Content processing methods and related equipment

Abstract

This application discloses a content processing method and related equipment. The relevant embodiments can be applied to various scenarios such as cloud technology, artificial intelligence, and smart transportation. It can acquire object attribute information of a target object, at least one historical interaction content, and interaction collaboration data. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed, whereby the interaction collaboration relationship graph represents the interaction relationship between the object and the content. Information transmission processing is performed on the object attribute information and the historical interaction content based on the interaction collaboration relationship graph to obtain interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, recommended content is selected from candidate content. This application can deeply mine the interaction relationship between the object and the content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a content processing method and related equipment. Background Technology

[0002] With the rapid development of internet technology, the amount of content on the internet has exploded. Filtering out content that users are interested in from this massive amount of content and recommending it to users has become increasingly important.

[0003] Current technologies typically rely on users' browsing history for personalized content recommendations. This history, containing previously viewed content, can, to some extent, indicate a user's browsing interests. However, current recommendation methods fail to consider the deeper interaction between users and content, resulting in lower accuracy. Summary of the Invention

[0004] This application provides a content processing method and related equipment. The related equipment may include a content processing device, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the accuracy of content recommendations.

[0005] This application provides a content processing method, including:

[0006] Obtain object attribute information of the target object, at least one historical interaction content, and interaction collaboration data, wherein the interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects;

[0007] Based on the interactive collaboration data, an interactive collaboration relationship graph corresponding to the target object is constructed. The interactive collaboration relationship graph represents the interactive relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interactive content corresponding to the target object and the historical interactive content corresponding to the associated objects.

[0008] Based on the interaction and collaboration relationship diagram, information transmission processing is performed on the object attribute information and the historical interaction content to obtain the interaction and collaboration feature information corresponding to the target object.

[0009] Based on the interactive collaboration feature information corresponding to the target object, content to be recommended is selected from the candidate content.

[0010] Accordingly, embodiments of this application provide a content processing apparatus, including:

[0011] The acquisition unit is used to acquire object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects.

[0012] The construction unit is used to construct an interaction collaboration relationship diagram corresponding to the target object based on the interaction collaboration data. The interaction collaboration relationship diagram represents the interaction relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects.

[0013] The information transmission unit is used to perform information transmission processing on the object attribute information and the historical interaction content according to the interaction collaboration relationship diagram, so as to obtain the interaction collaboration feature information corresponding to the target object.

[0014] The selection unit is used to select content to be recommended from candidate content based on the interactive collaboration feature information corresponding to the target object.

[0015] Optionally, in some embodiments of this application, the interaction collaboration relationship graph includes at least one node and connections between nodes. The node includes an attribute node corresponding to the object and a content node corresponding to the content. The connections represent the interaction relationship between the object and the content. Specifically, the information transmission unit can be used to transmit information through the connections between nodes in the interaction collaboration relationship graph to process the object attribute information and the historical interaction content, thereby obtaining the interaction collaboration feature information corresponding to the target object.

[0016] Optionally, in some embodiments of this application, the information transmission unit may include an extraction subunit, a node search subunit, and a transmission subunit, as follows:

[0017] The extraction subunit is used to extract features from each node in the interaction and collaboration relationship graph to obtain node feature information corresponding to each node. The node feature information includes node feature information of object attribute information corresponding to each attribute node and node feature information of historical interaction content corresponding to each content node.

[0018] The node search subunit is used to perform a node search in the interaction and collaboration relationship graph for each node, in order to determine the neighboring nodes of the node.

[0019] The transmission subunit is used to process the node feature information of the node according to the node feature information of the neighboring node to obtain the interaction and cooperation feature information corresponding to the target object.

[0020] Optionally, in some embodiments of this application, the transmission subunit may be specifically used to aggregate the node feature information of the node and the node feature information of the neighboring nodes to obtain the neighbor aggregated feature information of each node in the interaction collaboration relationship graph; update the node feature information of each node according to the neighbor aggregated feature information of each node to obtain the updated interaction collaboration relationship graph; and extract the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph.

[0021] Optionally, in some embodiments of this application, the step "updating the node feature information of each node based on the neighbor aggregation feature information of each node to obtain an updated interaction and cooperation relationship graph" may include:

[0022] The node feature information of each node is updated based on the neighbor aggregation feature information of each node.

[0023] Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0024] Optionally, in some embodiments of this application, the step of "extracting the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph" may include:

[0025] The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node.

[0026] The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

[0027] Optionally, in some embodiments of this application, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0028] For each content node in the interaction collaboration relationship graph, determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence. The target historical content sequence includes the interaction content of the target interaction object sorted in the order of interaction time within a historical time period. The target interaction object is the object that has an interaction relationship with the historical interaction content.

[0029] Based on the location information, the feature information of the target node corresponding to the content node is updated;

[0030] The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

[0031] Optionally, in some embodiments of this application, the selection unit may include a feature extraction subunit, a fusion subunit, and a selection subunit, as follows:

[0032] The feature extraction subunit is used to extract features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object.

[0033] The fusion subunit is used to fuse the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object;

[0034] The selection sub-unit is used to select content to be recommended from the candidate content based on the object's interest feature information.

[0035] Optionally, in some embodiments of this application, the fusion subunit may be specifically used to perform attention processing on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information; and to fuse the attribute feature information and interaction collaboration feature information corresponding to the target object according to the attention weight information to obtain the object interest feature information of the target object.

[0036] Optionally, in some embodiments of this application, the selection sub-unit may be specifically used to extract features from the object's interest feature information in at least one interest dimension to obtain interest sub-features in the at least one interest dimension; select candidate content to be recommended in each interest dimension from the candidate content based on the interest sub-features in each interest dimension; and perform aggregation processing on the candidate content to be recommended in each interest dimension to obtain the content to be recommended.

[0037] Optionally, in some embodiments of this application, the fusion subunit may be specifically used to obtain at least one reference historical interaction content of the target object in terms of reference content type; perform attention processing on the at least one reference historical interaction content to obtain reference content feature information corresponding to the target object; and fuse the attribute feature information, interaction collaboration feature information, and reference content feature information corresponding to the target object to obtain object interest feature information of the target object.

[0038] An electronic device provided in this application includes a processor and a memory. The memory stores multiple instructions, and the processor loads the instructions to execute the steps in the content processing method provided in this application.

[0039] This application also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps in the content processing method provided in this application.

[0040] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps in the content processing method provided in this application.

[0041] This application provides a content processing method and related equipment, which can acquire object attribute information of a target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in content interaction and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration relationship graph, information transmission processing is performed on the object attribute information and the historical interaction content to obtain interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, recommended content is selected from candidate content. This application can deeply mine the interaction relationship between objects and content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation. Attached Figure Description

[0042] 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.

[0043] Figure 1a This is a schematic diagram illustrating a scenario of the content processing method provided in an embodiment of this application;

[0044] Figure 1b This is a flowchart of the content processing method provided in the embodiments of this application;

[0045] Figure 1c This is a model structure diagram of the content processing method provided in the embodiments of this application;

[0046] Figure 1d This is another model structure diagram of the content processing method provided in the embodiments of this application;

[0047] Figure 2 This is another flowchart of the content processing method provided in the embodiments of this application;

[0048] Figure 3 This is a schematic diagram of the structure of the content processing device provided in the embodiments of this application;

[0049] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0050] 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.

[0051] This application provides a content processing method and related equipment. The related equipment may include a content processing apparatus, an electronic device, a computer-readable storage medium, and a computer program product. Specifically, the content processing apparatus may be integrated into an electronic device, which may be a terminal or a server, etc.

[0052] It is understood that the content processing method of this embodiment can be executed on a terminal, on a server, or jointly by a terminal and a server. The above examples should not be construed as limiting this application.

[0053] like Figure 1a As shown, the content processing method is executed jointly by a terminal and a server as an example. The content processing system provided in this application includes a terminal 10 and a server 11, etc.; the terminal 10 and the server 11 are connected via a network, such as a wired or wireless network, etc., wherein the content processing device can be integrated into the server.

[0054] Server 11 can be used to: acquire object attribute information of a target object, at least one historical interaction content, and interaction collaboration data, wherein the interaction collaboration data includes associated objects related to the target object in content interaction and the historical interaction content corresponding to the associated objects; construct an interaction collaboration relationship graph corresponding to the target object based on the interaction collaboration data, wherein the interaction collaboration relationship graph represents the interaction relationship between objects and content, the objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects; perform information transmission processing on the object attribute information and the historical interaction content based on the interaction collaboration relationship graph to obtain interaction collaboration feature information corresponding to the target object; and select content to be recommended from candidate content based on the interaction collaboration feature information corresponding to the target object. Server 11 can be a single server, a server cluster composed of multiple servers, or a cloud server. In the content processing method or apparatus disclosed in this application, multiple servers can form a blockchain, and the server is a node on the blockchain.

[0055] Terminal 10 can be used to receive content to be recommended from server 11 and display the content to be recommended to the target audience on the corresponding recommended content page. Terminal 10 can include mobile phones, smart TVs, tablets, laptops, or personal computers (PCs), etc. A client can also be set on terminal 10, which can be an application client or a browser client, etc.

[0056] The steps of server 11 obtaining the content to be recommended can also be performed by terminal 10.

[0057] The content processing method provided in this application relates to computer vision technology and natural language processing in the field of artificial intelligence.

[0058] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within 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. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. AI software technology mainly includes computer vision, speech processing, natural language processing, as well as machine learning / deep learning, autonomous driving, and intelligent transportation.

[0059] Computer vision (CV) is a science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes for target recognition and measurement, and further processes images to create images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, attempting to build artificial intelligence systems capable of extracting information from images or multidimensional data. Computer vision technologies typically include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping (SLAM), autonomous driving, intelligent transportation, and other technologies, as well as common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0060] Natural Language Processing (NLP) is an important area within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close connection with linguistic research. NLP technologies typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0061] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0062] This embodiment will be described from the perspective of a content processing device, which can be integrated into an electronic device, such as a server or a terminal.

[0063] It is understood that in the specific implementation of this application, user information, such as the user's historical interaction content and other related data, is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0064] The content processing method of this application embodiment can be applied to various scenarios requiring content recommendation, such as video recommendation and text recommendation. This embodiment can be applied to various scenarios such as cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0065] like Figure 1b As shown, the specific process of this content processing method can be as follows:

[0066] 101. Obtain the object attribute information of the target object, at least one historical interaction content, and interaction collaboration data, wherein the interaction collaboration data includes related objects that are related to the target object in terms of content interaction and the historical interaction content corresponding to the related objects.

[0067] The target object is the object to which the recommended content is to be submitted. The content processing method provided in this application allows for content recommendations to the target object. Object attribute information may include at least one dimension of attribute sub-information, such as the object's own attributes, short-term features, and long-term features. This embodiment does not impose any limitations on this. Specifically, short-term features may be features corresponding to historical interaction content that is relatively recent in time. For example, short-term features may be obtained by extracting features from the title information of historical interaction content within the last week. Long-term features may be features corresponding to historical interaction content that is relatively distant in time. For example, long-term features may be obtained by extracting features from the title information of historical interaction content within the last six months.

[0068] Specifically, a historical content sequence of the target object within a historical time period can be obtained. This historical content sequence includes at least one historical interaction content. Each historical interaction content can specifically be content viewed by the target object at a certain time within the historical time period. The viewed content can include information in various modalities such as video, audio, text, and images; this embodiment does not impose any limitations on this. Optionally, the historical interaction content in the historical content sequence can be arranged according to the interaction time order of the target object.

[0069] In this embodiment, for each historical interaction content, the historical interaction content may include at least one dimension of content-related information, or in other words, each historical interaction content may include the content itself, the publisher of the content, the content category, the content title, and the content's identifier, etc.

[0070] Specifically, the interactive collaboration data of the target object may include related objects in terms of content interaction and the historical interaction content corresponding to the related objects.

[0071] For example, the target object can be denoted as user1. The historical interaction content corresponding to the target object includes item1. Besides user1, user2 also has an interaction relationship with item1, meaning user2 has also viewed item1. Thus, user2 can be considered a related object associated with the target object user1. Other historical interactions of user2, such as item2, are historical interaction content corresponding to related objects. Similarly, user3, another object that interacts with item2, can also be considered a related object associated with the target object user1. Furthermore, other historical interactions of user3, such as item3, are also historical interaction content corresponding to related objects. Therefore, the interactive collaboration data of the target object can include user1, user2, user3, item1, item2, and item3. Here, an item (project) can specifically be an article or a video, etc.

[0072] It should be noted that the amount of interactive collaboration data for the target object can be set according to the actual situation. For example, if a deeper exploration of the interaction relationship between objects and content is needed, more interactive collaboration data can be obtained, that is, more related objects and historical interaction content related to the target object can be obtained.

[0073] 102. Based on the interactive collaboration data, construct an interactive collaboration relationship graph corresponding to the target object. The interactive collaboration relationship graph represents the interactive relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interactive content corresponding to the target object and the historical interactive content corresponding to the associated objects.

[0074] Specifically, the interaction and collaboration relationship graph may include at least one node and the connections between nodes. The node may include the attribute node corresponding to the object and the content node corresponding to the content. The connection may represent the interaction relationship between the object and the content.

[0075] Among them, attribute nodes are specifically the nodes corresponding to the object's attribute information. Lines can be used to connect attribute nodes and content nodes, meaning that the node types at both ends of the line are different.

[0076] Optionally, this embodiment can generate attribute nodes corresponding to each object and content corresponding to each content based on the various objects and content in the interactive collaboration data. Then, based on whether the objects corresponding to each attribute node and the content corresponding to each content node have an interactive relationship, the attribute nodes and content nodes are connected to obtain an interactive collaboration relationship diagram. For example, when an object and a piece of content have an interactive relationship, the attribute node corresponding to the object and the content node corresponding to the content can be connected. Specifically, the interactive relationship between the object and the content means that the object has viewed the content within a historical time period.

[0077] In a specific scenario, this embodiment can construct an interaction collaboration relationship graph corresponding to the target object based on the above-mentioned interaction collaboration data. Each node in the interaction collaboration relationship graph can be divided into multiple interaction collaboration relationship layers. The node types in each interaction collaboration relationship layer are consistent, and the node types corresponding to each interaction collaboration relationship layer are alternating. For example, if the node type corresponding to the 0th interaction collaboration relationship layer is an attribute node, then the node type corresponding to the 1st interaction collaboration relationship layer is a content node, the node type corresponding to the 2nd interaction collaboration relationship layer is an attribute node, the node type corresponding to the 3rd interaction collaboration relationship layer is a content node, and so on.

[0078] Specifically, the 0th interaction collaboration layer may include attribute nodes corresponding to the object attribute information of the target object, the 1st interaction collaboration layer may include content nodes corresponding to each historical interaction content of the target object, and the 2nd interaction collaboration layer may include attribute nodes corresponding to other objects that have interaction relationships with the historical interaction content of the target object.

[0079] For example, the object attribute information of the target object is denoted as user1. The historical interaction content corresponding to the target object includes item1, item2 and item3. The object that has an interaction relationship with item1 also includes user2. The object that has an interaction relationship with item3 also includes user3. Furthermore, the associated object user2 also includes historical interaction content item4 and item5. The associated object user3 also includes historical interaction content item4 and item6. Based on the interaction relationships between objects and content, an interaction collaboration relationship diagram corresponding to the target object can be constructed. The 0th interaction collaboration relationship layer (i.e., layer 0) can include the attribute nodes corresponding to the object attribute information of the target object user1. The 1st interaction collaboration relationship layer (i.e., layer 1) can include the content nodes corresponding to the historical interaction content item1, item2, and item3 of the target object. The 2nd interaction collaboration relationship layer (i.e., layer 2) can include the attribute nodes corresponding to the object attribute information of the object user2 that interacts with item1, and the attribute nodes corresponding to the object attribute information of the object user3 that interacts with item3. The 3rd interaction collaboration relationship layer (i.e., layer 3) can include the content nodes corresponding to the historical interaction content item4, item5, and item6.

[0080] 103. Based on the interaction and collaboration relationship diagram, perform information transmission processing on the object attribute information and the historical interaction content to obtain the interaction and collaboration feature information corresponding to the target object.

[0081] Optionally, in this embodiment, the interaction collaboration relationship graph includes at least one node and the connection between nodes. The node includes the attribute node corresponding to the object and the content node corresponding to the content. The connection represents the interaction relationship between the object and the content.

[0082] The step "based on the interaction collaboration relationship graph, perform information transmission processing on the object attribute information and the historical interaction content to obtain the interaction collaboration feature information corresponding to the target object" may include:

[0083] In the interactive collaboration relationship graph, information is transmitted through the connections between nodes to process the object attribute information and the historical interaction content, thereby obtaining the interactive collaboration feature information corresponding to the target object.

[0084] Specifically, this embodiment can explicitly encode and transmit the interaction information of objects and content according to the interaction and collaboration relationship diagram, realize the expression model of high-order connectivity of historical interaction content, obtain the representation of historical content sequence that integrates interaction and collaboration signals, thereby enhancing the representation power of historical content sequence.

[0085] Specifically, the interactive collaboration graph can be a graph neural network, which can be represented by a bipartite graph structure. This application utilizes a graph neural network as the main architecture and integrates the interaction relationships between objects and content into the feature embedding process using NGCF (Neural Graph Collaborative Filtering). NGCF specifically integrates the bipartite graph structure into the feature embedding process. Embedding refers to representing an object with a low-dimensional vector, which could be a word, a product, etc.

[0086] Optionally, in this embodiment, the step "transmitting information through connections between nodes in the interaction collaboration relationship graph to process the object attribute information and the historical interaction content, thereby obtaining the interaction collaboration feature information corresponding to the target object" may include:

[0087] Feature extraction is performed on each node in the interaction and collaboration relationship graph to obtain node feature information corresponding to each node. The node feature information includes node feature information of object attribute information corresponding to each attribute node and node feature information of historical interaction content corresponding to each content node.

[0088] For each node in the interaction and collaboration relationship graph, a node search is performed in the interaction and collaboration relationship graph based on the node to determine the neighboring nodes corresponding to the node;

[0089] Based on the node feature information of the neighboring nodes, information transmission processing is performed on the node feature information of the node to obtain the interaction and collaboration feature information corresponding to the target object.

[0090] The interaction and collaboration relationship graph can include attribute nodes and content nodes. Feature extraction is performed on attribute nodes, specifically by extracting features from the object attribute information of the object corresponding to the attribute node to obtain the node feature information of the attribute node. Feature extraction is also performed on content nodes, specifically by extracting features from the historical interaction content corresponding to the content node to obtain the node feature information of the content node.

[0091] Specifically, for each node in the interaction collaboration graph, a node search is performed within the graph. This search involves finding nodes located at the same interaction collaboration layer as the current node, and these nodes are identified as its neighbors. Specifically, the neighboring nodes have the same node type as the current node.

[0092] Optionally, in this embodiment, the step "extracting features from each node in the interaction and collaboration relationship graph to obtain node feature information corresponding to each node" may include:

[0093] For each attribute node in the interaction and collaboration relationship graph, feature extraction is performed on the object attribute information corresponding to the attribute node to obtain the node feature information of the attribute node;

[0094] For each content node in the interaction collaboration relationship graph, feature extraction is performed on the historical interaction content corresponding to the content node to obtain the initial content feature information corresponding to the content node.

[0095] Determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence. The target historical content sequence includes the interaction content of the target interaction object sorted in the order of interaction time within a historical time period. The target interaction object is an object that has an interaction relationship with the historical interaction content.

[0096] The initial content feature information and the location information are fused to obtain the node feature information corresponding to the content node.

[0097] Specifically, feature extraction of the object attribute information corresponding to the attribute node can be performed by obtaining attribute sub-information of the object in at least one dimension, extracting features from the attribute sub-information in each dimension to obtain sub-feature information in each dimension, and then fusing the sub-feature information in each dimension to obtain the node feature information of the attribute node.

[0098] There are various ways to fuse initial content feature information and location information. This embodiment does not limit this. For example, the fusion method can be splicing, specifically splicing the corresponding location information after the initial content feature information. Or, the fusion method can be weighted operation, etc.

[0099] Optionally, in this embodiment, the step "based on the node feature information of the neighboring nodes, performing information transmission processing on the node feature information of the node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0100] The node feature information of the node and the node feature information of the neighboring nodes are aggregated to obtain the neighbor aggregated feature information of each node in the interaction and cooperation relationship graph.

[0101] The node feature information of each node is updated based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph.

[0102] Extract the interaction and collaboration feature information corresponding to the target object from the updated interaction and collaboration relationship graph.

[0103] There are various ways to aggregate the node feature information of a node and its neighbors, and this embodiment does not limit this approach. After aggregation, the aggregated neighbor feature information of each node can be used as the new node feature information for each node.

[0104] In some embodiments, the updated node feature information corresponding to each node in the updated interaction and collaboration relationship graph can be fused to obtain the interaction and collaboration feature information corresponding to the target object.

[0105] Optionally, in this embodiment, the step "updating the node feature information of each node based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph" may include:

[0106] The node feature information of each node is updated based on the neighbor aggregation feature information of each node.

[0107] Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0108] In this process, the neighbor aggregation feature information of each node can be used as the new node feature information corresponding to each node. Then, a new round of aggregation processing is carried out based on the updated node feature information until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0109] The preset information transmission conditions can be set according to actual conditions, and this embodiment does not impose any restrictions on them. For example, the preset information transmission condition can be that the number of aggregation processes does not exceed a preset number. For instance, if the preset number is 2, then only 2 rounds of aggregation processes will be performed.

[0110] Specifically, the process of aggregating the node feature information of a node and its neighbors can be represented by the following formula (1):

[0111]

[0112] Where σ is the activation function, and W1 and W2 are parameters. This is the normalization coefficient. This represents the current node feature information of node v, specifically the neighbor aggregation feature information obtained by node v in the (k-1)th aggregation process. This represents the current node feature information of neighbor node u, specifically the neighbor aggregation feature information obtained by the (k-1)th aggregation process of neighbor node u; k represents the number of aggregation processes.

[0113] Formula (1) represents the node feature information of node v. Node characteristic information of its neighbor u Perform the k-th aggregation process to obtain the neighbor aggregation feature information of node v. Obtain neighbor aggregation feature information After that, you can This serves as new node characteristic information for node v.

[0114] The process of aggregating the node feature information of a node and its neighbors can specifically include: aggregating the node feature information of a node and its neighbors, then regularizing the aggregation result, and finally using an activation function to obtain the aggregated neighbor feature information of this node. The activation function can be the Tanh function, etc., where the Tanh function represents the hyperbolic activation function.

[0115] This embodiment can obtain node feature information with high-order interactivity by multiple aggregation processes based on the interactive collaboration relationship graph.

[0116] Specifically, before information transmission processing (i.e., aggregation processing), the node characteristic information of each node v in the interaction and collaboration relationship graph can be denoted as:

[0117] Optionally, in this embodiment, the step of "extracting the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph" may include:

[0118] The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node.

[0119] The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

[0120] In this process, for each node in the interaction and collaboration relationship graph, a new node feature information will be generated in each aggregation process. The node feature information corresponding to the node in each aggregation process can be fused. There are various fusion methods, such as splicing or weighted fusion. This embodiment does not limit this. Thus, the target node feature information that has been fused with the collaboration information can be obtained.

[0121] For example, for node v, the node feature information obtained in the k aggregation processes are as follows: Can The target node feature information of node v is obtained by fusion.

[0122] Optionally, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0123] Attention processing is performed on the target node feature information corresponding to each node to obtain attention weight information;

[0124] Based on the attention weight information, the target node feature information corresponding to each node is fused to obtain the interaction and collaboration feature information corresponding to the target object.

[0125] There are various ways to fuse the feature information of the target node corresponding to each node, such as weighted fusion.

[0126] Optionally, in this embodiment, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0127] For each content node in the interaction collaboration relationship graph, determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence. The target historical content sequence includes the interaction content of the target interaction object sorted in the order of interaction time within a historical time period. The target interaction object is the object that has an interaction relationship with the historical interaction content.

[0128] Based on the location information, the feature information of the target node corresponding to the content node is updated;

[0129] The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

[0130] To characterize the positional relationships between historical interactions within a historical content sequence, this embodiment incorporates position embedding encoding. This encoding considers the chronological order of historical interactions within the sequence, reflecting shifts in object interest. By integrating positional information into the feature information of the target node corresponding to the content node, it compensates for the loss of relevant positional information in the graph neural network. Specifically, position embedding refers to the vectorized representation of positional information.

[0131] Specifically, based on location information, the feature information of the target node corresponding to the content node is updated. This can be achieved by fusing the location information with the feature information of the target node corresponding to the content node to obtain fused feature information, which is then used as the new target node feature information for the content node. There are various ways to fuse location information and target node feature information, and this embodiment does not limit this approach. For example, the fusion method can be a concatenation process, where the location information is concatenated after the target node feature information to obtain the new target node feature information.

[0132] Optionally, in some embodiments, only the target node feature information corresponding to each updated content node can be fused to obtain the interactive collaboration feature information corresponding to the target object.

[0133] In one specific embodiment, for each historical interaction content in the historical content sequence corresponding to a certain object, the target node feature information corresponding to each historical interaction content is denoted as... Assume that the embedding vector corresponding to the position information of each historical interaction in its corresponding historical content sequence is... Then the updated target node feature information corresponding to the historical interaction content in this historical content sequence can be:

[0134] 104. Select recommended content from the candidate content based on the interactive collaboration feature information corresponding to the target object.

[0135] Optionally, in this embodiment, the step "selecting content to be recommended from candidate content based on the interactive collaboration feature information corresponding to the target object" may include:

[0136] Feature extraction is performed on the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object;

[0137] The attribute feature information and interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object;

[0138] Based on the object's interest characteristics, content to be recommended is selected from the candidate content.

[0139] There are various ways to fuse the attribute feature information and interaction collaboration feature information corresponding to the target object. This embodiment does not limit this. For example, the fusion method can be splicing or weighted fusion.

[0140] In this approach, recommended content can be selected from candidate content based on the similarity between the object's interest feature information and the content feature information corresponding to the candidate content. In some embodiments, candidate content with a similarity greater than a preset value can be identified as recommended content, and this preset value can be set according to actual conditions. In other embodiments, candidate content can be sorted according to the magnitude of similarity, such as sorting from largest to smallest, to obtain sorted candidate content, and then the top n candidate contents in the sorted candidate content can be selected as recommended content.

[0141] Optionally, in this embodiment, the step "extracting features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object" may include:

[0142] Feature extraction is performed on the attribute sub-information of the target object in each dimension to obtain the sub-feature information corresponding to the attribute sub-information in each dimension;

[0143] The sub-feature information of each dimension is fused to obtain the attribute feature information corresponding to the target object.

[0144] The object attribute information may include attribute sub-information in at least one dimension, such as the object's own attributes, short-term features, and long-term features. This embodiment does not impose any limitations on this. Specifically, short-term features may be features corresponding to historical interaction content that is relatively recent in time. For example, short-term features may include the first-level category, second-level category, and content tags of historical interaction content within the past week. Long-term features may be features corresponding to historical interaction content that is relatively distant in time. For example, long-term features may include the first-level category, second-level category, and content tags of historical interaction content within the past six months.

[0145] There are various ways to fuse sub-feature information from different dimensions, and this embodiment does not limit this approach. For example, the fusion method can be splicing or weighted fusion.

[0146] In one specific embodiment, the sub-feature information corresponding to the extracted attribute sub-information in each dimension can be pooled. Specifically, the pooling method can be average pooling with a mask. Then, the pooled sub-feature information is fused to obtain the attribute feature information corresponding to the target object. The representation ability of the attribute feature information can be further improved by two fully connected layers and the activation function Tanh.

[0147] In fully connected layers (FC), each node is connected to all nodes in the previous layer, and it can be used to synthesize the features extracted earlier.

[0148] Optionally, in this embodiment, the step "fusing the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object" may include:

[0149] Attention processing is performed on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information;

[0150] Based on the attention weight information, the attribute feature information and interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object.

[0151] Among these, attribute feature information and interaction collaboration feature information can be further fused using attention to generate interest latent space representations, i.e., object interest feature information, based on different levels of attention.

[0152] Optionally, in this embodiment, the step "selecting content to be recommended from candidate content based on the object's interest feature information" may include:

[0153] The object's interest feature information is subjected to feature extraction in at least one interest dimension to obtain interest sub-features in the at least one interest dimension;

[0154] Based on the interest sub-features of each interest dimension, candidate content to be recommended for each interest dimension is selected from the candidate content;

[0155] The candidate content to be recommended is aggregated from various interest dimensions to obtain the content to be recommended.

[0156] In this embodiment, interest sub-features in at least one interest dimension can be extracted through a self-attention mechanism, i.e., multiple interest vectors can be extracted.

[0157] In one embodiment, the feature embedding matrix of object interest feature information can be denoted as H. d×nWhere n is the sequence length of the object interest feature information, and d is the dimension of the feature embedding, a self-attention mechanism can be used to obtain the weight vector corresponding to the object interest feature information. As shown in formula (2):

[0158]

[0159] in, and These are trainable parameters. The softmax function can be used to convert the output values ​​of multi-class classification into a probability distribution ranging from [0, 1]. A vector can be obtained by weighting and summing the object interest feature information using the weight vector a.

[0160] In specific scenarios, target objects are usually not only interested in one type of content, and the types of content they are interested in can vary significantly. If only a single latent interest vector is extracted, it is difficult to fully represent all interests, and excessive differences may even lead to fragmented interest feature learning, resulting in poor convergence during modeling.

[0161] In the multi-interest extraction scheme, w2 in formula (2) can be extended to Where K is the number of interests (i.e., the number of interest dimensions to be acquired), thus transforming the weight vector a into a weight matrix A, as shown in formula (3):

[0162]

[0163] The final multi-interest vector of the target object It can be represented as: V u =HA, where the final V is obtained u It can be broken down into k interest sub-features on interest dimensions. For each interest sub-feature on interest dimension, the corresponding candidate content to be recommended on that interest dimension can be recalled from the candidate content. Then, the candidate content to be recommended on each interest dimension is aggregated to obtain the final k content to be recommended that the user may be interested in.

[0164] This process involves calculating the similarity between the interest sub-features of each interest dimension and the content feature information corresponding to the candidate content. Based on the similarity, candidate content for each interest dimension is determined from the candidate content. For example, for each interest dimension, candidate content with a similarity greater than a preset value to the interest sub-features of that interest dimension can be selected as candidate content for recommendation in that interest dimension. Alternatively, for each interest dimension, candidate content can be sorted from largest to smallest based on the similarity to the interest sub-features of that interest dimension, and the top n candidates in the sorted list can be determined as candidate content for recommendation in that interest dimension.

[0165] This involves aggregating candidate content for recommendation across various interest dimensions. Specifically, this can be achieved by performing a union operation on the candidate content for recommendation across each interest dimension to obtain the content to be recommended.

[0166] This embodiment can capture multiple interests of a target object from the object's interest feature information through a multi-interest model. Specifically, the multi-interest model can be a neural network model based on a controllable multi-interest framework for recommendation (ComiRec). Through this neural network model, multiple interests of the target object can be modeled and represented.

[0167] Optionally, in this embodiment, the step "fusing the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object" may include:

[0168] Obtain at least one reference history interaction content of the target object in terms of reference content type;

[0169] Attention processing is performed on the at least one reference historical interaction content to obtain reference content feature information corresponding to the target object;

[0170] The attribute feature information, interaction collaboration feature information, and reference content feature information corresponding to the target object are fused to obtain the object interest feature information of the target object.

[0171] In this embodiment, the reference historical interaction content of the target object and the content type of the historical interaction content can be different. For example, the historical interaction content is video content, while the reference historical interaction content can be text and image content, that is, the reference content type is text and image.

[0172] There are various ways to fuse the attribute features, interaction and collaboration features, and reference content features of the target object, and this embodiment does not limit this. For example, the fusion method can be splicing or weighted calculation.

[0173] This embodiment takes into account the scarcity of historical data and the cold start problem in the content recommendation process. Specifically, in some recommendation application scenarios, the amount of content data for different content types such as text and images, and videos, is different. For example, the number of videos is much smaller than the number of text and images. When making video recommendations, it is sometimes difficult to make accurate recommendations based solely on historically viewed videos. Furthermore, for the same object, different types of content may share common interests and preferences. Therefore, cross-domain methods can be used to fuse data from different domains, that is, to fuse interaction data from different content types, thereby increasing the amount of data in the target domain and improving the content recommendation effect in the target domain.

[0174] For example, when making content recommendations on video content types (target domains), interaction information on other content types (such as text and images) can be analyzed. For instance, the reference historical content sequence corresponding to the target object can be obtained. The reference historical content sequence includes at least one reference historical interaction content. Attention processing can be applied to the reference historical interaction content. Specifically, the Transformer can be used to pay attention to the global context in the reference historical content sequence to different degrees, thereby capturing the target object's preferences in certain aspects and achieving more accurate recommendations.

[0175] Specifically, two Transformers can be used to extract features from the reference historical interaction content. A core module of the Transformer is the self-attention mechanism module, which can model global dependencies. The specific calculation process can be found in formula (4):

[0176]

[0177] Here, Q and V represent the query (query vector) and value (value vector) of the current reference historical interaction content, respectively, and K is the key (key vector) of other reference historical interaction content in the same sequence. The Tranformer can extract strongly correlated feature information between reference historical content sequences, and further mapping yields a representation of the information in the reference historical content sequence, i.e., the reference content feature information.

[0178] Attention, specifically, is a weighted mechanism widely used in both Computer Vision (CV) and Natural Language Processing (NLP). It consists of a key, query, and value. The key (key vector) refers to the reference feature, typically the item feature; while the query and value vectors are derived from the input sequence features through linear transformations. If the key, query, and value are all the same input features, this type of attention is called "self-attention."

[0179] In specific recommendation scenarios, the online recommendation process based on the content processing method of this application can consist of recall logic, initial selection logic, and ranking logic.

[0180] The recall logic can include: obtaining interest sub-features across multiple interest dimensions based on the content processing methods described above, and then using these interest sub-features to find the recommended content that the target object might be most interested in. During the recall of recommended content, the content feature information corresponding to historical interactions can be stored offline, while the feature vectors corresponding to the object's attribute information are generated online. Because the online context of an object is dynamic, it is necessary to ensure that the feature vectors of the object's attribute information are generated online to more accurately capture the object's interests and preferences at that moment.

[0181] The initial selection logic can be responsible for initially screening a large number of recall results according to specific rules (such as user document relevance, timeliness, region, diversity, etc.) to reduce the scale of ranking calculations.

[0182] The sorting logic can estimate the click-through rate of the initially selected content to be recommended, and then sort it according to the estimation results to select the content to recommend to the target audience.

[0183] In one embodiment, such as Figure 1c The diagram shown is a model structure diagram for content recommendation. A multi-interest extraction module can be introduced based on the aggregation of object attribute information and historical interaction content features.

[0184] Specifically, the object attribute information of the target object can be obtained. This object attribute information can include attribute sub-information in at least one dimension, such as the object's own attributes, short-term features, and long-term features. Then, feature extraction layers are used to extract features from the attribute sub-information in each dimension to obtain the sub-feature information corresponding to the attribute sub-information in each dimension. Then, a masked average pooling layer is used to pool the sub-feature information corresponding to the attribute sub-information in each dimension. The pooled sub-feature information is then fused through a connection layer to obtain the attribute feature information corresponding to the target object. Finally, two fully connected layers and the Tanh activation function can be used to further improve the representation ability of the attribute feature information to obtain the user-side features.

[0185] Specifically, for the historical content sequence of the target object, a feature extraction layer can be used to extract the content feature information of each historical interaction in the historical content sequence, and a list group connection layer can be used to enhance the sequence relationship between the content feature information. In order to represent the positional relationship information between sequences, a position embedding layer can be used to represent the positional information of each historical interaction, thereby obtaining content feature information with added positional information. Then, two fully connected sequence layers are used to further improve the representation ability of content feature information, thus obtaining the features on the item side.

[0186] After obtaining user-side and item-side features, the user-side and item-side features can be further fused using an attention mechanism through an object-content attention layer to obtain an interest latent space representation, i.e., object interest feature information. Then, a multivariate interest extraction layer is used to extract features from the object interest feature information in at least one interest dimension to obtain at least one interest sub-feature in the interest dimension. For each interest sub-feature in the interest dimension, the corresponding candidate content to be recommended in the interest dimension can be recalled from the candidate content. Then, the candidate content to be recommended in each interest dimension is aggregated to obtain the final k content to be recommended that the user may be interested in.

[0187] During the offline training of the above model, training data can be acquired, which may include object interest feature information v of the sample objects. u and content feature information of the sample content e i The likelihood function of the interaction between the sample object and the sample content can be calculated using formula (4):

[0188]

[0189] Among them, P θ(i|u) represents the predicted probability that there is an interaction between the sample object and the sample content. The model parameters can be optimized based on this predicted probability and the label information corresponding to the sample object. This label information includes the expected probability that there is an interaction between the sample object and the sample content. Specifically, the loss value corresponding to the likelihood function can be calculated based on this predicted probability and the expected probability, and the model parameters can be adjusted accordingly.

[0190] The training data can be negative sample data, i.e., the expected probability that there is an interaction relationship between the sample object and the sample content is 0. In this embodiment, the identifier corresponding to the sample content in the negative sample can be obtained, and the content feature information (i.e., negative sample features) corresponding to the sample content can be obtained from the preset feature memory based on the identifier. The object attribute information of the sample object in the negative sample can be extracted through the feature extraction layer and the group average connection layer to obtain the attribute feature information corresponding to the sample object. Based on the content feature information corresponding to the sample content and the attribute feature information corresponding to the sample object, the predicted probability that there is an interaction relationship between the sample object and the sample content can be determined.

[0191] The above model neglects the collaborative information between the object and content when modeling the item-side features, resulting in insufficient extracted sequence representations. This embodiment can improve the above model based on the content processing method provided in this application. The improved model structure is as follows: Figure 1d As shown, graph neural networks can be introduced to enhance the fusion of collaborative signals based on the above model, and reference content feature information from historical interaction content can be used to solve the problem of scarce historical data in the content recommendation process.

[0192] The process of extracting the attribute feature information (i.e., user-side features) corresponding to the target object can be referred to in the previous embodiment, and will not be repeated in this embodiment.

[0193] Specifically, an interaction collaboration relationship graph can be constructed based on the interaction collaboration data of the target object. A ListWise Neighbor Sample Layer is used to determine the neighboring nodes corresponding to each node in the graph. A feature extraction layer is then used to extract features from each node, obtaining its corresponding node feature information. Next, a ListWise Neighbor Aggregator Layer aggregates the node feature information of the nodes and their neighbors in the graph, obtaining the neighbor aggregated feature information for each node. The feature extraction layer then updates the node feature information of each node based on its neighbor aggregated feature information, obtaining the node feature information corresponding to each node in each aggregation process. Finally, a list group connection layer fuses the node feature information corresponding to each node in each aggregation process, thus obtaining the target node feature information for each node. To avoid losing the relevant positional information corresponding to the historical content sequence, the positional information of each historical interaction in the historical content sequence can be explicitly encoded through the Position Embedding Layer, and further fused with the target node information corresponding to each node calculated from the interaction collaboration relationship graph. Finally, two fully connected layers with activation functions are used to further improve the expressive ability of the interaction collaboration feature information of the target object, so as to obtain the item-side features on the target domain.

[0194] To address the potential scarcity of historical data in content recommendation, an auxiliary interaction branch can be designed. This involves fusing data from different domains using cross-domain methods to increase the amount of data in the target domain, thereby improving content recommendation performance within that domain. Specifically, a sequence of historical reference content for the target object can be obtained. This sequence includes at least one historical reference interaction. Features can be extracted from each historical reference interaction using a feature extraction layer, and then the extracted features are fused using a list group connection layer. Finally, attention processing is applied to the fused features using two encoding layers (specifically, a Transformer) to obtain the sequence embedding, i.e., the reference content feature information.

[0195] After obtaining user-side and item-side features, the user-side and item-side features can be fused using an attention mechanism through an object-content attention layer. Then, the fused features of user-side and item-side features, as well as reference content feature information, are further fused through an object-content fusion layer (User Activity FuseLayer). This allows target domain recommendation to better utilize the basic features of objects and the potential interests of other domains to obtain an interest latent space representation, i.e., object interest feature information. Then, a multivariate interest extraction layer is used to extract features from the object interest feature information in at least one interest dimension to obtain at least one interest sub-feature in the interest dimension. For each interest sub-feature in the interest dimension, the corresponding candidate content to be recommended in the interest dimension can be recalled from the candidate content. Then, the candidate content to be recommended in each interest dimension is aggregated to obtain the final k content to be recommended that the user may be interested in.

[0196] As can be seen from the above, this embodiment can obtain object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration relationship graph, information transmission processing is performed on the object attribute information and the historical interaction content to obtain the interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, recommended content is selected from the candidate content. This application can deeply mine the interaction relationship between objects and content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation.

[0197] Based on the method described in the preceding embodiments, the following will provide a further detailed description, taking the specific integration of the content processing device into a server as an example.

[0198] This application provides a content processing method, such as... Figure 2 As shown, the specific process of this content processing method can be as follows:

[0199] 201. The server obtains the object attribute information of the target object, at least one historical interaction content, and interaction collaboration data, wherein the interaction collaboration data includes related objects that are related to the target object in terms of content interaction and the historical interaction content corresponding to the related objects.

[0200] Specifically, a historical content sequence of the target object within a historical time period can be obtained. This historical content sequence includes at least one historical interaction content. Each historical interaction content can specifically be content viewed by the target object at a certain time within the historical time period. The viewed content can include information in various modalities such as video, audio, text, and images; this embodiment does not impose any limitations on this. Optionally, the historical interaction content in the historical content sequence can be arranged according to the interaction time order of the target object.

[0201] Specifically, the interactive collaboration data of the target object may include related objects in terms of content interaction and the historical interaction content corresponding to the related objects.

[0202] For example, the target object can be denoted as user1. The historical interaction content corresponding to the target object includes item1. Besides user1, user2 also has an interaction relationship with item1, meaning user2 has also viewed item1. Thus, user2 can be considered a related object associated with the target object user1. Other historical interactions of user2, such as item2, are historical interaction content corresponding to related objects. Similarly, user3, another object that interacts with item2, can also be considered a related object associated with the target object user1. Furthermore, other historical interactions of user3, such as item3, are also historical interaction content corresponding to related objects. Therefore, the interactive collaboration data of the target object can include user1, user2, user3, item1, item2, and item3. Here, an item (project) can specifically be an article or a video, etc.

[0203] 202. The server constructs an interaction collaboration relationship graph corresponding to the target object based on the interaction collaboration data. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects.

[0204] Specifically, the interaction and collaboration relationship graph may include at least one node and the connections between nodes. The node may include the attribute node corresponding to the object and the content node corresponding to the content. The connection may represent the interaction relationship between the object and the content.

[0205] Among them, attribute nodes are specifically the nodes corresponding to the object's attribute information. Lines can be used to connect attribute nodes and content nodes, meaning that the node types at both ends of the line are different.

[0206] Optionally, this embodiment can generate attribute nodes corresponding to each object and content corresponding to each content based on the various objects and content in the interactive collaboration data. Then, based on whether the objects corresponding to each attribute node and the content corresponding to each content node have an interactive relationship, the attribute nodes and content nodes are connected to obtain an interactive collaboration relationship diagram. For example, when an object and a piece of content have an interactive relationship, the attribute node corresponding to the object and the content node corresponding to the content can be connected. Specifically, the interactive relationship between the object and the content means that the object has viewed the content within a historical time period.

[0207] In a specific scenario, this embodiment can construct an interaction collaboration relationship graph corresponding to the target object based on the above-mentioned interaction collaboration data. Each node in the interaction collaboration relationship graph can be divided into multiple interaction collaboration relationship layers. The node types in each interaction collaboration relationship layer are consistent, and the node types corresponding to each interaction collaboration relationship layer are alternating.

[0208] 203. The server performs information transmission processing on the object attribute information and the historical interaction content according to the interaction collaboration relationship diagram to obtain the interaction collaboration feature information corresponding to the target object.

[0209] Optionally, in this embodiment, the interaction collaboration relationship graph includes at least one node and the connection between nodes. The node includes the attribute node corresponding to the object and the content node corresponding to the content. The connection represents the interaction relationship between the object and the content.

[0210] The step "based on the interaction collaboration relationship graph, perform information transmission processing on the object attribute information and the historical interaction content to obtain the interaction collaboration feature information corresponding to the target object" may include:

[0211] In the interactive collaboration relationship graph, information is transmitted through the connections between nodes to process the object attribute information and the historical interaction content, thereby obtaining the interactive collaboration feature information corresponding to the target object.

[0212] Specifically, this embodiment can explicitly encode and transmit the interaction information of objects and content according to the interaction and collaboration relationship diagram, realize the expression model of high-order connectivity of historical interaction content, obtain the representation of historical content sequence that integrates interaction and collaboration signals, thereby enhancing the representation power of historical content sequence.

[0213] Optionally, in this embodiment, the step "transmitting information through connections between nodes in the interaction collaboration relationship graph to process the object attribute information and the historical interaction content, thereby obtaining the interaction collaboration feature information corresponding to the target object" may include:

[0214] Feature extraction is performed on each node in the interaction and collaboration relationship graph to obtain node feature information corresponding to each node. The node feature information includes node feature information of object attribute information corresponding to each attribute node and node feature information of historical interaction content corresponding to each content node.

[0215] For each node in the interaction and collaboration relationship graph, a node search is performed in the interaction and collaboration relationship graph based on the node to determine the neighboring nodes corresponding to the node;

[0216] Based on the node feature information of the neighboring nodes, information transmission processing is performed on the node feature information of the node to obtain the interaction and collaboration feature information corresponding to the target object.

[0217] Specifically, for each node in the interaction collaboration graph, a node search is performed within the graph. This search involves finding nodes located at the same interaction collaboration layer as the current node, and these nodes are identified as its neighbors. Specifically, the neighboring nodes have the same node type as the current node.

[0218] Optionally, in this embodiment, the step "based on the node feature information of the neighboring nodes, performing information transmission processing on the node feature information of the node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0219] The node feature information of the node and the node feature information of the neighboring nodes are aggregated to obtain the neighbor aggregated feature information of each node in the interaction and cooperation relationship graph.

[0220] The node feature information of each node is updated based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph.

[0221] Extract the interaction and collaboration feature information corresponding to the target object from the updated interaction and collaboration relationship graph.

[0222] In some embodiments, the updated node feature information corresponding to each node in the updated interaction and collaboration relationship graph can be fused to obtain the interaction and collaboration feature information corresponding to the target object.

[0223] Optionally, in this embodiment, the step "updating the node feature information of each node based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph" may include:

[0224] The node feature information of each node is updated based on the neighbor aggregation feature information of each node.

[0225] Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0226] In this process, the neighbor aggregation feature information of each node can be used as the new node feature information corresponding to each node. Then, a new round of aggregation processing is carried out based on the updated node feature information until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0227] The preset information transmission conditions can be set according to actual conditions, and this embodiment does not impose any restrictions on them. For example, the preset information transmission condition can be that the number of aggregation processes does not exceed a preset number. For instance, if the preset number is 2, then only 2 rounds of aggregation processes will be performed.

[0228] Optionally, in this embodiment, the step of "extracting the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph" may include:

[0229] The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node.

[0230] The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

[0231] In this process, for each node in the interaction and collaboration relationship graph, a new node feature information will be generated in each aggregation process. The node feature information corresponding to the node in each aggregation process can be fused. There are various fusion methods, such as splicing or weighted fusion. This embodiment does not limit this. Thus, the target node feature information that has been fused with the collaboration information can be obtained.

[0232] Optionally, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0233] Attention processing is performed on the target node feature information corresponding to each node to obtain attention weight information;

[0234] Based on the attention weight information, the target node feature information corresponding to each node is fused to obtain the interaction and collaboration feature information corresponding to the target object.

[0235] There are various ways to fuse the feature information of the target node corresponding to each node, such as weighted fusion.

[0236] Optionally, in this embodiment, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0237] For each content node in the interaction collaboration relationship graph, determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence. The target historical content sequence includes the interaction content of the target interaction object sorted in the order of interaction time within a historical time period. The target interaction object is the object that has an interaction relationship with the historical interaction content.

[0238] Based on the location information, the feature information of the target node corresponding to the content node is updated;

[0239] The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

[0240] To characterize the positional relationships between historical interactions within a historical content sequence, this embodiment incorporates position embedding encoding. This encoding considers the chronological order of historical interactions within the sequence, reflecting shifts in object interest. By integrating positional information into the feature information of the target node corresponding to the content node, it compensates for the loss of relevant positional information in the graph neural network. Specifically, position embedding refers to the vectorized representation of positional information.

[0241] 204. The server extracts features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object.

[0242] Optionally, in this embodiment, the step "extracting features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object" may include:

[0243] Feature extraction is performed on the attribute sub-information of the target object in each dimension to obtain the sub-feature information corresponding to the attribute sub-information in each dimension;

[0244] The sub-feature information of each dimension is fused to obtain the attribute feature information corresponding to the target object.

[0245] The object attribute information may include attribute sub-information in at least one dimension, such as the object's own attributes, short-term features, and long-term features. This embodiment does not impose any limitations on this. Specifically, short-term features may be features corresponding to historical interaction content that is relatively recent in time. For example, short-term features may include the first-level category, second-level category, and content tags of historical interaction content within the past week. Long-term features may be features corresponding to historical interaction content that is relatively distant in time. For example, long-term features may include the first-level category, second-level category, and content tags of historical interaction content within the past six months.

[0246] In one specific embodiment, the sub-feature information corresponding to the extracted attribute sub-information in each dimension can be pooled. Specifically, the pooling method can be average pooling with a mask. Then, the pooled sub-feature information is fused to obtain the attribute feature information corresponding to the target object. The representation ability of the attribute feature information can be further improved by two fully connected layers and the activation function Tanh.

[0247] 205. The server fuses the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object.

[0248] Optionally, in this embodiment, the step "fusing the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object" may include:

[0249] Attention processing is performed on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information;

[0250] Based on the attention weight information, the attribute feature information and interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object.

[0251] Among these, attribute feature information and interaction collaboration feature information can be further fused using attention to generate interest latent space representations, i.e., object interest feature information, based on different levels of attention.

[0252] 206. The server selects content to be recommended from the candidate content based on the object's interest feature information.

[0253] In this approach, recommended content can be selected from candidate content based on the similarity between the object's interest feature information and the content feature information corresponding to the candidate content. In some embodiments, candidate content with a similarity greater than a preset value can be identified as recommended content, and this preset value can be set according to actual conditions. In other embodiments, candidate content can be sorted according to the magnitude of similarity, such as sorting from largest to smallest, to obtain sorted candidate content, and then the top n candidate contents in the sorted candidate content can be selected as recommended content.

[0254] Optionally, in this embodiment, the step "selecting content to be recommended from candidate content based on the object's interest feature information" may include:

[0255] The object's interest feature information is subjected to feature extraction in at least one interest dimension to obtain interest sub-features in the at least one interest dimension;

[0256] Based on the interest sub-features of each interest dimension, candidate content to be recommended for each interest dimension is selected from the candidate content;

[0257] The candidate content to be recommended is aggregated from various interest dimensions to obtain the content to be recommended.

[0258] In this embodiment, interest sub-features in at least one interest dimension can be extracted through a self-attention mechanism, i.e., multiple interest vectors can be extracted.

[0259] As can be seen from the above, this embodiment can obtain the object attribute information, at least one historical interaction content, and interaction collaboration data of the target object through a server. The interaction collaboration data includes related objects that are related to the target object in terms of content interaction and the historical interaction content corresponding to the related objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the related objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the related objects. Based on the interaction collaboration relationship graph, information transmission processing is performed on the object attribute information and the historical interaction content to obtain the interaction collaboration feature information corresponding to the target object. Feature extraction is performed on the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object. The attribute feature information and the interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object. Based on the object interest feature information, content to be recommended is selected from the candidate content. This application can deeply mine the interaction relationship between objects and content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation.

[0260] To better implement the above methods, embodiments of this application also provide a content processing apparatus, such as... Figure 3 As shown, the content processing device may include an acquisition unit 301, a construction unit 302, an information transmission unit 303, and a selection unit 304, as follows:

[0261] (1) Obtain unit 301;

[0262] The acquisition unit is used to acquire object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects.

[0263] (2) Constructing unit 302;

[0264] The construction unit is used to construct an interaction collaboration relationship graph corresponding to the target object based on the interaction collaboration data. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects.

[0265] (3) Information transmission unit 303;

[0266] The information transmission unit is used to perform information transmission processing on the object attribute information and the historical interaction content according to the interaction collaboration relationship diagram, so as to obtain the interaction collaboration feature information corresponding to the target object.

[0267] Optionally, in some embodiments of this application, the interaction collaboration relationship graph includes at least one node and connections between nodes. The node includes an attribute node corresponding to the object and a content node corresponding to the content. The connections represent the interaction relationship between the object and the content. Specifically, the information transmission unit can be used to transmit information through the connections between nodes in the interaction collaboration relationship graph to process the object attribute information and the historical interaction content, thereby obtaining the interaction collaboration feature information corresponding to the target object.

[0268] Optionally, in some embodiments of this application, the information transmission unit may include an extraction subunit, a node search subunit, and a transmission subunit, as follows:

[0269] The extraction subunit is used to extract features from each node in the interaction and collaboration relationship graph to obtain node feature information corresponding to each node. The node feature information includes node feature information of object attribute information corresponding to each attribute node and node feature information of historical interaction content corresponding to each content node.

[0270] The node search subunit is used to perform a node search in the interaction and collaboration relationship graph for each node, in order to determine the neighboring nodes of the node.

[0271] The transmission subunit is used to process the node feature information of the node according to the node feature information of the neighboring node to obtain the interaction and cooperation feature information corresponding to the target object.

[0272] Optionally, in some embodiments of this application, the transmission subunit may be specifically used to aggregate the node feature information of the node and the node feature information of the neighboring nodes to obtain the neighbor aggregated feature information of each node in the interaction collaboration relationship graph; update the node feature information of each node according to the neighbor aggregated feature information of each node to obtain the updated interaction collaboration relationship graph; and extract the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph.

[0273] Optionally, in some embodiments of this application, the step "updating the node feature information of each node based on the neighbor aggregation feature information of each node to obtain an updated interaction and cooperation relationship graph" may include:

[0274] The node feature information of each node is updated based on the neighbor aggregation feature information of each node.

[0275] Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

[0276] Optionally, in some embodiments of this application, the step of "extracting the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph" may include:

[0277] The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node.

[0278] The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

[0279] Optionally, in some embodiments of this application, the step "fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object" may include:

[0280] For each content node in the interaction collaboration relationship graph, determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence. The target historical content sequence includes the interaction content of the target interaction object sorted in the order of interaction time within a historical time period. The target interaction object is the object that has an interaction relationship with the historical interaction content.

[0281] Based on the location information, the feature information of the target node corresponding to the content node is updated;

[0282] The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

[0283] (4) Select unit 304;

[0284] The selection unit is used to select content to be recommended from candidate content based on the interactive collaboration feature information corresponding to the target object.

[0285] Optionally, in some embodiments of this application, the selection unit may include a feature extraction subunit, a fusion subunit, and a selection subunit, as follows:

[0286] The feature extraction subunit is used to extract features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object.

[0287] The fusion subunit is used to fuse the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object;

[0288] The selection sub-unit is used to select content to be recommended from the candidate content based on the object's interest feature information.

[0289] Optionally, in some embodiments of this application, the fusion subunit may be specifically used to perform attention processing on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information; and to fuse the attribute feature information and interaction collaboration feature information corresponding to the target object according to the attention weight information to obtain the object interest feature information of the target object.

[0290] Optionally, in some embodiments of this application, the selection sub-unit may be specifically used to extract features from the object's interest feature information in at least one interest dimension to obtain interest sub-features in the at least one interest dimension; select candidate content to be recommended in each interest dimension from the candidate content based on the interest sub-features in each interest dimension; and perform aggregation processing on the candidate content to be recommended in each interest dimension to obtain the content to be recommended.

[0291] Optionally, in some embodiments of this application, the fusion subunit may be specifically used to obtain at least one reference historical interaction content of the target object in terms of reference content type; perform attention processing on the at least one reference historical interaction content to obtain reference content feature information corresponding to the target object; and fuse the attribute feature information, interaction collaboration feature information, and reference content feature information corresponding to the target object to obtain object interest feature information of the target object.

[0292] As can be seen from the above, in this embodiment, the acquisition unit 301 can acquire the object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects. The construction unit 302 constructs an interaction collaboration relationship graph corresponding to the target object based on the interaction collaboration data. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. The information transmission unit 303 performs information transmission processing on the object attribute information and the historical interaction content based on the interaction collaboration relationship graph to obtain the interaction collaboration feature information corresponding to the target object. The selection unit 304 selects content to be recommended from the candidate content based on the interaction collaboration feature information corresponding to the target object. This application can deeply mine the interaction relationship between objects and content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation.

[0293] This application also provides an electronic device, such as... Figure 4 The diagram shows a schematic representation of the structure of an electronic device according to an embodiment of this application. This electronic device can be a terminal or a server, specifically:

[0294] The electronic device may include components such as a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will understand that... Figure 4The 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:

[0295] The processor 401 is the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 402, and calls data stored in the memory 402, to perform various functions and process data. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 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 401.

[0296] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 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 electronic device, etc. In addition, the memory 402 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 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0297] The electronic device also includes a power supply 403 that supplies power to the various components. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 403 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.

[0298] The electronic device may also include an input unit 404, 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.

[0299] 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 401 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 402 according to the following instructions, and the processor 401 runs the applications stored in the memory 402 to realize various functions, as follows:

[0300] The process involves: acquiring object attribute information of a target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in content interaction and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. This graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration relationship graph, information transfer processing is performed on the object attribute information and the historical interaction content to obtain interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, content to be recommended is selected from candidate content.

[0301] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0302] As can be seen from the above, this embodiment can obtain object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration relationship graph, information transmission processing is performed on the object attribute information and the historical interaction content to obtain the interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, recommended content is selected from the candidate content. This application can deeply mine the interaction relationship between objects and content through the interaction collaboration data of the target object, thereby selecting content to recommend to the target object, which is beneficial to improving the accuracy of content recommendation.

[0303] 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 instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0304] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute steps in any of the content processing methods provided in embodiments of this application. For example, the instructions can execute the following steps:

[0305] The process involves: acquiring object attribute information of a target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in content interaction and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration data, an interaction collaboration relationship graph corresponding to the target object is constructed. This graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects, and the content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. Based on the interaction collaboration relationship graph, information transfer processing is performed on the object attribute information and the historical interaction content to obtain interaction collaboration feature information corresponding to the target object. Based on the interaction collaboration feature information corresponding to the target object, content to be recommended is selected from candidate content.

[0306] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0307] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0308] Since the instructions stored in the computer-readable storage medium can execute the steps of any of the content processing methods provided in the embodiments of this application, the beneficial effects that any of the content processing methods provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0309] According to one aspect of this application, a computer program product or computer program is provided, comprising 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 provided in various alternative implementations of the above-described processing aspects.

[0310] The above provides a detailed description of a content processing method and related equipment provided by the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A content processing method, characterized in that, include: Obtain object attribute information of the target object, at least one historical interaction content, and interaction collaboration data, wherein the interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects; Based on the interactive collaboration data, an interactive collaboration relationship graph corresponding to the target object is constructed. The interactive collaboration relationship graph represents the interactive relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interactive content corresponding to the target object and the historical interactive content corresponding to the associated objects. The interactive collaboration relationship graph includes at least one node. The node includes the attribute node corresponding to the object and the content node corresponding to the content. For each attribute node in the interaction collaboration graph, feature extraction is performed on the object attribute information corresponding to the attribute node to obtain the node feature information of the attribute node; for each content node in the interaction collaboration graph, feature extraction is performed on the historical interaction content corresponding to the content node to obtain the initial content feature information of the content node, and the position information of the historical interaction content corresponding to the content node in the target historical content sequence is determined. The target historical content sequence includes the interaction content of the target interaction object ordered according to the interaction time within a historical time period. The target interaction object is the object that has an interaction relationship with the historical interaction content. The initial content feature information and the position information are fused to obtain the node feature information corresponding to the content node; for each node in the interaction collaboration graph, a node search is performed on the interaction collaboration graph to determine the neighboring nodes corresponding to the node; based on the node feature information of the neighboring nodes, information transmission processing is performed on the node feature information of the node to obtain the interaction collaboration feature information corresponding to the target object; Feature extraction is performed on the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object. At least one reference historical interaction content of the target object in terms of reference content type is obtained. The reference historical interaction content of the target object has a different content type than the historical interaction content. Attention processing is performed on the at least one reference historical interaction content to obtain the reference content feature information corresponding to the target object. The attribute feature information, interaction collaboration feature information, and reference content feature information corresponding to the target object are fused to obtain the object interest feature information of the target object. Feature extraction is performed on the object interest feature information in at least one interest dimension to obtain the interest sub-features in the at least one interest dimension. Based on the interest sub-features in each interest dimension, candidate content to be recommended in each interest dimension is selected from the candidate content. The candidate content to be recommended in each interest dimension is aggregated to obtain the content to be recommended.

2. The method according to claim 1, characterized in that, The interactive collaboration graph also includes connections between nodes, whereby the connections represent the interaction relationship between the object and the content; the method further includes: In the interactive collaboration relationship graph, information is transmitted through the connections between nodes to process the object attribute information and the historical interaction content, thereby obtaining the interactive collaboration feature information corresponding to the target object.

3. The method according to claim 1, characterized in that, The step of processing the node feature information of the neighboring nodes to obtain the interaction and collaboration feature information corresponding to the target object by performing information transmission processing on the node feature information of the neighboring nodes includes: The node feature information of the node and the node feature information of the neighboring nodes are aggregated to obtain the neighbor aggregated feature information of each node in the interaction and cooperation relationship graph. The node feature information of each node is updated based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph. Extract the interaction and collaboration feature information corresponding to the target object from the updated interaction and collaboration relationship graph.

4. The method according to claim 3, characterized in that, The step of updating the node feature information of each node based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph includes: The node feature information of each node is updated based on the neighbor aggregation feature information of each node. Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

5. The method according to claim 4, characterized in that, The step of extracting the interaction collaboration feature information corresponding to the target object from the updated interaction collaboration relationship graph includes: The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node. The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

6. The method according to claim 5, characterized in that, The step of fusing the target node feature information corresponding to each node to obtain the interaction and collaboration feature information corresponding to the target object includes: Based on the location information, the feature information of the target node corresponding to the content node is updated; The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

7. The method according to claim 1, characterized in that, The process of fusing the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain the object interest feature information of the target object includes: Attention processing is performed on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information; Based on the attention weight information, the attribute feature information and interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object.

8. A content processing apparatus, characterized in that, include: The acquisition unit is used to acquire object attribute information of the target object, at least one historical interaction content, and interaction collaboration data. The interaction collaboration data includes associated objects related to the target object in terms of content interaction and the historical interaction content corresponding to the associated objects. A construction unit is configured to construct an interaction collaboration relationship graph corresponding to the target object based on the interaction collaboration data. The interaction collaboration relationship graph represents the interaction relationship between objects and content. The objects include the target object and the associated objects. The content includes the historical interaction content corresponding to the target object and the historical interaction content corresponding to the associated objects. The interaction collaboration relationship graph includes at least one node, and the node includes the attribute node corresponding to the object and the content node corresponding to the content. The information transmission unit includes an extraction subunit, a node search subunit, and a transmission subunit. The extraction subunit is used to: extract features from the object attribute information corresponding to each attribute node in the interaction collaboration relationship graph to obtain node feature information of the attribute node; extract features from the historical interaction content corresponding to each content node in the interaction collaboration relationship graph to obtain initial content feature information of the content node; determine the position information of the historical interaction content corresponding to the content node in the target historical content sequence, where the target historical content sequence includes interaction content of the target interaction object ordered by interaction time within a historical time period, and the target interaction object is an object that has an interaction relationship with the historical interaction content; and fuse the initial content feature information and the position information to obtain node feature information corresponding to the content node. The node search subunit is used to: perform node search in the interaction collaboration relationship graph for each node to determine the neighboring nodes corresponding to the node. The transmission subunit is used to: perform information transmission processing on the node feature information of the node based on the node feature information of the neighboring nodes to obtain the interaction collaboration feature information corresponding to the target object. The selection unit includes a feature extraction subunit, a fusion subunit, and a selection subunit: The feature extraction subunit is used to extract features from the object attribute information corresponding to the target object to obtain the attribute feature information corresponding to the target object. The fusion subunit is used to obtain at least one reference historical interaction content of the target object in terms of reference content type, wherein the reference historical interaction content of the target object is of a different content type than the historical interaction content; to perform attention processing on the at least one reference historical interaction content to obtain cross-domain reference content feature information corresponding to the target object; and to fuse the attribute feature information, interaction collaboration feature information and reference content feature information corresponding to the target object to obtain object interest feature information of the target object. A sub-unit is selected to extract features from the object's interest feature information in at least one interest dimension, thereby obtaining interest sub-features in the at least one interest dimension; based on the interest sub-features in each interest dimension, candidate content to be recommended in each interest dimension is selected from the candidate content; the candidate content to be recommended in each interest dimension is aggregated to obtain the content to be recommended.

9. The content processing apparatus according to claim 8, characterized in that, The interactive collaboration graph also includes connections between nodes, whereby the connections represent the interaction relationship between the object and the content. The information transmission unit is further used for: In the interactive collaboration relationship graph, information is transmitted through the connections between nodes to process the object attribute information and the historical interaction content, thereby obtaining the interactive collaboration feature information corresponding to the target object.

10. The content processing apparatus according to claim 9, characterized in that, The transmission subunit is used for: The node feature information of the node and the node feature information of the neighboring nodes are aggregated to obtain the neighbor aggregated feature information of each node in the interaction and cooperation relationship graph. The node feature information of each node is updated based on the neighbor aggregation feature information of each node to obtain the updated interaction and cooperation relationship graph. Extract the interaction and collaboration feature information corresponding to the target object from the updated interaction and collaboration relationship graph.

11. The content processing apparatus according to claim 10, characterized in that, The transmission subunit is also used for: The node feature information of each node is updated based on the neighbor aggregation feature information of each node. Return to the step of aggregating the node feature information of the node and the node feature information of the neighboring nodes until the node feature information corresponding to the node that meets the preset information transmission conditions is obtained.

12. The content processing apparatus according to claim 11, characterized in that, The transmission subunit is also used for: The node feature information of each node in each aggregation process is fused to obtain the target node feature information of each node. The target node feature information corresponding to each node is fused to obtain the interactive collaboration feature information corresponding to the target object.

13. The content processing apparatus according to claim 12, characterized in that, The transmission subunit is also used for: Based on the location information, the feature information of the target node corresponding to the content node is updated; The updated target node feature information of each node is fused to obtain the interactive collaboration feature information of the target object.

14. The content processing apparatus according to claim 8, characterized in that, The fusion subunit is also used for: Attention processing is performed on the attribute feature information and interaction collaboration feature information corresponding to the target object to obtain attention weight information; Based on the attention weight information, the attribute feature information and interaction collaboration feature information corresponding to the target object are fused to obtain the object interest feature information of the target object.

15. An electronic device, characterized in that, It includes a memory and a processor; the memory stores an application program, and the processor runs the application program within the memory to perform the operations in the content processing method according to any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to perform the steps of the content processing method according to any one of claims 1 to 7.

17. A computer program product comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the content processing method according to any one of claims 1 to 7.

Images