A method and device for processing cultural relic content information, an electronic device, and a storage medium

Abstract

The present application discloses a method and apparatus for processing cultural relic content information, an electronic device, and a storage medium. The method includes: extracting spatio-temporal features from the current spatio-temporal map of the target user to obtain node feature vectors of current target cultural relic content items; for each target cultural relic content item, combining the node feature vector of the current target cultural relic content item with the real-time behavior feature vector of the target user to obtain a real-time feature vector corresponding to the target cultural relic content item, and performing weighted fusion of the real-time feature vector with the feature vector of the cultural relic corresponding to the target cultural relic content item in the cultural relic knowledge graph to obtain an initial recommended content feature corresponding to the target cultural relic content item; combining the initial recommended content feature with the context feature of the current browsing content in the cultural relic knowledge graph to obtain a final recommended content feature corresponding to the target cultural relic content item; and based on the final recommended content feature corresponding to the target cultural relic content item, matching multi-modal current recommended cultural relic content items and pushing them.

Description

Technical Field

[0001] The present application relates to the technical field of content recommendation, and particularly relates to a method and device for processing cultural relic content information, an electronic device, and a storage medium. Background Art

[0002] Currently, in order to provide content that meets the needs or preferences of each user, many applications or web pages, etc., will perform personalized content recommendation for each user. Therefore, in the digital human live broadcast platform, there is also a need for personalized content recommendation for the content of cultural relic web pages.

[0003] Currently, for personalized recommendation, it is mainly based on the historical data of the user in the recent period of time, and through collaborative filtering and deep learning methods, the content that the user is interested in is analyzed. Then, in the current and future periods of time, other content similar to the analyzed content that the user is interested in will be recommended to the user, so as to meet the user's needs.

[0004] However, in the analysis process of the existing methods, the captured user behavior is static, lacking the analysis of spatio-temporal characteristics, that is, the changes of user behavior over time and space, such as the changes from weekdays to weekends, and the changes from text content to image content and other dynamic data. Therefore, the accuracy of the final analysis result is insufficient. And although the existing methods will try their best to analyze the recent historical data to reduce the lag, there is still a certain lag because they cannot be adjusted in time according to the current situation, which will also affect the accuracy of the result. Moreover, the existing methods only recommend similar content, so the recommended content is concentrated in a single modality. Due to the failure to fully consider the content connection between cultural relics, the depth of the recommended content is limited, so the richness of the recommended content is ineffective and cannot well meet the diverse needs of users. Summary of the Invention

[0005] Based on the above deficiencies of the prior art, the present application provides a method and device for processing cultural relic content information, an electronic device, and a storage medium to solve the problems that the prior art cannot achieve accurate recommendation and the depth of the recommended content is limited.

[0006] To achieve the above object, the present application provides the following technical solutions:

[0007] The first aspect of the present application provides a method for processing cultural relic content information, including:

[0008] Performing spatio-temporal feature extraction on the current cultural relic spatio-temporal map of the target user to obtain the node feature vectors of the current target cultural relic content; wherein, the historical behavior data of the target user for each target cultural relic content and the attributes of each target cultural relic content are recorded in the current cultural relic spatio-temporal map;

[0009] For each of the target cultural relic contents, combine the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user to obtain the real-time feature vector corresponding to the target cultural relic content;

[0010] Perform weighted fusion on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content;

[0011] Combine the initial recommended content feature corresponding to the target cultural relic content with the context feature of the currently browsed content in the cultural relic knowledge graph to obtain the final recommended content feature corresponding to the target cultural relic content;

[0012] Based on the final recommended content feature corresponding to the target cultural relic content, match the current recommended cultural relic content in multiple modalities and push it to the target user.

[0013] Optionally, in the above cultural relic content information processing method, before extracting the spatio-temporal features of the current cultural relic spatio-temporal graph of the target user to obtain the node feature vectors of the current target cultural relic contents, it further includes:

[0014] Obtain the latest historical behavior data of the target user;

[0015] For the target cultural relic content that generates behaviors recorded in the historical behavior data of the target user, create a node corresponding to the target cultural relic content based on the attributes of the target cultural relic content;

[0016] Initialize the node feature vector of the target cultural relic content according to the attributes of the target cultural relic content;

[0017] Connect the nodes corresponding to each target cultural relic content in the order of behaviors in the historical behavior data of the target user, and record the behavior time on the connected edges to obtain the current cultural relic spatio-temporal graph.

[0018] Optionally, in the above cultural relic content information processing method, extracting the spatio-temporal features of the current cultural relic spatio-temporal graph of the target user to obtain the node feature vectors of the current target cultural relic contents includes:

[0019] For each of the target cultural relic contents, determine each neighbor node of the target cultural relic content in the current cultural relic spatio-temporal graph;

[0020] Calculate the normalization coefficients between the target cultural relic content and each neighbor node;

[0021] Determine the current weight matrix corresponding to the target cultural relic content;

[0022] Based on the normalization coefficients of the target cultural relic content and each of the neighbor nodes, the node feature vector of the target cultural relic content, and the current weight matrix corresponding to the target cultural relic content, obtain the current weighted sum of the target cultural relic content;

[0023] Through the current weighting of the target cultural relic content, a preset bias term, and a preset activation function, calculate to obtain the node feature vector of the current target cultural relic content.

[0024] Optionally, in the above method for processing cultural relic content information, the combining the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user for each of the target cultural relic contents to obtain the real-time feature vector corresponding to the target cultural relic content includes:

[0025] Generate the real-time behavior feature vector of the target user based on the real-time behavior data of the target user;

[0026] For each of the target cultural relic contents, through the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user, obtain the real-time feature vector corresponding to the target cultural relic content.

[0027] Optionally, in the above method for processing cultural relic content information, it further includes:

[0028] Define each cultural relic and each cultural relic category as entities in the knowledge graph, and add the detailed information of each entity;

[0029] Connect the entity of each cultural relic with the entity of the cultural relic category to which it belongs, and with the entities of other cultural relics belonging to the same cultural relic category, to obtain a cultural relic knowledge graph.

[0030] Optionally, in the above method for processing cultural relic content information, the weighted fusion of the real-time feature vector corresponding to the target cultural relic content with the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content includes:

[0031] Perform weighted calculation on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph according to the fusion weight to obtain the initial recommended content feature corresponding to the target cultural relic content.

[0032] Optionally, in the above-mentioned cultural relic content information processing method, the combining of the initial recommended content features corresponding to the target cultural relic content with the context features of the currently viewed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content includes:

[0033] Weight the initial recommended content features corresponding to the target cultural relic content and the context features of the currently viewed content in the cultural relic knowledge graph according to the context weight to obtain the final recommended content features corresponding to the target cultural relic content; wherein, the context weight is determined according to the type of the currently viewed content and / or the key information in the currently viewed content.

[0034] The second aspect of the present application provides a cultural relic content information processing device, including:

[0035] A node feature update unit, configured to perform spatio-temporal feature extraction on the current cultural relic spatio-temporal graph of the target user to obtain the node feature vectors of each current target cultural relic content; wherein, the historical behavior data of the target user for each of the target cultural relic contents and the attributes of each of the target cultural relic contents are recorded in the current cultural relic spatio-temporal graph.

[0036] A behavior feature combination unit, configured to respectively combine the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user for each of the target cultural relic contents to obtain the real-time feature vector corresponding to the target cultural relic content.

[0037] A cultural relic feature fusion unit, configured to perform weighted fusion on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content features corresponding to the target cultural relic content.

[0038] A context feature combination unit, configured to combine the initial recommended content features corresponding to the target cultural relic content with the context features of the currently viewed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content.

[0039] A recommendation unit, configured to match and push multi-modal current recommended cultural relic contents to the target user based on the final recommended content features corresponding to the target cultural relic content.

[0040] Optionally, in the above-mentioned cultural relic content information processing device, it further includes:

[0041] A data acquisition unit, configured to acquire the latest historical behavior data of the target user.

[0042] A node creation unit, configured to create a node corresponding to the target cultural relic content for the target cultural relic content with generation behavior recorded in the historical behavior data of the target user, based on the attributes of the target cultural relic content;

[0043] A vector initialization unit, configured to initialize the node feature vector of the target cultural relic content according to the attributes of the target cultural relic content;

[0044] A node connection unit, configured to connect the nodes corresponding to each target cultural relic content in the order of behaviors in the historical behavior data of the target user, and record the behavior time on the connected edges to obtain the current cultural relic spatio-temporal graph.

[0045] Optionally, in the above cultural relic content information processing device, the node feature update unit includes:

[0046] A neighbor node determination unit, configured to respectively determine each neighbor node of the target cultural relic content in the current cultural relic spatio-temporal graph for each target cultural relic content;

[0047] A coefficient calculation unit, configured to calculate the normalization coefficients between the target cultural relic content and each neighbor node;

[0048] A weight determination unit, configured to determine the current weight matrix corresponding to the target cultural relic content;

[0049] An accumulation unit, configured to obtain the current weighted sum of the target cultural relic content based on the normalization coefficients between the target cultural relic content and each neighbor node, the node feature vector of the target cultural relic content, and the current weight matrix corresponding to the target cultural relic content;

[0050] An update unit, configured to calculate through the current weighting of the target cultural relic content, a preset bias term, and a preset activation function to obtain the latest node feature vector of the current target cultural relic content.

[0051] Optionally, in the above cultural relic content information processing device, the behavior feature combination unit includes:

[0052] A behavior feature generation unit, configured to generate a real-time behavior feature vector of the target user based on the real-time behavior data of the target user;

[0053] A node feature weighting unit, configured to respectively obtain the real-time feature vector corresponding to the target cultural relic content through the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user for each target cultural relic content.

[0054] Optionally, in the above cultural relic content information processing device, it further includes:

[0055] An entity creation unit for defining each cultural relic and each category of cultural relics as entities in a knowledge graph and adding detailed information of each of the entities;

[0056] An entity connection unit for connecting the entities of each cultural relic with the entities of the category of cultural relics to which they belong, and connecting with the entities of other cultural relics belonging to the same category of cultural relics to obtain a cultural relic knowledge graph.

[0057] Optionally, in the above cultural relic content information processing device, the cultural relic feature fusion unit includes:

[0058] A cultural relic feature weighting unit for performing weighted calculation on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph according to a fusion weight to obtain the initial recommended content feature corresponding to the target cultural relic content.

[0059] Optionally, in the above cultural relic content information processing device, the context feature combination unit includes:

[0060] A context feature combination subunit for weighting the initial recommended content feature corresponding to the target cultural relic content and the context feature of the currently viewed content in the cultural relic knowledge graph according to a context weight to obtain the final recommended content feature corresponding to the target cultural relic content; wherein, the context weight is determined according to the type of the currently viewed content and / or key information in the currently viewed content.

[0061] A third aspect of the present application provides an electronic device, including:

[0062] A memory and a processor;

[0063] Wherein, the memory is used for storing a program;

[0064] The processor is used for executing the program, and when the program is executed, it is specifically used to implement the cultural relic content information processing method as described in any one of the above.

[0065] A fourth aspect of the present application provides a computer storage medium for storing a computer program, and when the computer program is executed by a processor, it is used to implement the cultural relic content information processing method as described in any one of the above.

[0066] The embodiment of the present application provides a method for processing cultural relic content information, constructing a spatio-temporal map of cultural relics, and recording the historical behavior data of the latest target user for each target cultural relic content and the attributes of each target cultural relic content. Then, spatio-temporal features are extracted from the current spatio-temporal map of the target user to obtain the node feature vectors of each current target cultural relic content, so as to extract relevant features from the time dimension and space dimension, and obtain a more accurate reflection of the change of the user's interest over time and space. Furthermore, the accuracy of the recommended content can be improved. For each target cultural relic content, the node feature vector of the current target cultural relic content is combined with the real-time behavior feature vector of the target user to obtain the real-time feature vector corresponding to the target cultural relic content, so as to fully consider the user's behavior at the current moment, ensure that the obtained features reflect the user's current interest, and further improve the accuracy of the recommended content. Then, the real-time feature vector corresponding to the target cultural relic content is weighted and fused with the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content, so that the recommended content can be enriched and the depth of the recommended content can be increased by combining the features of the cultural relics in the knowledge graph. Then, the initial recommended content feature corresponding to the target cultural relic content is combined with the context feature of the current browsing content in the cultural relic knowledge graph to obtain the final recommended content feature corresponding to the target cultural relic content, so that the provided content conforms to the current browsing interest and improves the accuracy of the recommended content. Finally, based on the final recommended content feature corresponding to the target cultural relic content, multi-modal current recommended cultural relics are matched and pushed to the target user, thus realizing a method that can accurately provide multi-modal content. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0068] Figure 1 It is a flowchart of a method for processing cultural relic content information provided by an embodiment of the present application;

[0069] Figure 2 It is a flowchart of a method for constructing a current spatio-temporal map of cultural relics provided by an embodiment of the present application;

[0070] Figure 3 It is a flowchart of a method for updating node feature vectors provided by an embodiment of the present application;

[0071] Figure 4Schematic diagram of the architecture of an artifact content information processing device provided by an embodiment of the present application;

[0072] Figure 5 Schematic diagram of the architecture of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0073] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0074] In the present application, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0075] An embodiment of the present application provides an artifact content information processing method, as Figure 1 shown, specifically including the following steps:

[0076] S101. Extract spatio-temporal features from the current artifact spatio-temporal map of the target user to obtain node feature vectors of current target artifact contents.

[0077] Among them, the current artifact spatio-temporal map records the historical behavior data of the latest target user for each target artifact content and the attributes of each target artifact content. That is, the artifact spatio-temporal map records the target artifact contents on which the target user has recently performed behaviors, the types and times of the performed behaviors, and the order of behavior transformation between each target artifact content, so as to reflect the relationship between the behaviors of the target user for artifact contents and the time dimension, and reflect the relationship between artifact contents in the space dimension. For example, it records that the user browsed artifact A at time t1, and then jumped from browsing artifact content A to browsing artifact content B at time t2.

[0078] In order to analyze the target user's recent situation, at least the historical behavior data of the latest target user for each target cultural relic content and the attributes of each target cultural relic content need to be recorded in the current cultural relic spatio-temporal map. Therefore, the cultural relic spatio-temporal map needs to be continuously updated according to the user's behavior data. Specifically, it can be updated in real time or at regular intervals. When updating, usually new data is constructed in the original cultural relic spatio-temporal map, or of course, a new cultural relic spatio-temporal map can be reconstructed using the data.

[0079] Optionally, in another embodiment of the present application, a method for constructing a current cultural relic spatio-temporal map is provided, as Figure 2 shown, including:

[0080] S201. Obtain the historical behavior data of the latest target user.

[0081] Among them, the historical behavior data records the user's behavior regarding cultural relic content at various times, specifically including data on interactive behaviors such as user browsing and clicking. For example, the user browsed the text description of Song Dynasty celadon in the morning and watched the video of Song Dynasty celadon in the afternoon.

[0082] S202. For the target cultural relic content that generates behaviors recorded in the historical behavior data of the target user, create nodes corresponding to the target cultural relic content based on the attributes of the target cultural relic content.

[0083] Among them, the attributes of the target cultural relic content are the attributes of its corresponding cultural relic, such as name, age, material, discovery location, etc. Therefore, in the embodiments of the present application, contents such as the text description, pictures, videos, and interactive elements of the cultural relic are defined as nodes in the cultural relic spatio-temporal map, and the attributes of the corresponding cultural relic are used as the attributes of the nodes. Node 1 represents a text description page of Song Dynasty celadon, and its attributes include: age = Song Dynasty, material = celadon, discovery location = Zhejiang. Node 2 represents a picture of Song Dynasty celadon, and its attributes include age = Song Dynasty, material = celadon, discovery location = Zhejiang.

[0084] S203. Initialize the node feature vector of the target cultural relic content according to the attributes of the target cultural relic content.

[0085] It should be noted that for the initially created nodes, since their features have not been extracted yet, they are not related to the user's behavior. Therefore, initially, their attributes are used as the feature vectors of the corresponding nodes, so as to facilitate the subsequent update of the feature vectors of these nodes on this basis.

[0086] S204. Connect the nodes corresponding to each target cultural relic content in the order of the behaviors in the historical behavior data of the target user, and record the behavior time on the connected edges to obtain the current cultural relic spatio-temporal map.

[0087] Specifically, in the embodiments of the present application, the interaction relationships of the centennial logo users among different cultural relic contents include clicking, browsing time, scrolling behavior, etc. Moreover, the edges can further represent the association relationships among cultural relics, such as "belonging to the same dynasty" and "appearing at the same location".

[0088] Moreover, in order to record the occurrence time of each behavior, the behavior time will be recorded on the edges. Therefore, each node can be connected to other nodes through edges to form a user behavior path. The relationship between the edge and the time dimension is that each edge has a timestamp to record the time when the user jumps from one node to another node. The relationship between the node and the time dimension is that the feature vector of each node can be updated over time.

[0089] After updating the current cultural relic spatio-temporal graph, accordingly, the node feature vectors of the target cultural relic content will be continuously updated by analyzing the current cultural relic spatio-temporal graph, so that the current interests of the user can be reflected through the node feature vectors of the target cultural relic content.

[0090] Specifically, in order to capture the time characteristics and space characteristics of user behavior, in the embodiments of the present application, a spatio-temporal convolutional layer is used to capture the time dependence and space dependence of user behavior.

[0091] Among them, the time dependence is the change in the user's interest in different types of cultural relic contents at different times. For example, the user is more concerned about academic research contents during the day on weekdays, while more concerned about entertaining and highly interactive contents on weekends at night. Specifically, for example, the user browsed the text description of the Song Dynasty celadon at 10:00 in the morning and watched the video of the Song Dynasty celadon at 5:00 in the afternoon, indicating the change in the user's interest in different types of contents at different times.

[0092] The space dependence is the interaction relationship among the user's browsing of different cultural relic contents. For example, after viewing the text description of a certain cultural relic, the user may click on the relevant pictures or videos. Specifically, for example, after viewing the text description of the Song Dynasty celadon, the user then clicked on several pictures of the Song Dynasty celadon and then watched a video about the production process of the Song Dynasty celadon.

[0093] Optionally, in another embodiment of the present application, a specific implementation manner of step S101 is as Figure 3 shown and includes:

[0094] S301. For each target cultural relic content, determine each neighbor node of the target cultural relic content in the current cultural relic spatio-temporal graph.

[0095] Among them, the neighbor nodes of the target cultural relic content refer to other nodes directly connected to the node of the target cultural relic content, which reflect the situation where the user transfers from other nodes to the node of the target cultural relic content, or transfers from the node of the target cultural relic content to other nodes.

[0096] S302. Calculate the normalization coefficients between the target cultural relic content and each neighbor node.

[0097] In order to balance the contributions between different nodes and prevent some nodes from having too much influence on the results due to having too many neighbors, the contributions are restricted by calculating the normalization coefficients.

[0098] Specifically, for the node i of the target cultural relic content and a neighbor node j respectively, that is, for each pair of nodes (i, j), the normalization coefficient is specifically calculated as follows:

[0099]

[0100] Among them, N(i) is the set of neighbor nodes of the node i of the target cultural relic content. N(j) is the set of neighbor nodes of a neighbor node j of the target cultural relic content.

[0101] S303. Determine the current weight matrix corresponding to the target cultural relic content.

[0102] In order to consider the influence degree of different target cultural relic contents on the user's interests, it is necessary to determine the current weight matrix corresponding to the target cultural relic content. Optionally, the current weight matrix corresponding to the target cultural relic content is usually randomly initialized or obtained through a pre-trained model. The learning of the weight matrix can be optimized through the backpropagation algorithm.

[0103] S304. Based on the normalization coefficients between the target cultural relic content and each neighbor node, the node feature vector of the target cultural relic content, and the current weight matrix corresponding to the target cultural relic content, obtain the current weighted sum of the target cultural relic content.

[0104] Specifically, multiply the node feature vector of the current target cultural relic content by the normalization coefficients between the target cultural relic content and each neighbor node and the corresponding current weight matrix W (l) , and then perform accumulation to obtain the current weighted sum of the target cultural relic content. Specifically, it can be expressed as:

[0105]

[0106] Among them, is the node feature vector of the current target cultural relic content.

[0107] S305. Calculate using the current weighting of the target cultural relic content, a preset bias term, and a preset activation function to obtain the node feature vector of the latest current target cultural relic content.

[0108] Specifically, add the current weighting sum of the target cultural relic content and the preset bias term b (l) , then we can get:

[0109]

[0110] Finally, apply the preset activation function σ, then we can get the node feature vector of the new current target cultural relic content:

[0111]

[0112] Optionally, specifically, feature extraction can be performed through a spatio-temporal convolutional network. Specifically, when processing the current cultural relic spatio-temporal graph of the target user through the convolutional layer of the spatio-temporal convolutional network, the above-mentioned processing process is realized. And, the spatio-temporal convolutional network can include multiple convolutional layers, that is, the above can be repeatedly executed through multiple convolutional layers.

[0113] Therefore, after capturing the temporal dependence and spatial dependence of the user's behavior through the spatio-temporal graph convolutional layer, the updated feature vector, compared with the original feature vector, adds the temporal and spatial information of the user's recent behavior and can more accurately reflect the user's interest changes. For example, if the user has browsed the text description page of Song Dynasty celadon multiple times in the past week and watched a teaching video a few days ago, then through spatio-temporal feature extraction, the system will generate a comprehensive feature vector, that is, the updated feature vector, reflecting the user's long-term interest in Song Dynasty celadon.

[0114] S102. For each target cultural relic content, combine the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user to obtain the real-time feature vector corresponding to the target cultural relic content.

[0115] Although the node feature vector of the current target cultural relic content is obtained based on recent data analysis, it is still obtained based on historical behavior data. Therefore, in order to analyze the user's interest in real time according to the user's current behavior, so as to more accurately reflect the current user's interest and make the final recommendation result more accurate. Therefore, capture the user's latest behavior data in real time, combine its corresponding feature vector with the node feature vector of the current target cultural relic content, and adjust the feature vector in real time to ensure that the recommended content based on the adjusted feature vector is closer to the user's current interest.

[0116] Optionally, in another embodiment of the present application, a specific implementation manner of step S102 includes:

[0117] Generate a real-time behavior feature vector of the target user based on the real-time behavior data of the target user, and then, for each target cultural relic content, obtain the real-time feature vector corresponding to the target cultural relic content through the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user.

[0118] Specifically, generate a real-time behavior feature vector of the target user based on the real-time behavior data of the target user, and then, for each target cultural relic content, multiply the difference between the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user by the learning rate, and add the obtained product to the node feature vector of the current target cultural relic content to obtain the real-time feature vector corresponding to the target cultural relic content.

[0119] Therefore, specifically, when a new behavior occurs to the target user, the system immediately updates the corresponding node feature vector in the following manner to obtain the real-time feature vector corresponding to the target cultural relic content:

[0120]

[0121] Among them, represents the real-time feature vector corresponding to the target cultural relic content, that is, the updated node feature vector finally obtained. x r is the real-time behavior feature vector of the target user generated based on the real-time behavior data. η is the learning rate, which controls the update speed.

[0122] And the operation of "x r -h i " is actually calculating the difference between the feature vector x r of the behavior data and the existing feature vector h i . This difference can be understood as the change in user interest or the transition vector from the existing state to the new state. In other words, the result of this operation is the incremental information brought by the real-time behavior. η(x r -h i ) represents the degree of adjusting the existing feature vector according to the user's latest behavior data, and the learning rate η determines how much to trust the new observation value x r , and accordingly adjusts the existing feature vector. Multiply this difference by the learning rate η and add it to the original feature vector h i to obtain the updated feature vector , which represents a state closer to the user's current interest after combining the latest user behavior.

[0123] Because the feature vector h iIt is based on the user's historical behavior data and reflects the user's comprehensive interests over a period of time. The subsequent dynamic feature update is to capture the user's latest behavior and ensure that the recommended content is more in line with the user's current interests, that is, to combine the user's real-time behavior data to generate the final feature vector . Therefore, these two complement each other, enabling the recommendation system to more intelligently understand the user's behavior patterns and interest changes, and provide more accurate and personalized recommended content. In short, the dynamic feature update is to ensure the timeliness and relevance of the recommendation system, further enhance the user experience, and adapt to various changes in the user's interests

[0124] For example, in the scenario of completely retaining the original recommendations: The user has mainly browsed content related to Song Dynasty celadon in the past month, including text descriptions, pictures, and videos. In recent days, the user's behavior has not changed significantly, and they continue to focus on Song Dynasty celadon. So the user's initial interest feature vector h i indicates a high degree of attention to Song Dynasty celadon. The system generates a new real-time feature vector x r based on the user's latest behavior, capturing the user's continued interest in Song Dynasty celadon. Calculate the small difference between x r and the existing feature vector h i , and update it using a moderate learning rate η. So when presenting the recommendation results, the system continues to recommend content related to Song Dynasty celadon, such as newly discovered celadon objects and expert interpretation articles, ensuring that the recommended content is highly consistent with the user's current interests

[0125] For the scenario of completely switching to new recommendations: The user has mainly browsed various content related to Song Dynasty celadon in the past month, including text descriptions, pictures, and videos. However, in recent days, the user's behavior has changed significantly, and they have developed a strong interest in Tang Dynasty figurines, continuously viewed several articles about Tang Dynasty figurines, watched related archaeological excavation videos, and even participated in an online discussion forum. So the user's initial interest feature vector h i reflects a high degree of attention to Song Dynasty celadon. The system generates a new real-time feature vector x r based on the user's latest behavior, capturing the user's new interest in Tang Dynasty figurines. Calculate the significant difference between x r and the existing feature vector h i , and update it using a higher learning rate η to quickly adapt to the change in the user's interests. So when presenting the recommendation results, the system quickly adjusts the recommended content, focusing more on Tang Dynasty figurines and their related materials, reflecting the significant shift in the user's interests. This approach ensures that the recommended content is close to the user's immediate interests and provides a more personalized and timely experience

[0126] For a scenario with a gradual change in interest between the above two cases: The user has mainly browsed content related to celadon wares of the Song Dynasty in the past month, including written descriptions, pictures, and videos. In recent days, the user has started to show interest in Ru kiln porcelain of the Song Dynasty, viewing several related articles and high-definition pictures and videos. Although the interest has expanded, it still focuses on the field of Song Dynasty porcelain. So the user's initial interest feature vector h i reflects a high degree of attention to celadon wares of the Song Dynasty. The system generates a new real-time feature vector x r based on the user's latest behavior, capturing the user's new interest in Ru kiln porcelain of the Song Dynasty. Calculate the medium difference between x r and the existing feature vector h i and update it using a moderate learning rate η. Therefore, when making recommendations, the system continues to recommend content related to celadon wares of the Song Dynasty while gradually introducing information about Ru kiln porcelain of the Song Dynasty. This way not only retains the recommended content familiar to the user but also gradually explores the user's new interest points, ensuring the continuity and freshness of the recommended content.

[0127] S103. Weightedly fuse the real-time feature vector corresponding to the target cultural relic content with the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content.

[0128] The real-time feature vector corresponding to the target cultural relic content analyzed can accurately reflect the cultural relic content that the user is interested in. In order to accurately make recommendations based on the inherent connections of cultural relics and enrich the recommended content. Therefore, in the embodiment of the present application, a full-scale cultural relic knowledge graph is pre-constructed. The knowledge graph records the detailed information of each cultural relic and the associations between each cultural relic. Therefore, weightedly fuse the feature vectors of each cultural relic related to the target cultural relic content in the cultural relic knowledge graph. Among them, the feature vector of the cultural relic can be generated according to the information recorded in the knowledge graph. Specifically, the feature vector of the cultural relic corresponding to the target cultural relic content can be combined with the real-time feature vector corresponding to the target cultural relic content according to a certain ratio. So this process is not limited to the user's current interest points but can also be extended to relevant literature, videos, and other background information, thereby enhancing the depth and breadth of the recommended content.

[0129] Optionally, in another embodiment of the present application, the method for constructing a cultural relic knowledge graph includes:

[0130] Define each cultural relic and each cultural relic category as entities in the knowledge graph, add the detailed information of each entity, and connect the entity of each cultural relic with the entity of the cultural relic category to which it belongs, and connect it with the entities of other cultural relics belonging to the same cultural relic category to obtain the cultural relic knowledge graph.

[0131] Among them, the categories of cultural relics can specifically include dynasties, cultural backgrounds, relevant documents, etc. Therefore, in the embodiments of the present application, cultural relics, dynasties, cultural backgrounds, relevant documents, etc. are defined as entities in the cultural relic knowledge graph. Therefore, the feature vector of the cultural relic corresponding to the target cultural relic content is generated not only by using the information of the cultural relic involved in the target cultural relic content, but also by using the information of entities such as the dynasties, cultural backgrounds, and relevant documents related to the cultural relic. Therefore, it is also possible to enrich the recommended content based on dynasties, cultural backgrounds, relevant documents, etc. Moreover, detailed attribute information is added, such as age, material, discovery location, historical background, etc. Then, the associated entities are connected, that is, the relationships between the entities are defined, such as "belonging to the same dynasty", "citing the same document", "appearing at the same location".

[0132] In this way, the cultural relic knowledge graph can reveal the complex semantic relationships between cultural relics, helping users understand the stories and meanings behind the cultural relics. This deep-level association is based on detailed research and professional knowledge, rather than simply relying on user behavior data. Moreover, compared with the cultural relic spatio-temporal map, the content of the cultural relic knowledge graph is more stable because it is based on the inherent attributes and historical facts of the cultural relics themselves, rather than instantaneous user behavior changes. Therefore, it can serve as a solid foundation for the recommendation system, maintaining a certain degree of coherence even when users' interests change significantly.

[0133] Among them, the data of the cultural relic knowledge graph can be sourced from professional archaeological materials, historical documents, and other authoritative resources, emphasizing the accuracy and depth of knowledge. The data of the cultural relic spatio-temporal map comes from users' historical behavior logs and real-time interaction records, emphasizing the immediacy and dynamics of user behavior. The cultural relic knowledge graph focuses on the inherent attributes and complex semantic relationships of the cultural relics themselves, providing in-depth background information and support, so it is mainly used to enrich the recommended content, increase the depth and breadth of recommendations, and support long-term and stable content recommendations. The cultural relic spatio-temporal map focuses on user behavior patterns and dependencies in time and space, capturing users' immediate interests and behavior changes, and is mainly used to generate user interest vectors, achieve precise personalized recommendations, and make dynamic adjustments according to user behavior. Therefore, although both the cultural relic knowledge graph and the cultural relic spatio-temporal map involve cultural relics and their attributes, their roles and functions are complementary, jointly constituting a complete personalized recommendation system. The cultural relic knowledge graph provides a solid knowledge foundation, while the cultural relic spatio-temporal map enhances the flexibility and response speed of the system.

[0134] Optionally, in another embodiment of the present application, a specific implementation manner of step S103 includes:

[0135] The real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph are weighted and calculated according to the fusion weight to obtain the initial recommended content feature corresponding to the target cultural relic content.

[0136] Specifically, the initial recommended content features corresponding to the target cultural relic content can be expressed as:

[0137]

[0138] where k i is the feature vector of the cultural relic corresponding to the target cultural relic content in the cultural relic knowledge graph. α is the fusion weight.

[0139] Based on the fused features, more abundant content can be recommended to users, and multi-model recommendations can be achieved. For example, assuming that a user is interested in celadon wares of the Song Dynasty, the system will generate the fused recommended content features according to the user's interest vector, that is, the real-time feature vector corresponding to the target cultural relic content and the relevant entities (celadon wares of the Song Dynasty) in the cultural relic knowledge graph.

[0140] S104. Combine the initial recommended content features corresponding to the target cultural relic content with the context features of the currently viewed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content.

[0141] In order to further improve the accuracy of recommendations, in the embodiments of the present application, not only the current behavior of the target user is considered, but also the content currently viewed by the target user needs to be considered. Therefore, in the present application, the context features of the currently viewed content are fused into the features, such as the features of relevant literature and videos of the currently viewed content, so that the content recommended based on the features is closer to the current user's interests. For example, when a user is viewing pictures of celadon wares of the Song Dynasty, the system can recommend relevant literature and video materials, such as research papers on celadon wares of the Song Dynasty and teaching videos on the production process.

[0142] Optionally, in another embodiment of the present application, a specific implementation manner of step S104 includes:

[0143] Weight the initial recommended content features corresponding to the target cultural relic content and the context features of the currently viewed content in the cultural relic knowledge graph according to the context weight to obtain the final recommended content features corresponding to the target cultural relic content.

[0144] Therefore, the specific calculation method of the final recommended content features corresponding to the target cultural relic content is:

[0145]

[0146] where c iis the context feature of the current viewed content. β is the context weight, which is determined according to the type of the current viewed content and / or the key information in the current viewed content. For example, if the user is currently viewing a picture of a cultural relic, research papers on the cultural relic and teaching videos on its production process may be given higher weights. It can also be based on the detailed matching of the current viewed content. For example, if the user is viewing the discovery location information of a celadon of the Song Dynasty, archaeological reports or on-site inspection videos of the location may receive higher weights.

[0147] S105. Based on the final recommended content features corresponding to the target cultural relic content, match the multi-modal current recommended cultural relic content and push it to the target user.

[0148] Since the final recommended content features corresponding to the target cultural relic content not only consider the user's recent behaviors and related cultural relics, but also consider the current behavior and the context of the current behavior, based on the final recommended content features corresponding to the target cultural relic content, multi-modal cultural relic content that meets the user's current interests can be accurately matched and recommended to the user.

[0149] Optionally, specifically, it can be by calculating the similarity between each cultural relic content and the final recommended content features, matching the content that meets the current interests of the target user, and then based on the richness of the content, screening a certain amount of different-modal content from the matched content for recommendation as the current recommended cultural relic content and pushing it to the target user.

[0150] Another embodiment of the present application provides a device for processing cultural relic content information, as Figure 4 shown, including:

[0151] The node feature update unit 401 is used to perform spatio-temporal feature extraction on the current cultural relic spatio-temporal graph of the target user to obtain the node feature vectors of each current target cultural relic content.

[0152] Among them, the historical behavior data of the target user on each target cultural relic content and the attributes of each target cultural relic content are recorded in the current cultural relic spatio-temporal graph.

[0153] The behavior feature combination unit 402 is used to combine the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user for each target cultural relic content to obtain the real-time feature vector corresponding to the target cultural relic content.

[0154] The cultural relic feature fusion unit 403 is used to perform weighted fusion on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content features corresponding to the target cultural relic content.

[0155] A context feature combination unit 404 is configured to combine the initial recommended content features corresponding to the target cultural relic content with the context features of the currently browsed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content.

[0156] A recommendation unit 405 is configured to match the multimodal currently recommended cultural relic content based on the final recommended content features corresponding to the target cultural relic content and push it to the target user.

[0157] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, it further includes:

[0158] A data acquisition unit is configured to acquire the historical behavior data of the latest target user.

[0159] A node creation unit is configured to create a node corresponding to the target cultural relic content based on the attributes of the target cultural relic content for the target cultural relic content with generated behaviors recorded in the historical behavior data of the target user.

[0160] A vector initialization unit is configured to initialize the node feature vector of the target cultural relic content according to the attributes of the target cultural relic content.

[0161] A node connection unit is configured to connect the nodes corresponding to each target cultural relic content in the order of behaviors in the historical behavior data of the target user, and record the behavior time on the connected edges to obtain the current cultural relic spatio-temporal graph.

[0162] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, the node feature update unit includes:

[0163] A neighbor node determination unit is configured to determine each neighbor node of the target cultural relic content in the current cultural relic spatio-temporal graph for each target cultural relic content respectively.

[0164] A coefficient calculation unit is configured to calculate the normalization coefficients between the target cultural relic content and each neighbor node.

[0165] A weight determination unit is configured to determine the current weight matrix corresponding to the target cultural relic content.

[0166] An accumulation unit is configured to obtain the current weighted sum of the target cultural relic content based on the normalization coefficients between the target cultural relic content and each neighbor node, the node feature vector of the target cultural relic content, and the current weight matrix corresponding to the target cultural relic content.

[0167] An update unit is configured to calculate through the current weighting of the target cultural relic content, a preset bias term, and a preset activation function to obtain the latest node feature vector of the current target cultural relic content.

[0168] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, the behavior feature combination unit includes:

[0169] A behavior feature generation unit, configured to generate a real-time behavior feature vector of a target user based on the real-time behavior data of the target user.

[0170] A node feature weighting unit, configured to, for each target cultural relic content, multiply the difference between the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user by a learning rate, and add the obtained product to the node feature vector of the current target cultural relic content to obtain a real-time feature vector corresponding to the target cultural relic content.

[0171] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, it further includes:

[0172] An entity creation unit, configured to define each cultural relic and each cultural relic category as entities in a knowledge graph, and add detailed information of each entity.

[0173] An entity connection unit, configured to connect the entity of each cultural relic with the entity of the cultural relic category to which it belongs, and connect with the entities of other cultural relics belonging to the same cultural relic category to obtain a cultural relic knowledge graph.

[0174] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, the cultural relic feature fusion unit includes:

[0175] A cultural relic feature weighting unit, which performs weighted calculation on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph according to a fusion weight to obtain an initial recommended content feature corresponding to the target cultural relic content.

[0176] Optionally, in the cultural relic content information processing device provided in another embodiment of the present application, the context feature combination unit includes:

[0177] A context feature combination subunit, configured to perform weighting on the initial recommended content feature corresponding to the target cultural relic content and the context feature of the current browsing content in the cultural relic knowledge graph according to a context weight to obtain a final recommended content feature corresponding to the target cultural relic content. Wherein, the context weight is determined according to the type of the current browsing content and / or key information in the current browsing content.

[0178] It should be noted that for the specific working processes of the respective units provided in the above embodiments of the present application, reference may be made correspondingly to the implementation processes of the corresponding steps in the above method embodiments, which will not be elaborated herein.

[0179] Another embodiment of the present application provides an electronic device, such as Figure 5As shown, it includes:

[0180] A memory 501 and a processor 502.

[0181] Among them, the memory 501 is used to store programs.

[0182] The processor 502 is used to execute the program stored in the memory 501. When the program is executed, it is specifically used to implement the cultural relic content information processing method provided in any one of the above embodiments.

[0183] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed by a processor, it is used to implement the cultural relic content information processing method provided in any one of the above embodiments.

[0184] The computer storage medium includes permanent and non-permanent, removable and non-removable media. Information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0185] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0186] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for processing cultural relics content information, characterized in that: include: Extracting spatiotemporal features from the target user's current spatiotemporal graph of cultural relics to obtain node feature vectors of each current target cultural relic content; wherein the current spatiotemporal graph of cultural relics records the target cultural relic content on which the target user has recently performed actions, the type and time of the actions performed, and the order of behavior changes between each target cultural relic content; the node feature vectors of the target cultural relic content are used to reflect the current interests of the target user; For each target cultural relic content, the node feature vector of the current target cultural relic content is combined with the real-time behavior feature vector of the target user to obtain a real-time feature vector corresponding to the target cultural relic content; Performing weighted fusion on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content; Combining the initial recommended content features corresponding to the target cultural relic content with the context features of the currently browsed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content; Based on the final recommended content features corresponding to the target cultural relic content, multimodal current recommended cultural relic content is matched and pushed to the target user.

2. The method according to claim 1, characterized in that Before extracting the spatiotemporal features of the target user's current spatiotemporal graph of cultural relics to obtain the node feature vectors of the current contents of each target cultural relic, the method further includes: Obtain the latest historical behavior data of the target user; For the target cultural relic content that generates the behavior recorded in the historical behavior data of the target user, creating a node corresponding to the target cultural relic content based on the attributes of the target cultural relic content; Initializing a node feature vector of the target cultural relic content according to the attributes of the target cultural relic content; According to the behavior sequence in the historical behavior data of the target user, the nodes corresponding to the target cultural relic contents are connected, and the behavior time is recorded on the connected edges to obtain the current cultural relic spatiotemporal graph.

3. The method according to claim 1, characterized in that: The step of extracting spatiotemporal features from the current spatiotemporal graph of cultural relics of the target user to obtain node feature vectors of the current contents of each target cultural relic includes: For each target cultural relic content, determine each neighbor node of the target cultural relic content in the current cultural relic spatiotemporal graph; Calculating the normalization coefficients of the target cultural relic content and each of the neighboring nodes; Determining a current weight matrix corresponding to the target cultural relic content; Based on the normalization coefficients of the target cultural relic content and each of the neighboring nodes, the node feature vector of the target cultural relic content, and the current weight matrix corresponding to the target cultural relic content, a current weighted sum of the target cultural relic content is obtained; The latest node feature vector of the current target cultural relic content is obtained by calculating the current weight of the target cultural relic content, the preset bias term and the preset activation function.

4. The method according to claim 1, characterized in that The step of combining the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user for each target cultural relic content to obtain a real-time feature vector corresponding to the target cultural relic content includes: Generating a real-time behavior feature vector of the target user based on the real-time behavior data of the target user; For each target cultural relic content, a real-time feature vector corresponding to the target cultural relic content is obtained through the node feature vector of the current target cultural relic content and the real-time behavior feature vector of the target user.

5. The method according to claim 1, characterized in that Also includes: Define each cultural relic and each cultural relic category as an entity in the knowledge graph, and add detailed information for each of the entities; The entities of each of the cultural relics are connected with the entities of the cultural relic category to which they belong, and with the entities of other cultural relics belonging to the same cultural relic category, to obtain a cultural relic knowledge graph.

6. The method according to claim 1, characterized in that The step of weighting and fusing the real-time feature vector corresponding to the target cultural relic content with the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph to obtain the initial recommended content feature corresponding to the target cultural relic content includes: The real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph are weightedly calculated according to the fusion weight to obtain the initial recommended content features corresponding to the target cultural relic content.

7. The method according to claim 1, characterized in that The combining of the initial recommended content features corresponding to the target cultural relic content with the context features of the currently browsed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content includes: The initial recommended content features corresponding to the target cultural relic content and the context features of the currently browsed content in the cultural relic knowledge graph are weighted according to the context weight to obtain the final recommended content features corresponding to the target cultural relic content; wherein the context weight is determined according to the type of the currently browsed content and / or the key information in the currently browsed content.

8. A cultural relic content information processing device, characterized in that: include: A node feature updating unit is used to extract the spatiotemporal features of the current spatiotemporal graph of cultural relics of the target user to obtain the node feature vectors of each current target cultural relic content; wherein the current spatiotemporal graph of cultural relics records the target cultural relic content on which the target user has recently performed actions, the type and time of the actions performed, and the order of behavior changes between each target cultural relic content; the node feature vectors of the target cultural relic content are used to reflect the current interests of the target user; A behavior feature combining unit is used to combine the node feature vector of the current target cultural relic content with the real-time behavior feature vector of the target user for each target cultural relic content, so as to obtain a real-time feature vector corresponding to the target cultural relic content; A cultural relic feature fusion unit, used to perform weighted fusion on the real-time feature vector corresponding to the target cultural relic content and the feature vector of the cultural relic corresponding to the target cultural relic content in the pre-constructed cultural relic knowledge graph, to obtain the initial recommended content feature corresponding to the target cultural relic content; A context feature combining unit, used to combine the initial recommended content features corresponding to the target cultural relic content with the context features of the currently browsed content in the cultural relic knowledge graph to obtain the final recommended content features corresponding to the target cultural relic content; The recommendation unit is used to match the multimodal current recommended cultural relic content based on the final recommended content features corresponding to the target cultural relic content and push it to the target user.

9. An electronic device, characterized in that: include: Memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the cultural relics content information processing method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, is used to implement the cultural relics content information processing method as described in any one of claims 1 to 7.

Images