multimedia resources. This refers to the final hidden layer output of the nth second access action at step t. This refers to the subset of multimedia resource features corresponding to the nth second access behavior.

[0207] Based on this, the final second feature vector is shown in formula (17):

[0208] (17)

[0209] in, This refers to the second eigenvector. This refers to the degree of preference a target object has for a pre-defined multimedia resource accessed by a second access action. This refers to the final hidden layer output at step t.

[0210] In this embodiment, another method for determining target feature vectors is provided. Using the above method, feature modeling is performed based on behavioral sequences, user profile features of the target object, and multimedia resource feature sets. The resulting first feature vector can accurately represent the target object's preference for different dimensions of behavior in the first time period, while the second feature vector can accurately represent the target object's preference for the preset multimedia resources accessed in the second time period. Therefore, the completeness of the information represented by the subsequently obtained target feature vector is further improved, so that the target feature vector can more comprehensively represent the feature information of the target object. Furthermore, the accuracy of the subsequently obtained interest is higher, thereby improving the accuracy of multimedia resource recommendation.

[0211] Optionally, in the above Figure 3Based on the corresponding embodiments, in one optional embodiment of the multimedia resource recommendation method provided in this application, obtaining a multimedia resource feature set specifically includes:

[0212] Obtain a preset set of multimedia resources, wherein the preset set of multimedia resources includes L preset multimedia resources, and the L preset multimedia resources include F features.

[0213] Based on F dimensions, feature extraction is performed on L preset multimedia resources to obtain a multimedia resource feature subset corresponding to each dimension;

[0214] A multimedia resource feature set is generated based on the multimedia resource feature subset corresponding to each dimension.

[0215] In this embodiment, the terminal device acquires a preset multimedia resource set, which includes L preset multimedia resources, and the L preset multimedia resources include F dimensions of features. Specifically, in practical applications, the target object usually not only focuses on the preset multimedia resources themselves, but also needs to focus on the different dimensions of multimedia resource attributes corresponding to the preset multimedia resources. For example, the type of preset multimedia resources, the industry of preset multimedia resources, and the price of preset multimedia resources are all different dimensions of multimedia resource attributes that the target object may focus on.

[0216] Therefore, this solution uses different dimensions of multimedia resource attributes to characterize and represent preset multimedia resources. For example, it uses multimedia resource identifiers, industry classifications, and primary and secondary classifications based on actual conditions. Based on this, the terminal device extracts features from L preset multimedia resources using F dimensions for each preset multimedia resource, obtaining a subset of multimedia resource features corresponding to each dimension, i.e., the subsets of multimedia resource features corresponding to the F dimensions. This results in subsets of multimedia resource features corresponding to different dimensions of multimedia resource attributes. Therefore, the dimensions of features included in each subset of multimedia resource features are different. It should be understood that not all preset multimedia resources include features of the F dimensions; the aforementioned F dimensions are features shared by the L dimensions. Finally, the terminal device concatenates the subsets of multimedia resource features corresponding to each dimension to generate a multimedia resource feature set.

[0217] For example, the preset multimedia resource set includes five preset multimedia resources as an example for illustration, namely, preset multimedia resource 1, preset multimedia resource 2, preset multimedia resource 3, preset multimedia resource 4, and preset multimedia resource 10. If the aforementioned five preset multimedia resources collectively include five dimensions of features, that is, preset multimedia resources 1 to preset multimedia resources 5 collectively include multimedia resource identification features, multimedia resource category features, multimedia resource industry features, text semantic features, and image features, and the text semantic features are features obtained after text processing of the preset multimedia resources, and the image features are features obtained after image processing of the preset multimedia resources. Based on this, feature extraction is performed on the aforementioned five preset multimedia resources according to the aforementioned five dimensions to obtain multimedia resource feature subsets corresponding to the aforementioned five dimensions. That is, the obtained five multimedia resource feature subsets include multimedia resource identification feature subset, multimedia resource category feature subset, multimedia resource industry feature subset, text semantic feature subset, and image feature subset. Then, the multimedia resource feature subsets corresponding to the aforementioned five dimensions are concatenated to generate a multimedia resource feature set, that is, the multimedia resource feature set includes multimedia resource identification feature subset, multimedia resource category feature subset, multimedia resource industry feature subset, text semantic feature subset, and image feature subset.

[0218] This embodiment provides a method for generating a multimedia resource feature set. Using this method, different dimensions of multimedia resource attributes are used to characterize and represent preset multimedia resources, thereby obtaining a multimedia resource feature subset corresponding to each dimension. The multimedia resource feature subsets corresponding to each dimension are then concatenated to generate a multimedia resource feature set. Therefore, the obtained multimedia resource feature set can more accurately reflect the feature information of the preset multimedia resources in each dimension, thereby improving the comprehensiveness of feature information learned by the subsequent model, thus improving the accuracy of interest and ultimately enhancing the accuracy of multimedia resource recommendations.

[0219] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, the multimedia resource recommendation method further includes:

[0220] The first set of data samples, the second set of data samples, the preset multimedia resource set, and the real tag set of the target object are obtained. The first set of data samples includes multiple first set of data samples, the second set of data samples includes multiple second set of data samples, the first set of data samples and the second set of data samples are data samples from different time periods, the preset multimedia resource set includes L preset multimedia resources, and the real tag set includes the real tags corresponding to each preset multimedia resource of the target object.

[0221] Based on the first row of data sample set and the second row of data sample set, the first row of sample sequence set and the second row of sample sequence set are determined. The first row of sample sequence set includes multiple first row of sample sequences, and the second row of sample sequence set includes multiple second row of sample sequences. The first row of sample sequences includes features of the same dimension, and the second row of sample sequences includes features of different dimensions.

[0222] Based on the first row of sample sequence sets and the second row of sample sequence sets, determine the target feature vectors of the target object for L preset multimedia resources;

[0223] Based on the target feature vectors of the target object to L preset multimedia resources and the preset multimedia resource set, the predicted interest degree of the target object to each preset multimedia resource is obtained through the interest degree prediction model to be trained.

[0224] Based on the real tag set and the predicted interest level of the target object for each preset multimedia resource, the interest prediction model is trained to obtain the interest prediction model.

[0225] In this embodiment, before recommending multimedia resources, the terminal device needs to obtain the interest prediction model required in this solution through the interest prediction model. First, the terminal device can obtain a first behavioral data sample set and a second behavioral data sample set of the target object. The first behavioral data sample set includes multiple first behavioral data samples, and the second behavioral data sample set includes multiple second behavioral data samples. The first and second behavioral data sample sets are behavioral data samples from different time periods, and the time span of the first behavioral data sample time period includes the time span of the second behavioral data sample time period. For example, the first behavioral data sample set may include multiple behavioral data samples from the second quarter of 2021 (April to June), then the first behavioral data sample set may include multiple behavioral data samples from April 2021. The specifics are similar to the first and second behavioral data described in the previous embodiment, and will not be repeated here.

[0226] Secondly, the terminal device can also obtain a preset multimedia resource set and a real tag set. The preset multimedia resource set includes L preset multimedia resources, the specific L preset multimedia resources being similar to those described in the previous embodiments and will not be repeated here. The real tag set includes real tags corresponding to each preset multimedia resource for the target object. The real tags are determined by manual annotation. For example, if the target object is interested in preset multimedia resource 1, the real tag is 1; if it is not interested in preset multimedia resource 1, the real tag is 0. Thus, the real tag corresponding to preset multimedia resource 1 is obtained. The aforementioned tag annotation is performed on all L preset multimedia resources to obtain the real tag corresponding to each preset multimedia resource, thereby generating the aforementioned real tag set.

[0227] Furthermore, the terminal device determines a first-row sample sequence set and a second-row sample sequence set based on the first-row data sample set and the second-row data sample set. The first-row sample sequence set includes multiple first-row sample sequences, and the second-row sample sequence set includes multiple second-row sample sequences. The first-row sample sequences include features of the same dimension, and the second-row sample sequences include features of different dimensions. The method for obtaining the corresponding behavior sample sequence set from the behavior data sample set is similar to that in the aforementioned embodiments and will not be described again here.

[0228] Therefore, the terminal device determines the target feature vector of the target object for L preset multimedia resources based on the first row of sample sequence sets and the second row of sample sequence sets. Specifically, in order to ensure that the features of the input model can comprehensively represent the basic feature information of the target object, the terminal device can also obtain the basic attribute information of the target object, and perform feature processing on the basic attribute information of the target object using a method similar to that in the aforementioned embodiments to obtain the user profile features of the target object. Secondly, in order to ensure that the features of the input model can more accurately reflect the feature information of the preset multimedia resources in various dimensions, the terminal device can also obtain a multimedia resource feature set based on the preset multimedia resource set using a method similar to that in the aforementioned embodiments. Based on this, the terminal device determines the first feature vector of the target object for L preset multimedia resources by combining multiple first row of sample sequences, the user profile features of the target object, and the multimedia resource feature set, and determines the second feature vector of the target object for L preset multimedia resources based on multiple second row of sample sequences, the user profile features of the target object, and the multimedia resource feature set. Finally, the terminal device fuses the first feature vector, the second feature vector, and the user profile features of the target object using a method similar to that in the aforementioned embodiments to obtain the aforementioned target feature vector.

[0229] Furthermore, the terminal device obtains a multimedia resource feature set based on a preset multimedia resource set in the aforementioned manner. Therefore, it can specifically use the aforementioned target feature vector and multimedia resource feature set as input to the interest prediction model to be trained. Then, the interest prediction model to be trained outputs the predicted interest of the target object to each preset multimedia resource. Based on the real label set and the predicted interest of the target object to each preset multimedia resource, the interest prediction model to be trained is trained to obtain the interest prediction model.

[0230] Optionally, in the above Figure 3 Based on the corresponding embodiments, in an optional embodiment of the multimedia resource recommendation method provided in this application, the interest prediction model to be trained is trained according to the real tag set and the predicted interest of the target object for each preset multimedia resource to obtain the interest prediction model, specifically including:

[0231] Based on the predicted interest of the target object in each preset multimedia resource, the predicted label corresponding to the target object for each preset multimedia resource is determined;

[0232] Based on the real tag set and the predicted tag corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model to be trained are updated according to the target loss function to obtain the interest prediction model.

[0233] In this embodiment, the terminal device determines the predicted label for each preset multimedia resource based on the predicted interest level of the target object. Specifically, it judges the predicted label for each preset multimedia resource by comparing it with a preset interest level threshold. A predicted label greater than or equal to the preset interest level threshold is classified as one type, and one less than the threshold as another. For example, taking a preset interest level threshold of 60% as an example, if the predicted interest level of the target object for the preset multimedia resource is 80% (greater than 60%), then the predicted label for that preset multimedia resource is determined to be 1. Conversely, if the predicted interest level is 40% (less than 60%), then the predicted label is determined to be 0. Based on this, the terminal device updates the model parameters of the interest prediction model to be trained according to the target loss function, based on the real label set and the predicted label for each preset multimedia resource, to obtain the interest prediction model.

[0234] Specifically, the target object's real label for each preset multimedia resource is used as the target for iterative training. That is, the loss value of the target loss function is determined based on the difference between the target object's real label for each preset multimedia resource and the predicted label for each preset multimedia resource. The loss value of the target loss function is used to determine whether the loss function has reached the convergence condition. If the convergence condition has not been reached, the model parameters of the interest prediction model to be trained are updated using the loss value of the target loss function until the target loss function reaches the convergence condition. Then, the loss value of the target loss function at this step is determined as the model parameters of the interest prediction model to be trained, thereby obtaining the interest prediction model.

[0235] To understand the aforementioned training process, Figure 5 This is a schematic diagram illustrating a process for training an interest prediction model provided in an embodiment of this application, such as... Figure 5 As shown, the terminal device first acquires a first-row data sample set C1, a second-row data sample set C2, a preset multimedia resource set C3, and a real label set C4 for the target object. Then, based on the first-row data sample set C1, it determines a first-row sample sequence set C5, and based on the second-row data sample set C2, it determines a second-row sample sequence set C6. Next, based on the first-row sample sequence set C5 and the second-row sample sequence set C6, it determines the target feature vector C7 for the target object with respect to L preset multimedia resources. The target feature vector C7 for the target object with respect to the L preset multimedia resources and the preset multimedia resource set C3 are used as input to an interest prediction model C8 to be trained. The interest prediction model C8 then outputs the predicted interest level C9 of the target object for each preset multimedia resource. Finally, based on the predicted interest level C9 of the target object with respect to each preset multimedia resource, it determines the predicted label C10 corresponding to the target object for each preset multimedia resource. Therefore, based on the true label set C4 and the predicted label C10 corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model C8 to be trained are updated according to the target loss function. That is, based on the predicted label C10 corresponding to each preset multimedia resource for the target object, the true label set C4, and the target loss function, the interest prediction model C8 to be trained is iteratively trained until the updated model parameters make the target loss function converge, thereby generating the interest prediction model. It should be understood that... Figure 5 The examples in this paper are for the convenience of understanding this solution and are not intended to limit this solution.

[0236] Furthermore, the target loss function in this scheme is as shown in formula (18), which is as follows:

[0237] (18)

[0238] in, This refers to the target loss function. This refers to the predicted tags for each preset multimedia resource for the target object. G refers to the real label corresponding to each preset multimedia resource for the target object, and G refers to the number of samples.

[0239] It should be understood that the convergence condition of the aforementioned target loss function can be that the value of the target loss function is less than or equal to a first preset threshold. For example, the value of the first preset threshold can be 0.005, 0.01, 0.02 or other values ​​close to 0. Alternatively, the difference between two consecutive values ​​of the target loss function can be less than or equal to a second preset threshold. The value of the second threshold can be the same as or different from the value of the first threshold. For example, the value of the second preset threshold can be 0.005, 0.01, 0.02 or other values ​​close to 0. The terminal device can also use other convergence conditions, which are not limited here.

[0240] This embodiment provides a method for training an interest prediction model. Using the above method, an interest prediction model is trained based on a target feature vector, a multimedia resource feature set, and a set of real labels. Since the target feature vector can accurately represent the characteristics of the target object in various dimensions at different time periods, and the multimedia resource feature set can more accurately reflect the feature information of the preset multimedia resources in various dimensions, the interest prediction model learns more comprehensive and accurate information about the object's relationship with the preset multimedia resources through multiple iterations during the training process. Therefore, the obtained interest prediction model has high reliability, ensuring that the subsequent output of interest values ​​is more consistent with the actual behavioral data of the target object, thereby improving the accuracy of multimedia resource recommendations.

[0241] Optionally, in the above Figure 3 Based on the corresponding embodiments, in one optional embodiment of the multimedia resource recommendation method provided in this application, the predicted interest of the target object in each preset multimedia resource is ordered, and the target multimedia resource recommended to the target object is determined based on the order order result. Specifically, this includes:

[0242] The maximum heap sort method is used to sort the predicted interest of the target object in each preset multimedia resource in order;

[0243] The order of the results is determined by sorting the predicted interest in each preset multimedia resource from highest to lowest.

[0244] Extract target interest levels from high to low based on the sequentially arranged results;

[0245] The preset multimedia resources corresponding to the target interest level are identified as the target multimedia resources.

[0246] In this embodiment, the terminal device uses max-heap sort to sequentially arrange the predicted interest levels of the target object for each preset multimedia resource, and determines the sequential arrangement result as a descending order of the predicted interest levels of each preset multimedia resource. For example, the preset multimedia resource set includes preset multimedia resource 1, preset multimedia resource 2, preset multimedia resource 3, preset multimedia resource 4, and preset multimedia resource 5, and the predicted interest level of the target object for preset multimedia resource 1 is 75%, the predicted interest level for preset multimedia resource 2 is 20%, the predicted interest level for preset multimedia resource 3 is 60%, the predicted interest level for preset multimedia resource 4 is 45%, and the predicted interest level for preset multimedia resource 5 is 90%. Based on this, the obtained sequential arrangement result is the predicted interest level of preset multimedia resource 5, the predicted interest level of preset multimedia resource 1, the predicted interest level of preset multimedia resource 3, the predicted interest level of preset multimedia resource 4, and so on down to the predicted interest level of preset multimedia resource 2.

[0247] Furthermore, the terminal device extracts the target interest scores sequentially from high to low based on the ordered results, and identifies the preset multimedia resources corresponding to the target interest scores as the target multimedia resources. The aforementioned target interest scores may include the predicted interest scores of one or more preset multimedia resources, without limitation. For example, based on the aforementioned example, since the ordered results are the predicted interest scores of preset multimedia resource 5, preset multimedia resource 1, preset multimedia resource 3, preset multimedia resource 4, and so on, up to the predicted interest scores of preset multimedia resource 2, if only one multimedia resource needs to be recommended based on the scenario and application requirements, then the terminal device can identify the predicted interest score of preset multimedia resource 5 as the target interest score, meaning preset multimedia resource 5 can be identified as the target multimedia resource. Similarly, if three multimedia resources need to be recommended based on the scenario and application requirements, then the terminal device can identify the predicted interest scores of preset multimedia resource 5, preset multimedia resource 1, and preset multimedia resource 3 as the target interest scores, meaning preset multimedia resource 5, preset multimedia resource 1, and preset multimedia resource 3 can be identified as the target multimedia resources. It should be understood that the foregoing examples are for the purpose of understanding this solution only and should not be construed as specific limitations of this solution.

[0248] In this embodiment, another method for recommending multimedia resources is provided. Using the above method, multiple preset multimedia resources are sorted from high to low in terms of interest. Based on the demand, the required number of target interest values ​​are extracted from high to low. Target multimedia resources are then determined based on the target interest values, making the multimedia resource recommendations more consistent with the target user's interest in the preset multimedia resources, thereby ensuring the reliability of the multimedia resource recommendations.

[0249] Figure 6 This is a schematic diagram of the structure of a multimedia resource recommendation device provided in an embodiment of this application, as shown below. Figure 6 As shown, the multimedia resource recommendation device 600 includes:

[0250] The acquisition module 601 is used to acquire the first row data of the target object in a first time period and the second row data in a second time period, wherein the length of the first time period is greater than the length of the second time period, and the first time period includes the second time period.

[0251] The acquisition module 601 is also used to obtain N first-row sequences and M second-row sequences based on the first-row data and the second-row data, wherein the first-row sequences include features of the same dimension within the first time period, and the second-row sequences include features of different dimensions within the second time period, and M and N are both positive integers greater than 1.

[0252] The determination module 602 is used to determine the target feature vector of the target object to L preset multimedia resources based on N first row sequences and M second row sequences, where L is a positive integer greater than 1;

[0253] The acquisition module 601 is also used to obtain the predicted interest degree of the target object on each preset multimedia resource based on the target feature vector of the target object on L preset multimedia resources through the interest degree prediction model;

[0254] The determining module 602 is further configured to sequentially arrange the predicted interest of the target object in each preset multimedia resource, and determine the target multimedia resource to be recommended to the target object from the L preset multimedia resources based on the sequential arrangement result, wherein the target multimedia resource includes at least one preset multimedia resource.

[0255] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the acquisition module 601 is specifically used to obtain N first row sequences based on the first row data;

[0256] Based on the data in the second row, M sequences of the second row are obtained.

[0257] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the acquisition module 601 is specifically used to perform information extraction processing on the first behavior data to obtain the first access behavior corresponding to N dimensions in the first time period, wherein the first access behavior is generated when the target object accesses the preset multimedia resource;

[0258] Based on the first access behavior corresponding to N dimensions within the first time period, N first behavior sequences are generated.

[0259] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the acquisition module 601 is specifically used to perform information extraction processing on the second behavior data to obtain M second access behaviors within a second time period, wherein the second access behaviors are generated when the target object accesses a preset multimedia resource;

[0260] Based on M second access behaviors within the second time period, generate M second behavior sequences.

[0261] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the determining module 602 is specifically used to determine the first feature vector of the target object to L preset multimedia resources according to N first action sequences;

[0262] The second feature vector of the target object for L preset multimedia resources is determined based on M second action sequences;

[0263] The first feature vector and the second feature vector are fused together to obtain the target feature vector.

[0264] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the multimedia resource recommendation device 600 further includes a processing module 603;

[0265] The acquisition module 601 is also used to acquire basic attribute information of the target object;

[0266] Processing module 603 is used to perform feature processing on the basic attribute information of the target object to obtain the user profile features of the target object;

[0267] The determination module 602 is specifically used to determine the first feature vector based on N first behavior sequences and the user profile features of the target object;

[0268] The determination module 602 is specifically used to determine the first feature vector based on N first behavior sequences and the user profile features of the target object;

[0269] The determination module 602 is specifically used to fuse the first feature vector, the second feature vector, and the user profile features of the target object to obtain the target feature vector.

[0270] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the acquisition module 601 is further used to acquire a multimedia resource feature set, wherein the multimedia resource feature set includes F multimedia resource feature subsets, each multimedia resource feature subset includes L preset multimedia resources with the same dimension of features, and different multimedia resource feature subsets include features with different dimensions, F is a positive integer greater than 1, and F is greater than or equal to N;

[0271] The determination module 602 is specifically used to determine the first feature vector based on N first behavior sequences, user profile features of the target object, and multimedia resource feature set;

[0272] The determination module 602 is specifically used to determine the second feature vector based on M second behavior sequences, user profile features of the target object, and multimedia resource feature set;

[0273] The acquisition module 601 is specifically used to obtain the predicted interest of the target object on each preset multimedia resource based on the target feature vector and the multimedia resource feature set, through an interest prediction model.

[0274] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the acquisition module 601 is specifically used to acquire a preset multimedia resource set, wherein the preset multimedia resource set includes L preset multimedia resources, and the L preset multimedia resources include F dimensions of features.

[0275] Based on F dimensions, feature extraction is performed on L preset multimedia resources to obtain a feature subset of multimedia resources corresponding to each dimension;

[0276] A multimedia resource feature set is generated based on the multimedia resource feature subset corresponding to each dimension.

[0277] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the multimedia resource recommendation device 600 further includes a training module 604;

[0278] The acquisition module 601 is also used to acquire a first set of behavioral data samples, a second set of behavioral data samples, a set of preset multimedia resources, and a set of real tags for the target object. The first set of behavioral data samples includes multiple first set of behavioral data samples, the second set of behavioral data samples includes multiple second set of behavioral data samples, the first set of behavioral data samples and the second set of behavioral data samples are behavioral data samples within different time periods, the set of preset multimedia resources includes L preset multimedia resources, and the set of real tags includes the real tags corresponding to each preset multimedia resource for the target object.

[0279] The determining module 602 is further configured to determine a first row sample sequence set and a second row sample sequence set based on the first row data sample set and the second row data sample set, wherein the first row sample sequence set includes multiple first row sample sequences, the second row sample sequence set includes multiple second row sample sequences, and the first row sample sequences include features of the same dimension, while the second row sample sequences include features of different dimensions.

[0280] The determining module 602 is also used to determine the target feature vector of the target object for L preset multimedia resources based on the first row of the sample sequence set and the second row of the sample sequence set;

[0281] The acquisition module 601 is also used to obtain the predicted interest of the target object on each preset multimedia resource based on the target feature vector of the target object on L preset multimedia resources and the preset multimedia resource set, through the interest prediction model to be trained.

[0282] The training module 604 is used to train the interest prediction model to be trained based on the real label set and the predicted interest of the target object for each preset multimedia resource, so as to obtain the interest prediction model.

[0283] Optionally, in the above Figure 6 Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the training module 604 is specifically used to determine the predicted label corresponding to each preset multimedia resource by the target object based on the predicted interest degree of the target object to each preset multimedia resource.

[0284] Based on the real tag set and the predicted tag corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model to be trained are updated according to the target loss function to obtain the interest prediction model.

[0285] Optionally, in the above Figure 6Based on the corresponding embodiments, in another embodiment of the multimedia resource recommendation device 600 provided in this application, the determining module 602 is specifically used to use the max heap sort method to sequentially arrange the predicted interest of the target object for each preset multimedia resource;

[0286] The order of the results is determined by sorting the predicted interest in each preset multimedia resource from highest to lowest.

[0287] Extract target interest levels from high to low based on the sequentially arranged results;

[0288] The preset multimedia resources corresponding to the target interest level are identified as the target multimedia resources.

[0289] This application also provides another multimedia resource recommendation device, which is deployed on terminal devices, such as... Figure 7 As shown, for ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application. The following explanation uses a mobile phone as an example of a terminal device:

[0290] Figure 7 This is a block diagram illustrating a portion of the structure of a mobile phone related to the terminal provided in the embodiments of this application. (Reference) Figure 7 The mobile phone includes components such as a radio frequency (RF) circuit 1110, a memory 1120, an input unit 1130, a display unit 1140, a sensor 1150, an audio circuit 1160, a wireless fidelity (WiFi) module 1170, a processor 1180, and a power supply 1190. Those skilled in the art will understand that... Figure 7 The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0291] The following is combined with Figure 7 A detailed introduction to each component of a mobile phone:

[0292] RF circuit 1110 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with processor 1180; additionally, it transmits uplink data to the base station. Typically, RF circuit 1110 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, etc. Furthermore, RF circuit 1110 can also communicate wirelessly with networks and other devices. The aforementioned wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0293] The memory 1120 can be used to store software programs and modules. The processor 1180 executes various mobile phone functions and data processing by running the software programs and modules stored in the memory 1120. The memory 1120 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 1120 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0294] The input unit 1130 can be used to receive input numerical or character information, and to generate key signal inputs related to the object settings and function control of the mobile phone. Specifically, the input unit 1130 may include a touch panel 1131 and other input devices 1132. The touch panel 1131, also known as a touch screen, can collect touch operations on or near the object (such as operations performed by the object using a finger, stylus, or any suitable object or accessory on or near the touch panel 1131), and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 1131 may include two parts: a touch detection device and a touch controller. The touch detection device detects the touch position of the object and detects the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 1180, and can also receive and execute commands sent by the processor 1180. In addition, the touch panel 1131 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 1131, the input unit 1130 may also include other input devices 1132. Specifically, other input devices 1132 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0295] Display unit 1140 can be used to display information input by an object or information provided to an object, as well as various menus of the mobile phone. Display unit 1140 may include display panel 1141, optionally configured as a Liquid Crystal Display (LCD), Organic Light-Emitting Diode (OLED), or similar display panel 1141. Further, touch panel 1131 may cover display panel 1141. When touch panel 1131 detects a touch operation on or near it, it transmits the information to processor 1180 to determine the type of touch event. Subsequently, processor 1180 provides corresponding visual output on display panel 1141 based on the type of touch event. Although in Figure 7 In this embodiment, the touch panel 1131 and the display panel 1141 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 1131 and the display panel 1141 can be integrated to realize the input and output functions of the mobile phone.

[0296] The mobile phone may also include at least one sensor 1150, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 1141 according to the ambient light level, and the proximity sensor can turn off the display panel 1141 and / or the backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, are not described in detail here.

[0297] Audio circuit 1160, speaker 1161, and microphone 1162 provide an audio interface between the device and the mobile phone. Audio circuit 1160 converts received audio data into electrical signals and transmits them to speaker 1161, where speaker 1161 converts them into sound signals for output. On the other hand, microphone 1162 converts collected sound signals into electrical signals, which are received by audio circuit 1160, converted into audio data, and then processed by processor 1180 before being transmitted via RF circuit 1110 to, for example, another mobile phone, or the audio data can be output to memory 1120 for further processing.

[0298] WiFi is a short-range wireless transmission technology. A mobile phone using the WiFi module 1170 can help users send and receive emails, browse web pages, and access streaming media, providing them with wireless broadband internet access. Although Figure 7 WiFi module 1170 is shown, but it is understood that it is not an essential component of a mobile phone.

[0299] The processor 1180 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes various functions and processes data by running or executing software programs and / or modules stored in the memory 1120, and by calling data stored in the memory 1120. Optionally, the processor 1180 may include one or more processing units; preferably, the processor 1180 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1180.

[0300] The mobile phone also includes a power supply 1190 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 1180 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0301] Although not shown, mobile phones may also include a camera, Bluetooth module, etc., which will not be elaborated here.

[0302] In this embodiment of the application, the processor 1180 included in the terminal is used to perform, for example... Figure 3 The illustrated embodiments and Figure 3 The corresponding embodiments.

[0303] This application also provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the aforementioned actions. Figure 3 The steps performed by the terminal device in the method described in the illustrated embodiment.

[0304] This application also provides a computer program product including a program, which, when run on a computer, causes the computer to perform the aforementioned actions. Figure 3 The steps performed by the terminal device in the method described in the illustrated embodiment.

[0305] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0306] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, at least two units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0307] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across at least two network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0308] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0309] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0310] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for recommending multimedia resources, characterized in that, include: Acquire the first behavior data of the target object within a first time period and the second behavior data within a second time period, wherein the length of the first time period is greater than the length of the second time period, and the first time period includes the second time period. The first behavior data is processed by information extraction to obtain the first access behavior corresponding to N dimensions within the first time period, wherein the first access behavior is generated when the target object accesses the preset multimedia resources; Based on the first access behaviors corresponding to N dimensions within the first time period, N first behavior sequences are generated; the first behavior sequences include features of the same dimension within the first time period. The second behavior data is processed by information extraction to obtain M second access behaviors within the second time period, wherein the second access behaviors are generated when the target object accesses the preset multimedia resource; Based on the M second access behaviors within the second time period, M second behavior sequences are generated, wherein the second behavior sequence includes features of different dimensions within the second time period, where M and N are both positive integers greater than 1, and the dimension is the dimension corresponding to the preset multimedia resource indicated by the user's access behavior. Based on the N first line sequences and the M second line sequences, the target feature vectors of the target object for L preset multimedia resources are determined, where L is a positive integer greater than 1; Based on the target feature vectors of the target object to L preset multimedia resources, the predicted interest degree of the target object to each preset multimedia resource is obtained through the interest degree prediction model. The predicted interest of the target object in each preset multimedia resource is ordered, and a target multimedia resource recommended to the target object is determined from the L preset multimedia resources based on the ordering result. The target multimedia resource includes at least one preset multimedia resource.

2. The method according to claim 1, characterized in that, The step of determining the target feature vector of the target object for L preset multimedia resources based on the N first action sequences and the M second action sequences includes: The first feature vector of the target object for the L preset multimedia resources is determined based on the N first action sequences; The second feature vector of the target object for the L preset multimedia resources is determined based on the M second behavior sequences; The first feature vector and the second feature vector are fused together to obtain the target feature vector.

3. The method according to claim 2, characterized in that, The method further includes: Obtain the basic attribute information of the target object; The basic attribute information of the target object is subjected to feature processing to obtain the user profile features of the target object; The step of determining the first feature vector of the target object for the L preset multimedia resources based on the N first action sequences includes: The first feature vector is determined based on the N first behavior sequences and the user profile features of the target object; Determining the second feature vector of the target object for the L preset multimedia resources based on the M second action sequences includes: The second feature vector is determined based on the M second behavior sequences and the user profile features of the target object; The step of fusing the first feature vector and the second feature vector to obtain the target feature vector includes: The first feature vector, the second feature vector, and the user profile features of the target object are fused together to obtain the target feature vector.

4. The method according to claim 3, characterized in that, The method further includes: Obtain a multimedia resource feature set, wherein the multimedia resource feature set includes F multimedia resource feature subsets, each multimedia resource feature subset includes features of the same dimension of the L preset multimedia resources, and different multimedia resource feature subsets include features of different dimensions, F is a positive integer greater than 1, and F is greater than or equal to N; The step of determining the first feature vector based on the N first behavior sequences and the user profile features of the target object includes: The first feature vector is determined based on the N first behavior sequences, the user profile features of the target object, and the multimedia resource feature set; The step of determining the second feature vector based on the M second behavior sequences and the user profile features of the target object includes: The second feature vector is determined based on the M second behavior sequences, the user profile features of the target object, and the multimedia resource feature set; The step of obtaining the predicted interest degree of the target object for each preset multimedia resource based on the target feature vector of the target object for L preset multimedia resources through an interest degree prediction model includes: Based on the target feature vector and the multimedia resource feature set, the predicted interest of the target object in each preset multimedia resource is obtained through the interest prediction model.

5. The method according to claim 4, characterized in that, The acquisition of the multimedia resource feature set includes: Obtain a preset multimedia resource set, wherein the preset multimedia resource set includes the L preset multimedia resources, and the L preset multimedia resources include F dimensions of features; Based on the F dimensions, feature extraction is performed on the L preset multimedia resources to obtain a multimedia resource feature subset corresponding to each dimension; The multimedia resource feature set is generated based on the multimedia resource feature subset corresponding to each dimension.

6. The method according to claim 1, characterized in that, The method further includes: The first set of behavioral data samples, the second set of behavioral data samples, the preset set of multimedia resources, and the set of real tags for the target object are obtained. The first set of behavioral data samples includes multiple first behavioral data samples, the second set of behavioral data samples includes multiple second behavioral data samples, the first set of behavioral data samples and the second set of behavioral data samples are behavioral data samples from different time periods, the preset set of multimedia resources includes the L preset multimedia resources, and the set of real tags includes the real tags corresponding to each preset multimedia resource for the target object. Based on the first behavioral data sample set and the second behavioral data sample set, a first behavioral sample sequence set and a second behavioral sample sequence set are determined. The first behavioral sample sequence set includes multiple first behavioral sample sequences, and the second behavioral sample sequence set includes multiple second behavioral sample sequences. The first behavioral sample sequences include features of the same dimension, and the second behavioral sample sequences include features of different dimensions. Based on the first set of behavioral sample sequences and the second set of behavioral sample sequences, the target feature vectors of the target object for L preset multimedia resources are determined; Based on the target feature vectors of the target object to L preset multimedia resources and the preset multimedia resource set, the predicted interest degree of the target object to each preset multimedia resource is obtained through the interest degree prediction model to be trained. The interest prediction model is trained based on the real tag set and the predicted interest of the target object for each preset multimedia resource to obtain the interest prediction model.

7. The method according to claim 6, characterized in that, The step of training the interest prediction model based on the real tag set and the predicted interest of the target object for each preset multimedia resource to obtain the interest prediction model includes: Based on the predicted interest of the target object in each preset multimedia resource, the predicted label corresponding to the target object for each preset multimedia resource is determined; Based on the real tag set and the predicted tag corresponding to each preset multimedia resource for the target object, the model parameters of the interest prediction model to be trained are updated according to the target loss function to obtain the interest prediction model.

8. The method according to claim 1, characterized in that, The step of sequentially arranging the predicted interest of the target object in each preset multimedia resource, and determining the target multimedia resource recommended to the target object based on the sequential arrangement result, includes: The predicted interest of the target object in each preset multimedia resource is ordered using the max-heap sort method. The order is determined to be the predicted interest level of each preset multimedia resource from high to low. Based on the ordered results, the target interest level is extracted sequentially from high to low. The preset multimedia resource corresponding to the target interest level is determined as the target multimedia resource.

9. A multimedia resource recommendation device, characterized in that, The multimedia resource recommendation device includes: The acquisition module is used to acquire the first behavior data of the target object in a first time period and the second behavior data in a second time period, wherein the length of the first time period is greater than the length of the second time period, and the first time period includes the second time period. The acquisition module is further configured to perform information extraction processing on the first behavior data to obtain first access behaviors corresponding to N dimensions within the first time period, wherein the first access behaviors are generated when the target object accesses the preset multimedia resources; based on the first access behaviors corresponding to N dimensions within the first time period, generate N first behavior sequences; the first behavior sequences include features of the same dimension within the first time period; perform information extraction processing on the second behavior data to obtain M second access behaviors within the second time period, wherein the second access behaviors are generated when the target object accesses the preset multimedia resources; based on the M second access behaviors within the second time period, generate M second behavior sequences, wherein the second behavior sequences include features of different dimensions within the second time period, where M and N are both positive integers greater than 1, and the dimension is the dimension corresponding to the preset multimedia resource indicated by the user's access behavior; The determining module is used to determine the target feature vector of the target object for L preset multimedia resources based on the N first action sequences and the M second action sequences, where L is a positive integer greater than 1; The acquisition module is further configured to obtain the predicted interest degree of the target object on each preset multimedia resource based on the target feature vector of the target object on L preset multimedia resources through an interest degree prediction model. The determining module is further configured to sequentially arrange the predicted interest of the target object in each preset multimedia resource, and determine the target multimedia resource to be recommended to the target object from the L preset multimedia resources based on the sequential arrangement result, wherein the target multimedia resource includes at least one preset multimedia resource.

10. The apparatus according to claim 9, characterized in that, The determining module is specifically used to determine the first feature vector of the target object for the L preset multimedia resources based on the N first action sequences; The second feature vector of the target object for the L preset multimedia resources is determined based on the M second behavior sequences; The first feature vector and the second feature vector are fused together to obtain the target feature vector.

11. The apparatus according to claim 10, characterized in that, The device also includes a processing module; The acquisition module is also used to acquire the basic attribute information of the target object; The processing module is used to perform feature processing on the basic attribute information of the target object to obtain the user profile features of the target object; The determining module is specifically used to determine the first feature vector based on the N first behavior sequences and the user profile features of the target object; The determining module is specifically used to determine the second feature vector based on the M second behavior sequences and the user profile features of the target object; The determining module is specifically used to fuse the first feature vector, the second feature vector, and the user profile features of the target object to obtain the target feature vector.

12. The apparatus according to claim 11, characterized in that, The acquisition module is further configured to acquire a multimedia resource feature set, wherein the multimedia resource feature set includes F multimedia resource feature subsets, each of which includes features of the same dimension of the L preset multimedia resources, and different multimedia resource feature subsets include features of different dimensions, where F is a positive integer greater than 1 and F is greater than or equal to N. The determining module is specifically used to determine the first feature vector based on the N first behavior sequences, the user profile features of the target object, and the multimedia resource feature set; The determining module is specifically used to determine the second feature vector based on the M second behavior sequences, the user profile features of the target object, and the multimedia resource feature set; The acquisition module is specifically used to obtain the predicted interest of the target object in each preset multimedia resource based on the target feature vector and the multimedia resource feature set, through the interest prediction model.

13. A computer device, characterized in that, include: Memory, transceiver, processor, and bus system; The memory is used to store programs; The processor is used to execute the program in the memory to implement the method of any one of claims 1 to 8; The bus system is used to connect the memory and the processor to enable communication between the memory and the processor.

14. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method as claimed in any one of claims 1 to 8.

15. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images