Media content recall method and apparatus, storage medium, and electronic device
By classifying media asset content features and processing information masking, a media asset content feature vector is generated, which solves the shortcomings of cold start recall in media asset content recommendation services and achieves effective recall of missing and newly generated media asset content.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN TCL NEW-TECH CO LTD
- Filing Date
- 2022-08-05
- Publication Date
- 2026-05-12
AI Technical Summary
In existing media asset recommendation services, cold start recall cannot effectively recall missing or newly generated media asset content, resulting in the inability to effectively recall media asset content lacking relevance.
By classifying content based on pre-collected media asset content features, obtaining browsing trajectory sequences and performing information masking encoding, and generating media asset content feature vectors, cold start recall of missing and newly generated media asset content can be achieved.
It enables personalized and relevant characterization of media asset content, effectively linking missing and newly generated media asset content, and improving the accuracy of cold start recall.
Smart Images
Figure CN117150110B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of Internet technology, specifically to a method, apparatus, storage medium, and electronic device for retrieving media content. Background Technology
[0002] Conventional internet search typically includes media asset recommendation services. These services often involve cold start recall of media asset content. Currently, conventional cold start recall usually involves directly returning trending media asset content to users based on statistics. However, conventional cold start recall is often ineffective in recalling missing media asset content, and such content is usually discarded without further recall. Even with newly generated media asset content, relevant but missing content may not be effectively recalled.
[0003] Therefore, currently, there is a problem that cold start recall cannot be effectively carried out for both missing media assets and newly generated media assets. Summary of the Invention
[0004] This application provides a solution that can effectively achieve cold start recall of media asset content with missing content and newly generated media asset content.
[0005] The embodiments of this application provide the following technical solutions:
[0006] According to one embodiment of this application, a media asset content recall method includes: classifying content based on predetermined features of media asset content in a pre-collected media asset content set to obtain category information for each media asset content; obtaining at least one browsing trajectory sequence of media asset content in the media asset content set; combining each browsing trajectory sequence with associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; performing information masking encoding on the extended trajectory sequence to obtain a media asset content feature vector for each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0007] In some embodiments of this application, the step of classifying content based on predetermined features of media asset content in a pre-collected media asset content set to obtain category information for each media asset content includes: performing clustering processing based on predetermined features of media asset content in the media asset content set to obtain multiple media asset content clusters, each media asset content cluster corresponding to a category information; and obtaining category information for each media asset content according to the media asset content cluster to which each media asset content belongs.
[0008] In some embodiments of this application, before classifying content based on predetermined features of media asset content in a pre-collected media asset content set, the method further includes: obtaining source information and media asset content description information for each media asset content in the media asset content set; and using a combination of the source information and media asset content description information for each media asset content as a predetermined feature for each media asset content in the media asset content set.
[0009] In some embodiments of this application, the source information is source channel information, and the media asset content description information includes preset media asset content tags or image representation information.
[0010] In some embodiments of this application, the browsing trajectory sequence is an identifier of at least one media asset content linked together in the browsing order; the step of combining each of the browsing trajectory sequences with associated category information to obtain at least one extended trajectory sequence includes: inserting the category information of the media asset content corresponding to the identifier before each identifier in each of the browsing trajectory sequences to obtain an extended trajectory sequence corresponding to each of the browsing trajectory sequences.
[0011] In some embodiments of this application, after performing information masking modeling on the expanded trajectory sequence to obtain the media asset content feature vectors of each media asset content, the method further includes: classifying media asset content based on the media asset content-related features of the newly recalled media asset content to obtain category information of the newly recalled media asset content, wherein the newly recalled media asset content includes missing media asset content and newly generated media asset content; generating a sequence of newly recalled media asset content based on the category information and the identifier of the newly recalled media asset content; performing information masking encoding on the sequence of newly recalled media asset content to obtain the media asset content-related feature vectors of the newly recalled media asset content; calculating the similarity between the media asset content-related feature vectors and the media asset content feature vectors of the media asset content in the media asset content set; and performing cold start recall of the newly recalled media asset content based on the similarity.
[0012] In some embodiments of this application, the step of performing information masking encoding on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content includes: performing masking training on the at least one extended trajectory sequence using a pre-trained language model to obtain the category description vector corresponding to the category information of each media asset content; and using the category description vector as the media asset content feature vector.
[0013] According to one embodiment of this application, a media asset content recall device includes: a classification module, configured to classify content based on predetermined features of media asset content in a pre-collected media asset content set, to obtain category information for each media asset content; an acquisition module, configured to acquire at least one browsing trajectory sequence of media asset content in the media asset content set; a combination module, configured to combine each browsing trajectory sequence with associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; and an encoding module, configured to perform information mask encoding processing on the extended trajectory sequence to obtain a media asset content feature vector for each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0014] According to another embodiment of this application, a storage medium stores a computer program thereon, which, when executed by a computer's processor, causes the computer to perform the methods described in the embodiments of this application.
[0015] According to another embodiment of this application, an electronic device may include: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the methods described in the embodiments of this application.
[0016] In this embodiment, content is classified based on predetermined features of media assets in a pre-collected media asset content set to obtain category information for each media asset content; at least one browsing trajectory sequence of the media asset content in the media asset content set is obtained; each browsing trajectory sequence is combined with associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; the extended trajectory sequence is subjected to information masking encoding to obtain a media asset content feature vector for each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0017] In this way, media asset content can be represented in a personalized and relevant manner, resulting in a media asset content feature vector for each content. Based on these media asset content feature vectors, missing media asset content and newly generated media asset content can be effectively associated, thereby achieving effective cold start recall for both missing and newly generated media asset content. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 A flowchart of a media asset content recall method according to an embodiment of this application is shown.
[0020] Figure 2 A schematic diagram illustrating the combination of sequence and category information according to an embodiment of this application is shown.
[0021] Figure 3 A schematic diagram of information mask encoding according to an embodiment of this application is shown.
[0022] Figure 4 A block diagram of a media asset content recall device according to an embodiment of this application is shown.
[0023] Figure 5 A block diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation
[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0025] Figure 1 A flowchart illustrating a media asset content retrieval method according to an embodiment of this application is shown. The entity executing this media asset content retrieval method can be any device, such as a server or terminal.
[0026] like Figure 1 As shown, the media asset content recall method may include steps S110 to S140.
[0027] Step S110: Classify the content based on the predetermined features of the media asset content in the pre-collected media asset content set to obtain the category information of each media asset content; Step S120: Obtain at least one browsing trajectory sequence of the media asset content in the media asset content set; Step S130: Combine each browsing trajectory sequence with the associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; Step S140: Perform information masking encoding on the extended trajectory sequence to obtain the media asset content feature vector of each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0028] A media asset content set can be a collection of media asset content from a specific platform. This content can include articles, videos, and other media assets. Predefined features can be related characteristics of the media asset content; for example, predefined features may include source information and description information of the media asset content.
[0029] By classifying content using predetermined features of each media asset in a media asset content set, category information for each media asset can be obtained. One approach is to cluster the media asset content using predetermined features to obtain category information for each content. Alternatively, another approach is to input the predetermined features into a pre-trained classification model for classification processing to obtain category information for each media asset.
[0030] Based on users' browsing behavior of media assets within a media asset content set, a browsing trajectory sequence for each user can be obtained. Furthermore, for each media asset within the set, at least one browsing trajectory sequence can be obtained based on the browsing behavior of at least one user. In one approach, the browsing trajectory sequence can be obtained by concatenating the identifiers (i.e., media asset content IDs) of the media assets browsed by the user in the order they were viewed. Optionally, in another approach, the browsing trajectory sequence can be obtained by concatenating the identifiers (i.e., media asset content IDs) of the media assets browsed by the user according to predetermined weights. Each browsing trajectory sequence can be a browsing trajectory sequence within a truncated predetermined time period (e.g., one day).
[0031] Each browsing trajectory sequence is combined with its associated category information to obtain at least one extended trajectory sequence. The associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence. For example, if one browsing trajectory sequence is A and another is B, A can be combined with its corresponding associated category information A1, and B can be combined with its corresponding associated category information B1, thus obtaining the extended trajectory sequence corresponding to each browsing trajectory sequence. In one approach, the category information of the media asset content corresponding to each identifier can be inserted before each identifier in each browsing trajectory sequence to obtain the extended trajectory sequence corresponding to each browsing trajectory sequence. Alternatively, in another approach, each browsing trajectory sequence can be directly concatenated with its associated category information to obtain the extended trajectory sequence corresponding to each browsing trajectory sequence.
[0032] Finally, the expanded trajectory sequence undergoes information masking encoding. This involves masking each media asset separately and then encoding it along with the information corresponding to other unmasked media assets in the sequence. This yields the media asset feature vectors of the masked media assets, resulting in the media asset feature vectors for each individual media asset. These feature vectors can represent the media assets in a personalized and relevant way. Based on these feature vectors, effective cold-start recall can be performed for both missing and newly generated media assets.
[0033] In this way, based on steps S110 to S140, the media asset content in the media asset content set can be represented in a personalized and relevant manner, and a media asset content feature vector corresponding to each media asset content can be obtained. Based on these media asset content feature vectors, missing media asset content and newly generated media asset content can be effectively associated, thereby achieving effective cold start recall for missing media asset content and newly generated media asset content.
[0034] The following description Figure 1 Further specific optional embodiments of each step performed when recalling media asset content are described in the examples.
[0035] In one embodiment, step S110, which involves classifying content based on predetermined features of media asset content in a pre-collected media asset content set to obtain category information for each media asset content, includes: performing clustering processing based on predetermined features of media asset content in the media asset content set to obtain multiple media asset content clusters, each media asset content cluster corresponding to a category information; and obtaining category information for each media asset content according to the media asset content cluster to which each media asset content belongs.
[0036] In this embodiment, the media asset content in the media asset content set is clustered based on predetermined features to obtain multiple media asset content clusters. Each media asset content cluster includes at least one media asset content, and each cluster corresponds to a category information. The category information of each media asset content can be obtained based on the media asset content cluster it belongs to. Clustering allows for classification based on the correlation between media asset content. Compared to other methods, the category information obtained through this method can be used in subsequent steps to obtain media asset content feature vectors with stronger correlation performance. Clustering can be performed based on predetermined features of the media asset content in the media asset content set, using algorithms such as K-means clustering, hierarchical clustering, and DBSCAN.
[0037] In one embodiment, before performing content classification based on predetermined features of media asset content in a pre-collected media asset content set in step S110, the method further includes: obtaining source information and media asset content description information for each media asset content in the media asset content set; and using the combination of source information and media asset content description information for each media asset content as predetermined features for each media asset content in the media asset content set.
[0038] Source information refers to the source of the media asset content, such as the source name or code. Examples of sources include channel information, platform, or region. Media asset content description information describes the content itself, such as predefined tags, image representations of images within the content, a summary, and the size of the content. By combining source and description information as predefined features, media asset content can be effectively classified, particularly based on clustering to achieve effective classification based on content relevance.
[0039] In one embodiment, the source information is source channel information, and the media asset content description information includes preset media asset content tags or image representation information. The preset media asset content tags are tags pre-defined by relevant users for each media asset content. Each media asset content can have multiple tags, and the preset media asset content tags can be a predetermined number (e.g., 3) of tags selected from these multiple tags. The image representation information can be the color information of an image. In one example, the image representation information is the three HSV color representations of the image; in another example, the image representation information is the HSV color representation of the image's segmentation range expanded to 6 categories. The source channel information can be channel information such as the television channel or program channel corresponding to the media asset content.
[0040] In one embodiment, the browsing trajectory sequence is an identifier of at least one media asset content linked together in browsing order; the step of combining each of the browsing trajectory sequences with associated category information to obtain at least one extended trajectory sequence includes: inserting the category information of the media asset content corresponding to each identifier before each identifier in each of the browsing trajectory sequences to obtain an extended trajectory sequence corresponding to each of the browsing trajectory sequences.
[0041] See Figure 2 A browsing trajectory sequence includes identifiers id1, id2, and id3 of at least one media asset content linked together in the browsing order. Before each identifier (id1, id2, and id3) in the browsing trajectory sequence, the category information of the corresponding media asset content (cluster K1 of id1, cluster K2 of id2, and cluster K3 of id3) is inserted to obtain an expanded trajectory sequence: cluster K1 of id1, id1, cluster K2 of id2, id2, cluster K3 of id3, and id3. This process is repeated for each browsing trajectory sequence. Compared to other methods, the expanded trajectory sequences obtained in this way can be used in subsequent steps to generate better media asset content feature vectors through information masking encoding.
[0042] In one embodiment, after performing information masking modeling on the expanded trajectory sequence to obtain the media asset content feature vectors of each media asset content, the method further includes: classifying media asset content based on the media asset content-related features of the newly recalled media asset content to obtain category information of the newly recalled media asset content, wherein the newly recalled media asset content includes missing media asset content and newly generated media asset content; generating a sequence of newly recalled media asset content based on the category information and the identifier of the newly recalled media asset content; performing information masking encoding on the sequence of newly recalled media asset content to obtain the media asset content-related feature vectors of the newly recalled media asset content; calculating the similarity between the media asset content-related feature vectors and the media asset content feature vectors of the media asset content in the media asset content set; and performing cold start recall of the newly recalled media asset content based on the similarity.
[0043] Newly recalled media assets refer to media assets not present in the existing media asset set. These newly recalled media assets can include both missing and newly generated content. Media asset classification is performed based on the content-related features of these newly recalled media assets. One approach is to use these features to cluster the media assets, thus obtaining category information for each newly recalled media asset. Alternatively, another approach is to input the content-related features into a pre-trained classification model for classification processing, thereby obtaining category information for each newly recalled media asset.
[0044] Combining the category information and identifier of newly recalled media assets can generate a sequence of newly recalled media assets. One method involves inserting the category information of the newly recalled media assets before the identifier in each asset and then concatenating them to obtain the sequence. Alternatively, another method involves directly concatenating the sequence of category information and the identifier of the newly recalled media assets to obtain the same sequence. The length of each newly recalled media asset sequence can be specified according to the actual situation.
[0045] Then, the newly added recalled media asset content sequence can be processed by information masking encoding to obtain the media asset content-related feature vector of the newly added recalled media asset content. By using similarity calculation methods such as Euclidean distance, the similarity between the media asset content-related feature vector and the media asset content feature vector of the media asset content in the media asset content set can be calculated.
[0046] Based on similarity, newly added media assets can be cold-started and recalled. For example, by statistically analyzing media asset content hotspots, trending media assets under these hotspots can be pushed to users. Trending media assets can include similar newly added media assets and media asset content clusters.
[0047] In one embodiment, the step of performing information masking encoding on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content includes: performing masking training on the at least one extended trajectory sequence using a pre-trained language model to obtain the category description vector corresponding to the category information of each media asset content; and using the category description vector as the media asset content feature vector.
[0048] Using a pre-trained language model (BERT) for mask training (MLM mask training) yields category description vectors corresponding to the category information of each media asset. These category description vectors exhibit good personalization and relevance. See also... Figure 3 For the extended trajectory sequences: clusters K1, K2, K3, and K3 of id1 and id2, when using a pre-trained language model (BERT) for mask training (MLM mask training), after masking id1, the identifier feature vector of id1 can be obtained based on the sequence information in the sequence (MLM mask training). After masking id3, the category description vector of cluster K3 of id3 can be obtained based on the sequence information in the sequence (MLM mask training).
[0049] By using the category description vector corresponding to the category information of media asset content as the media asset content feature vector, the applicant found that it is possible to carry out extremely effective cold start recall for media asset content with missing content and newly generated media asset content.
[0050] In one embodiment, the step of performing information masking encoding on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content includes:
[0051] Determine the extended trajectory sequence corresponding to each media asset content; hide the media asset content node information of each media asset content in the corresponding extended trajectory sequence to obtain the hidden trajectory sequence corresponding to each media asset content; perform feature extraction processing on the hidden trajectory sequence corresponding to each media asset content to obtain the media asset content feature vector of each media asset content.
[0052] Based on the expanded trajectory sequence containing the identifier of each media asset, the corresponding expanded trajectory sequence can be determined. The media asset node information of each media asset in the corresponding expanded trajectory sequence is hidden to obtain a hidden trajectory sequence for each media asset. For example, for media asset id1, in the expanded trajectory sequences: clusters K1 and id1 of id1, clusters K2 and K3 of id2, and clusters K3 and id3 of id3, clusters K1 and id1 of id1 are hidden to obtain the hidden trajectory sequences K2, K2, K3, and id3 of id2 and id3 of id1. Finally, feature extraction processing can be performed on the hidden trajectory sequences using a feature extraction network to obtain the media asset feature vector for each media asset.
[0053] To facilitate better implementation of the media asset content retrieval method provided in this application, this application also provides a media asset content retrieval device based on the above-described media asset content retrieval method. The meanings of the terms used are the same as in the above-described media asset content retrieval method, and specific implementation details can be found in the descriptions within the method embodiments. Figure 4 A block diagram of a media asset content recall device according to an embodiment of this application is shown.
[0054] like Figure 4 As shown, the media asset content retrieval device 200 may include a classification module 210, an acquisition module 220, a combination module 230, and an encoding module 240.
[0055] The system includes a classification module for classifying content based on predetermined features of media assets in a pre-collected media asset content set, obtaining category information for each media asset content; an acquisition module for acquiring at least one browsing trajectory sequence of media assets in the media asset content set; a combination module for combining each browsing trajectory sequence with associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; and an encoding module for performing information mask encoding on the extended trajectory sequences to obtain a media asset content feature vector for each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0056] In some embodiments of this application, the classification module is used to: perform clustering processing based on predetermined features of media asset content in the media asset content set to obtain multiple media asset content clusters, each media asset content cluster corresponding to a category information; and obtain the category information of each media asset content according to the media asset content cluster to which each media asset content belongs.
[0057] In some embodiments of this application, the device further includes a collection module, configured to: acquire source information and media content description information for each media content in the media content set; and combine the source information and media content description information of each media content as a predetermined feature of each media content in the media content set.
[0058] In some embodiments of this application, the source information is source channel information, and the media asset content description information includes preset media asset content tags or image representation information.
[0059] In some embodiments of this application, the browsing trajectory sequence is an identifier of at least one media asset content linked together in the browsing order; the combination module is used to: insert the category information of the media asset content corresponding to the identifier before each identifier in each of the browsing trajectory sequences to obtain an extended trajectory sequence corresponding to each of the browsing trajectory sequences.
[0060] In some embodiments of this application, the apparatus further includes a recall module, configured to: classify media asset content based on media asset content-related features of newly recalled media asset content to obtain category information of the newly recalled media asset content, wherein the newly recalled media asset content includes missing media asset content and newly generated media asset content; generate a sequence of newly recalled media asset content based on the category information and the identifier of the newly recalled media asset content; perform information masking encoding on the sequence of newly recalled media asset content to obtain a media asset content-related feature vector of the newly recalled media asset content; calculate the similarity between the media asset content-related feature vector and the media asset content feature vector of the media asset content in the media asset content set; and perform cold start recall of the newly recalled media asset content based on the similarity.
[0061] In some embodiments of this application, the encoding module is configured to: perform mask training on the at least one extended trajectory sequence using a pre-trained language model to obtain a category description vector corresponding to the category information of each media asset content; and use the category description vector as the feature vector of the media asset content.
[0062] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.
[0063] Furthermore, embodiments of this application also provide an electronic device, which can be a terminal or a server, such as... Figure 5 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:
[0064] The electronic device may include components such as a processor 301 with one or more processing cores, a memory 302 with one or more computer-readable storage media, a power supply 303, and an input unit 304. Those skilled in the art will understand that... Figure 5 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:
[0065] The processor 301 is the control center of the electronic device. It connects to various parts of the computer device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 302, and by calling data stored in the memory 302, it performs various functions of the computer device and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user page, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.
[0066] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and data processing by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs 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 computer device, etc. In addition, the memory 302 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. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.
[0067] The electronic device also includes a power supply 303 that supplies power to various components. Preferably, the power supply 303 can be logically connected to the processor 301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 303 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.
[0068] The electronic device may also include an input unit 304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
[0069] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more computer programs into the memory 302 according to the following instructions, and the processor 301 runs the computer programs stored in the memory 302, thereby realizing the various functions in the foregoing embodiments of this application. For example, the processor 301 can perform the following steps:
[0070] Based on predetermined features of media assets in a pre-collected media asset content set, content is classified to obtain category information for each media asset content; at least one browsing trajectory sequence of the media asset content in the media asset content set is obtained; each browsing trajectory sequence is combined with associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; the extended trajectory sequence is subjected to information masking encoding to obtain a media asset content feature vector for each media asset content, wherein the media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content.
[0071] In some embodiments of this application, the step of classifying content based on predetermined features of media asset content in a pre-collected media asset content set to obtain category information for each media asset content includes: performing clustering processing based on predetermined features of media asset content in the media asset content set to obtain multiple media asset content clusters, each media asset content cluster corresponding to a category information; and obtaining category information for each media asset content according to the media asset content cluster to which each media asset content belongs.
[0072] In some embodiments of this application, before classifying content based on predetermined features of media asset content in a pre-collected media asset content set, the method further includes: obtaining source information and media asset content description information for each media asset content in the media asset content set; and using a combination of the source information and media asset content description information for each media asset content as a predetermined feature for each media asset content in the media asset content set.
[0073] In some embodiments of this application, the source information is source channel information, and the media asset content description information includes preset media asset content tags or image representation information.
[0074] In some embodiments of this application, the browsing trajectory sequence is an identifier of at least one media asset content linked together in the browsing order; the step of combining each of the browsing trajectory sequences with associated category information to obtain at least one extended trajectory sequence includes: inserting the category information of the media asset content corresponding to the identifier before each identifier in each of the browsing trajectory sequences to obtain an extended trajectory sequence corresponding to each of the browsing trajectory sequences.
[0075] In some embodiments of this application, after performing information masking modeling on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content, the method further includes: classifying media asset content based on the media asset content-related features of the newly recalled media asset content to obtain category information of the newly recalled media asset content, wherein the newly recalled media asset content includes missing media asset content and newly generated media asset content; generating a sequence of newly recalled media asset content based on the category information and the identifier of the newly recalled media asset content; performing information masking encoding on the sequence of newly recalled media asset content to obtain the media asset content-related feature vector of the newly recalled media asset content; calculating the similarity between the media asset content-related feature vector and the media asset content feature vector of the media asset content in the media asset content set; and performing cold start recall of the newly recalled media asset content based on the similarity.
[0076] In some embodiments of this application, the step of performing information masking encoding on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content includes: performing masking training on the at least one extended trajectory sequence using a pre-trained language model to obtain the category description vector corresponding to the category information of each media asset content; and using the category description vector as the media asset content feature vector.
[0077] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by a computer program, or by a computer program controlling related hardware. The computer program can be stored in a computer-readable storage medium and loaded and executed by a processor.
[0078] Therefore, embodiments of this application also provide a storage medium storing a computer program that can be loaded by a processor to execute the steps in any of the methods provided in embodiments of this application.
[0079] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.
[0080] Since the computer program stored in the storage medium can execute the steps of any of the methods provided in the embodiments of this application, the beneficial effects that the methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.
[0081] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.
[0082] It should be understood that this application is not limited to the embodiments described above and shown in the accompanying drawings, but various modifications and changes can be made without departing from its scope.
Claims
1. A method for recalling media asset content, characterized in that, include: Based on the predetermined features of the media asset content collected in advance, the content is classified to obtain the category information of each media asset content. The predetermined features are the related media asset content features of the media asset content. Obtain at least one browsing trajectory sequence of the media assets in the media asset content set; Each of the browsing trajectory sequences is combined with the associated category information to obtain at least one extended trajectory sequence, wherein the associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence; The expanded trajectory sequence is processed by information masking encoding to obtain the media asset content feature vector of each media asset content. The media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content. The browsing trajectory sequence is an identifier of at least one media asset content linked together in the browsing order; The step of combining each of the browsing trajectory sequences with the associated category information to obtain at least one extended trajectory sequence includes: inserting the category information of the media asset content corresponding to each identifier before each identifier in each of the browsing trajectory sequences to obtain the extended trajectory sequence corresponding to each of the browsing trajectory sequences; After performing information masking encoding on the extended trajectory sequence to obtain the media asset content feature vectors of each media asset content, the method further includes: Media asset content is classified based on its content-related features to obtain category information. This category includes both missing and newly generated media asset content. A sequence of newly recalled media asset content is generated based on its category information and identifier. The sequence is then subjected to information masking encoding to obtain a content-related feature vector. The similarity between this feature vector and the content-related feature vectors of other media asset content in the set is calculated. Finally, a cold-start recall is performed on the newly recalled media asset content based on this similarity.
2. The method according to claim 1, characterized in that, The content classification based on predetermined characteristics of media assets in a pre-collected media asset content set yields category information for each media asset, including: Clustering is performed on the media asset content based on the predetermined features of the media asset content in the media asset content set to obtain multiple media asset content clusters, and each media asset content cluster corresponds to a category information; Based on the media asset content cluster to which each media asset content belongs, the category information of each media asset content is obtained.
3. The method according to claim 1, characterized in that, Before classifying content based on predetermined features of media assets in a pre-collected media asset content set, the method further includes: Obtain the source information and description information of each media asset in the media asset content set; The combination of source information and description information of each media asset content is used as a predetermined feature of each media asset content in the media asset content set.
4. The method according to claim 3, characterized in that, The source information is the source channel information, and the media asset content description information includes preset media asset content tags or image representation information.
5. The method according to claim 1, characterized in that, The step of performing information masking encoding on the at least one extended trajectory sequence to obtain the media asset content feature vector of each media asset content includes: For the at least one extended trajectory sequence, a pre-trained language model is used for mask training to obtain the category description vector corresponding to the category information of each media asset content; The category description vector is used as the media asset content feature vector.
6. A media asset content recall device, characterized in that, include: The classification module is used to classify content based on predetermined features of media asset content in a pre-collected media asset content set, and to obtain category information for each media asset content. The predetermined features are the related media asset content features of the media asset content. The acquisition module is used to acquire at least one browsing trajectory sequence of media asset content in the media asset content set; The combination module is used to combine each of the browsing trajectory sequences with the associated category information to obtain at least one extended trajectory sequence. The associated category information is the category information of the media asset content corresponding to each browsing trajectory sequence. The browsing trajectory sequence is the identifier of at least one media asset content linked together in the browsing order. The step of combining each of the browsing trajectory sequences with the associated category information to obtain at least one extended trajectory sequence includes: inserting the category information of the media asset content corresponding to each identifier before each identifier in each of the browsing trajectory sequences to obtain the extended trajectory sequence corresponding to each of the browsing trajectory sequences; The encoding module is used to perform information mask encoding on the extended trajectory sequence to obtain the media asset content feature vector of each media asset content. The media asset content feature vector is used for cold start recall of missing media asset content and newly generated media asset content. The recall module is used to classify media assets based on their content-related features to obtain category information, including missing and newly generated media assets; generate a sequence of newly recalled media assets based on their category information and identifiers; perform information masking encoding on the sequence to obtain content-related feature vectors; calculate the similarity between the content-related feature vectors and the content-related feature vectors of other media assets in the content set; and perform cold-start recall of the newly recalled media assets based on the similarity.
7. A storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the method described in any one of claims 1 to 5.
8. An electronic device, characterized in that, include: Memory, which stores computer programs; A processor reads a computer program stored in memory to perform the method described in any one of claims 1 to 5.