Digital cultural transmission content recommendation method and system based on knowledge graph

Abstract

The invention relates to the technical field of digital content recommendation, and discloses a digital cultural transmission content recommendation method and system based on a knowledge graph, and the method comprises the steps: constructing a space-time perception knowledge representation model, and introducing a space-time marked tetrad knowledge representation structure; a culture event detection model based on multi-source data fusion is constructed, and real-time detection of culture events is realized; constructing a new culture content discovery and value evaluation model, and automatically identifying and evaluating the emerging culture content; realizing a knowledge graph dynamic updating and expansion algorithm, and responding to cultural events and new contents; and constructing a timeliness-perceived adaptive recommendation algorithm, and balancing the recommendation of hot contents and novel contents. According to the invention, accurate, timely and innovative recommendation of the digital culture content is realized.

Description

Technical Field

[0001] The present invention relates to the field of digital content recommendation technology, and more specifically, to a method and system for recommending digital cultural communication content based on a knowledge graph. Background Art

[0002] With the rapid development of digital technology, digital cultural content continues to be produced and evolved at an unprecedented rate. Major cultural events (such as exhibitions, festivals, and commemorative events) often trigger a surge in interest in related cultural content, while emerging cultural content continues to emerge. In this context, how to effectively recommend massive amounts of digital cultural content to target users has become a major technical issue in the field of digital cultural communication.

[0003] Traditional content recommendation systems are primarily based on collaborative filtering or content-based recommendation methods. These methods perform well when dealing with static content collections, but they have significant limitations when dealing with digital cultural content that exhibits distinct spatiotemporal evolution and event-driven characteristics. While current mainstream knowledge graph-based recommendation systems incorporate domain knowledge, their knowledge representations often utilize static triple structures, which are unable to effectively capture the evolution and development trends of cultural content across different spatiotemporal contexts. Consequently, these recommendation systems lack an accurate grasp of cultural dynamics.

[0004] Furthermore, traditional knowledge graph update mechanisms are mostly based on fixed cycles or cumulative change thresholds, lacking the ability to respond sensitively to cultural events and timely update relevant knowledge and provide recommendation services when cultural hotspots emerge. Existing cultural content discovery methods primarily rely on manual intervention or simple similarity matching, lacking the ability to automatically identify and assess the value of new cultural content, making it difficult to promptly incorporate emerging cultural content into recommendation systems. Furthermore, recommendation algorithms struggle to strike a balance between exploration and exploitation, either being overly conservative, resulting in a single recommended content, or being too random, resulting in insufficient relevance of recommended content, impacting user acceptance and satisfaction with emerging cultural content. Summary of the Invention

[0005] The present invention provides a method and system for recommending digital cultural communication content based on knowledge graphs, which solves the technical problem in the existing technology that it is impossible to effectively perceive and respond to the spatiotemporal evolution characteristics and event-driven features of cultural content.

[0006] The first aspect of the present invention discloses a method for recommending digital cultural communication content based on a knowledge graph, comprising: Build a spatiotemporal knowledge representation model, introduce spatiotemporal tags to construct a quadruple knowledge representation structure, and realize the spatiotemporal awareness of the knowledge graph; Based on the knowledge representation model, a cultural event detection model integrating multi-source data is constructed to collect multi-source data and extract time series features, calculate the suddenness index, and detect cultural events; Based on the detected cultural events, a new cultural content discovery and value assessment model is constructed to extract content features for multi-dimensional evaluation, calculate novelty and value scores, and screen high-value content; Based on the detected cultural events and high-value content, the knowledge graph is dynamically updated and expanded by the algorithm to update the knowledge graph; Based on the updated knowledge graph, a timeliness-aware adaptive recommendation algorithm is constructed, which combines the hotspot decay model and the exploration and utilization balance strategy to generate a recommendation list.

[0007] Furthermore, the construction of the spatiotemporal-aware knowledge representation model includes: constructing a spatiotemporal-labeled quadruple knowledge representation structure; defining a spatiotemporal-labeled structure; constructing spatiotemporal-aware entity and relationship representations; and implementing spatiotemporal propagation rules.

[0008] Furthermore, the construction of a multi-source data fusion cultural event detection model includes: multi-source data collection and preprocessing; construction of a temporal feature extraction model for cultural events; implementation of a suddenness index calculation model; and construction of a multi-source fusion event detection model.

[0009] Furthermore, the construction of a new cultural content discovery and value assessment model includes: constructing a cultural content feature representation model; implementing a novelty measurement model; constructing a cultural value assessment model; and constructing a comprehensive screening model.

[0010] Furthermore, the algorithm for implementing dynamic update and expansion of knowledge graph includes: constructing an event-related knowledge subgraph recognition algorithm; implementing an event-based knowledge update algorithm; constructing a new content knowledge graph mapping model; implementing an automatic connection algorithm for graph structure prediction; and updating the spatiotemporal evolution model.

[0011] Furthermore, the construction of a timeliness-aware adaptive recommendation algorithm includes: constructing a hotspot decay model; implementing a content relevance calculation model based on a knowledge graph; constructing an exploration and utilization balance strategy model; implementing a comprehensive recommendation score calculation model; and constructing a final recommendation list generation algorithm.

[0012] Furthermore, the construction of the cultural value assessment model includes: establishing cultural content quality assessment indicators, combining expert evaluation and user feedback; constructing a cultural content influence assessment model, analyzing the scope and speed of dissemination; implementing a cultural content sentiment analysis model, evaluating sentiment tendencies and intensity; establishing a cultural content historical value assessment model, analyzing the correlation with historical culture.

[0013] Furthermore, the automatic connection algorithm for implementing graph structure prediction includes: learning the association pattern between entities based on the existing knowledge graph structure; using graph neural networks to predict the potential relationship between new entities and existing entities; calculating the relationship confidence and setting a threshold to filter high-confidence relationships; and verifying the consistency of the newly added relationships through knowledge reasoning.

[0014] Furthermore, the construction of the exploration and utilization balance strategy model includes: designing a dynamic exploration rate adjustment mechanism to adaptively adjust according to user feedback and content characteristics; implementing a multi-armed bandit algorithm based on Thompson sampling to balance the recommendation ratio of hot content and novel content; building a user interest drift detection model to timely adjust the exploration and utilization strategy; and designing a diversity guarantee mechanism to ensure the content diversity of the recommendation list.

[0015] The second aspect of the present invention discloses a digital cultural communication content recommendation system based on a knowledge graph, which is used to execute the above-mentioned digital cultural communication content recommendation method based on a knowledge graph, including: The spatiotemporal knowledge representation module is used to construct a spatiotemporal labeled quadruple knowledge representation structure, define the structure of spatiotemporal labels, build spatiotemporal aware entity and relationship representations, and implement spatiotemporal propagation rules; The cultural event detection module of multi-source data fusion is used to collect and preprocess multi-source data, build a temporal feature extraction model for cultural events, implement a suddenness index calculation model, and build a multi-source fusion event detection model; New cultural content discovery and value assessment module, used to build a cultural content feature representation model, implement a novelty measurement model, build a cultural value assessment model, and build a comprehensive screening model; The knowledge graph dynamic update and expansion module is used to build an event-related knowledge subgraph recognition algorithm, implement an event-based knowledge update algorithm, build a new content knowledge graph mapping model, implement an automatic connection algorithm for graph structure prediction, and update the spatiotemporal evolution model; The timeliness-aware adaptive recommendation module is used to build a hotspot decay model, implement a content relevance calculation model based on the knowledge graph, build an exploration and utilization balance strategy model, implement a comprehensive recommendation score calculation model, and build a final recommendation list generation algorithm.

[0016] The beneficial effects of the present invention are as follows: the event-driven knowledge graph dynamic update mechanism and hot spot attenuation model are adopted, so that the recommendation system can timely capture hot events in the cultural field, quickly update relevant knowledge, and provide users with content recommendations related to the current cultural background and hot spots; through the new cultural content discovery and value evaluation model, and the exploration and utilization balance strategy, the system can automatically identify and evaluate emerging cultural content, and reasonably introduce these contents in the recommendation process to provide users with more emerging and valuable cultural content; based on the knowledge representation model of time and space perception and the capture of time and space evolution laws, the system can understand the evolution trend of cultural content, predict possible hot spots and directions in the future, and help users grasp the trend of cultural development; by balancing the relevance, popularity and novelty of recommended content, and dynamically adjusting the ratio of exploration and utilization, the system can guide users to discover new valuable content while satisfying their current interests; through the dual-driven knowledge graph dynamic update and expansion, the system can continuously enrich and optimize the knowledge graph, improve its comprehensiveness and timeliness, and provide a better knowledge base for various cultural content recommendation tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a method for recommending digital cultural communication content based on knowledge graphs provided by the present invention. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a method for recommending digital cultural communication content based on knowledge graph, such as Figure 1 As shown, the following steps are included: Step 1: Build a spatiotemporal knowledge representation model. By introducing spatiotemporal tags to construct a quadruple knowledge representation structure, the spatiotemporal awareness of the knowledge graph can be realized. The specific implementation is as follows: Step 1.1: Construct a spatiotemporal labeled quadruple knowledge representation structure.

[0020] Traditional knowledge graphs use triples represents knowledge, where Represents the head entity, Indicates relationship, Represents the tail entity. This application expands it into a four-tuple representation: ; in, Represents a collection of entities, Represents a set of relations, represents a set of spatiotemporal tags, Represents a spatiotemporal tag, containing information in both time and space dimensions. represents the knowledge graph, Represents the head entity, Indicates relationship, Represents the tail entity, Represents space-time markers, Indicates that the conditions are met. Indicates belonging to a set.

[0021] Step 1.2: Define the structure of the spatiotemporal marker.

[0022] Time and space markers Defined as a two-tuple: ; in, Indicates a timestamp, which can be a precise time point or a time interval ; Indicates geographic location information, which can be precise coordinates , or a region or geographic entity name.

[0023] Step 1.3: Construct spatiotemporal-aware entity and relationship representations.

[0024] Combining the vector representation of entities and relationships with spatiotemporal information to form a spatiotemporal-aware representation method: ; ; in, Representing an entity The spatiotemporal perception vector representation of Representing relationships The spatiotemporal perception vector representation of and Represent entities separately and relationships The basis vector representation of and Represents and entities respectively and relationships Associated spatiotemporal markers and The vector representation of Representation and Entity associated spatiotemporal tags (containing temporal and spatial information), Representation and Relationship associated spatiotemporal tags (containing temporal and spatial information), Represents a vector concatenation operation.

[0025] In practical applications, pre-trained language models (such as BERT, RoBERTa, etc.) can be used to obtain basic vector representations, and spatiotemporal vector representations can be obtained through time encoding and space encoding. Time encoding can use a periodic encoding method: ; in, represents the time encoding vector elements, represents the encoding dimension, Represents a timestamp (convertible to a numeric value), represents the index of a vector element, represents the sine function, Represents the cosine function. Spatial encoding can use geographic coordinate embedding or region embedding methods.

[0026] Step 1.4: Implement the spatiotemporal propagation rules.

[0027] Define the rules for the dissemination of knowledge in the spatial and temporal dimensions, and capture the temporal and spatial evolution of cultural content: ; in, Represents a given header entity ,relation and its space-time marking Under the condition of The probability distribution of is the spatiotemporal propagation function, which is used to model the propagation law of spatiotemporal characteristics in the knowledge graph. Represents the header entity The spatiotemporal perception vector representation of Representing relationships The spatiotemporal perception vector representation of represents the spatiotemporal tag of the head entity, Represents the spatiotemporal tag of the tail entity.

[0028] Space-time propagation function This can be achieved through graph neural networks, such as using spatiotemporal graph attention networks: ; in, represents the space-time propagation function, represents the activation function, represents the weight matrix, represents the bias vector, The concatenation of the vectors representing the head entity, relation, and spatiotemporal tags. represents the vector representation of the head entity, A vector representation of the relationship, represents the spatiotemporal token vector representation of the head entity, represents the matrix multiplication operation, Represents vector addition operation.

[0029] In cultural communication scenarios, this function can be used to predict the spread of cultural activities. For example, if a Chinese opera performance first premieres in Beijing, this function can predict which cities it might tour to and when it will be performed in each city, thereby helping the recommendation system prepare relevant content recommendations in advance.

[0030] It should be noted that through the above steps, a knowledge graph representation model with time and space perception capabilities will be obtained. This model can effectively express the existence status and evolution laws of cultural content in different time and space backgrounds, laying the foundation for subsequent cultural event detection, dynamic updating of knowledge graphs and content recommendations.

[0031] Step 2: Based on the knowledge representation model, a cultural event detection model that integrates multi-source data is constructed. Multi-source data is collected and time series features are extracted. The suddenness index is calculated to detect cultural events.

[0032] The specific implementation is as follows: Step 2.1: Multi-source data collection and preprocessing.

[0033] Collect real-time data related to the cultural field from multiple data sources: ; in, Represents a collection of multi-source data, 、 、 The data sets for the first, second, and nth data sources are represented, respectively, where n represents the total number of data sources. These data sources can include social media data (such as Weibo and Twitter), news reports, user behavior data (such as clicks, favorites, and comments), and search query data. For each data source, clean, standardize, and extract features to convert it into a unified representation: ; in, Indicates the processed The feature set of a data source, Represents the preprocessing function, Indicates the A data collection from a data source.

[0034] The specific implementation of the preprocessing function includes the following steps: (1) word segmentation, stop word removal, entity recognition and other processing of text data; (2) feature extraction of image data to extract visual content features; (3) statistical aggregation of user behavior data to calculate indicators such as behavior frequency and conversion rate; (4) time alignment and format standardization of all data.

[0035] In practical applications, named entity recognition (NER) models can be used to identify cultural entities in text, such as cultural activity entities such as "Shanghai Museum Special Exhibition" and "Dunhuang Cultural Exhibition" in news reports. For multilingual data, multilingual pre-trained models can be used for unified representation, ensuring that cultural content in different languages can be compared within the same semantic space.

[0036] Step 2.2: Build a temporal feature extraction model for cultural events.

[0037] For the time series data of each data source, extract its time series features: ; in, Indicates the Data sources at time The eigenvector of represents the time series feature extraction function, Indicates the processed The feature set of a data source, Indicates the current time point, Indicates the time window size. Time series features can include data volume, growth rate, fluctuation, keyword distribution, etc.

[0038] The specific implementation of the time series feature extraction function can adopt a variety of time series analysis methods, including: (1) window aggregation calculation, such as the calculation of statistical indicators such as data volume and growth rate within the time window; (2) spectrum analysis, such as Fourier transform of data time series to extract periodic features; (3) trend analysis, such as using exponential smoothing and other methods to extract trend features; (4) semantic clustering, clustering analysis of content themes within the time window.

[0039] For example, for social media data, we can calculate the change curve of the mention volume of specific cultural topics or tags within 24 consecutive hours to identify peaks and abnormal growth points; for search query data, we can analyze the changes in the frequency of search terms related to specific cultural heritage and capture sudden increases in public interest.

[0040] Step 2.3: Implement the burst index calculation model.

[0041] Based on the time series characteristics, calculate the burstiness index of each data source: ; in, Indicates the Data sources at time The burst index, Indicates the Data sources at time The eigenvector of Indicates the past In the time window The mean of the data source characteristics, Indicates the past In the time window The standard deviation of the data source characteristics, Indicates the time window size.

[0042] The burst index can be calculated separately for different feature dimensions and then considered comprehensively. In addition, to improve the stability of the index, robust statistical methods can be used, such as using the median and interquartile range instead of the mean and standard deviation: ; in, Indicates the Data sources at time The robust burstiness index of Indicates the Data sources at time The eigenvector of Indicates the past In the time window The median of the data source characteristics, Indicates the past The time window The interquartile range (i.e., the difference between the upper quartile and the lower quartile) of the data source characteristics.

[0043] In the digital cultural communication scenario, this method can effectively detect the sudden popularity of cultural events such as "museum digital collection release" and "intangible cultural heritage skills exhibition and performance activities" to avoid being interfered by noise data.

[0044] Step 2.4: Build a multi-source fusion event detection model.

[0045] The burstiness index of each data source is weighted and integrated to generate a comprehensive burstiness index: ; in, represents the comprehensive suddenness index, Indicates the The weight of each data source satisfies , Indicates the Data sources at time The burst index, Indicates the total number of data sources. Indicates the current time point, represents the sum operation, Represents a multiplication operation. The weight can be dynamically adjusted based on the reliability, relevance, and timeliness of the data source.

[0046] Weight adjustment can be achieved through an adaptive mechanism: ; in, Indicates the The weight of the data source, Represents a data source The reliability score can be calculated based on historical accuracy, data freshness, coverage and other indicators. represents the base of natural logarithms, Indicates the total number of data sources. represents the sum operation, The index variable representing the sum.

[0047] At the same time, an event feature extraction model is built to extract the key features of events from burst data: ; in, represents the event feature vector, represents the event feature extraction function, Indicates time Data that is determined to be bursty.

[0048] Event feature extraction uses topic models (such as LDA) and keyword extraction algorithms (such as TextRank) to extract event topics, key entities, and descriptive features from burst data. These features can be used to classify events and assess their impact.

[0049] Finally, based on the event characteristics, determine the type, scope of impact, and duration of the event: ; in, Indicates event information, including event type Scope of impact and duration , represents the event classification function, Represents the event feature vector.

[0050] Event classification functions can be implemented using supervised learning methods, such as training a multi-label classification model using pre-labeled cultural event data to map event features to a predefined set of event types. The impact range can be determined through geographic entity recognition and user distribution analysis, while the duration can be predicted based on the persistence patterns of similar historical events.

[0051] In actual application scenarios, the model can detect cultural events such as "The digital exhibition of Dunhuang murals opened in Shanghai" and "The work of a certain intangible cultural heritage inheritor won an international award", and determine the event type (such as exhibition, award, commemoration, etc.), the scope of influence (such as national, regional, global, etc.) and the expected duration (such as one week, one month, etc.), providing event trigger signals for subsequent knowledge graph updates and content recommendations.

[0052] It should be noted that through the above steps, a multi-source data fusion event detection model will be obtained that can monitor and identify important events in the cultural field in real time. This model can capture the emergence and evolution of cultural hotspots and provide trigger signals for the dynamic update of the knowledge graph.

[0053] Step 3: Based on the detected cultural events, a new cultural content discovery and value assessment model is constructed to extract content features for multi-dimensional evaluation, calculate novelty and value scores, and screen high-value content; The specific implementation is as follows: Step 3.1: Construct a cultural content feature representation model.

[0054] For various cultural contents, extract their multi-dimensional features and form a unified feature representation: ; in, Representing cultural content The vector representation of Indicates the cultural content to be processed (such as articles, pictures, videos, audio and other digital cultural resources), Represents the content encoding function, which can be implemented based on deep learning models such as BERT and ResNet, targeting different types of cultural content such as text and images.

[0055] The specific implementation of the content encoding function can adopt a multimodal feature fusion method and select different feature extractors according to the content type: For text content (such as articles and comments): Use pre-trained language models (such as BERT, RoBERTa, etc.) to extract semantic features and output document-level representation vectors.

[0056] For image content (such as artworks and photos of cultural relics): Use pre-trained visual models (such as ResNet and ViT) to extract visual features and capture the visual elements and style characteristics of the image.

[0057] For audio content (such as folk music and opera excerpts): use audio feature extraction models (such as VGGish and wav2vec) to extract acoustic features and identify musical style and emotional characteristics.

[0058] For video content (such as dance performances and cultural documentaries): combine visual models and audio models, and use temporal aggregation methods (such as attention mechanisms and LSTM) to integrate multi-frame features.

[0059] For multimodal cultural content, the features of different modalities can be fused into a unified representation through a multimodal fusion model: ; in, Representing cultural content The multimodal fusion vector representation of represents the multimodal feature fusion function, represents the text modality feature vector, represents the image modality feature vector, represents the audio modal feature vector, Represents the video modality feature vector.

[0060] The fusion function can be implemented using an attention mechanism or an MLP network to dynamically adjust the weights of each modality according to the content characteristics.

[0061] In practical applications, for cultural relics exhibits in digital museums, their images, explanatory texts and voice explanations can be processed simultaneously to form a unified multimodal representation, enabling a more comprehensive understanding of the cultural connotations of the cultural relics.

[0062] Step 3.2: Implement the novelty measurement model.

[0063] Calculate the similarity between new content and content in the existing content library, and define the novelty index based on the maximum similarity: ; in, Display content The novelty index, Represents an existing content library, Represents the existing content in the library. Represents the similarity function, which can be cosine similarity, Euclidean distance, etc. Display content The vector representation of Display content The higher the novelty index, the more unique the content is and the greater the difference from existing content. max represents the maximum value operation.

[0064] To improve the accuracy of novelty calculation, locality sensitive hashing (LSH) or vector indexing technology can be used to accelerate similarity search and process large-scale content libraries; at the same time, a hierarchical similarity calculation strategy can be adopted to first screen potential similar content at a coarse-grained level, and then accurately calculate the similarity at a fine-grained level.

[0065] In addition, novelty can be evaluated from multiple dimensions, such as novelty of theme, novelty of expression, novelty of style, etc. ; in, Display content Multi-dimensional novelty scoring, 、 and Represent the weight coefficients of theme novelty, expression novelty and style novelty respectively, represents the topic novelty score, represents the novelty score of the expression form, represents the style novelty score.

[0066] In the digital cultural communication scenario, this multi-dimensional novelty assessment can more accurately identify innovative cultural content such as "innovative expressions that integrate traditional cultural elements with modern art forms" and "modern interpretations of ancient literary works."

[0067] Step 3.3: Construct a cultural value assessment model.

[0068] Evaluate the intrinsic value of cultural content by combining multiple factors:

[0069] in, Display content The value rating of represents the value evaluation function, Indicates content quality factors (such as production quality, professionalism, etc.), Indicates content influence factors (such as author influence, dissemination potential, etc.), Indicates the social value of the content (such as educational significance, cultural protection value, etc.), Indicates the historical value of the content (such as scarcity, historical significance, etc.).

[0070] The specific evaluation methods for each factor are as follows: Content quality factors : It can be evaluated through technical quality assessment models (such as clarity and professionalism scores) and audience feedback indicators (such as completion and like rate); Content influence factors : This can be assessed through creator influence indicators (such as past work dissemination data) and initial dissemination indicators (such as sharing rate and interaction rate); Social value factors : It can identify social value themes such as education and protection through semantic analysis and evaluate the strength of the content's relevance to these themes; Historical value factors : The relevance and scarcity of the content to rare cultural heritage can be evaluated through correlation analysis with historical and cultural databases.

[0071] The value assessment model can adopt weighted summation or nonlinear mapping: ; or ; in, Indicates the The weight of the factors, represents the scoring function of the corresponding factor, represents the neural network model, Indicates content quality factors, Indicates the content influence factor, Indicates the social value factor of the content, Indicates the historical value of the content. Indicates the cultural content to be evaluated, Display content The final value score.

[0072] For neural network implementation, a multi-layer perceptron architecture can be used, taking various factors as input features and outputting a value score through multiple layers of nonlinear transformations. The network parameters can be supervised through expert-labeled data or user feedback data.

[0073] In specific application scenarios, the model can evaluate the multi-dimensional value of cultural content such as digitally restored Dunhuang murals, video recordings of intangible cultural heritage skills, and digitized versions of ancient books, providing a value basis for screening and recommending content.

[0074] Step 3.4: Build a comprehensive screening model. Combine novelty and value assessment to screen out emerging cultural content with dissemination value: ; ; in, Display content The comprehensive rating of and is the weight coefficient that balances novelty and value, Display content The novelty index, Display content The value rating of represents a collection of newly discovered cultural content, represents the screening threshold, It represents the final selected set of emerging cultural content with dissemination value.

[0075] Weight coefficient and It can be adjusted dynamically according to the strategy of the recommendation system. For example, in the early stage of cultural communication, the The value of priority is to recommend novel content first and expand the scope of cultural influence; in the deepening stage of communication, it can be increased. The value of , give priority to recommending high-value content, and deepen cultural understanding. In addition, the screening threshold It can also be adjusted dynamically based on the size of the content library and recommendation needs.

[0076] The comprehensive screening process can also take into account diversity constraints to ensure that the screening results cover different cultural categories and expressions: ; in, Represents a collection of filtered results diversity indicators, Indicates the size of the filter result set. and Respectively represent the filter result sets The i-th and j-th cultural content items are used for pairwise comparison to calculate content differences. Indicates the similarity between two content items.

[0077] By maximizing diversity indicators, we can prevent the recommendation system from falling into the trap of recommending a single type of popular content and more comprehensively present cultural diversity.

[0078] In practical applications, this model can screen out novel and valuable content from a large amount of newly generated digital cultural content, such as innovative digital expressions of intangible cultural heritage, immersive experiences combining traditional culture with modern technology, and high-quality digital restorations of rare cultural resources, providing high-quality content sources for knowledge graph expansion and recommendation systems.

[0079] In addition, through the above steps, a model that can automatically discover and evaluate emerging cultural content will be obtained. This model can identify novel and valuable content from a large amount of newly generated cultural content, providing a content source for the expansion of the knowledge graph and innovative recommendations.

[0080] Step 4: Based on the detected cultural events and high-value content, implement the knowledge graph dynamic update and expansion algorithm to update the knowledge graph.

[0081] The specific implementation is as follows: Step 4.1: Construct an event-related knowledge subgraph recognition algorithm.

[0082] Based on the detected cultural events, identify the subgraphs in the knowledge graph that are related to the event: ; in, Representation and Events A collection of related entities, Represents the entire knowledge graph, Represents an entity node in the knowledge graph, Representing an entity and events The correlation of Represents the relevance threshold. Relevance calculation can be based on the vector representation of entities and events: ; in, Representing an entity and events The correlation of Representing an entity The spatiotemporal perception representation of Representing an event The vector representation of Represents the cosine similarity function.

[0083] Step 4.2: Implement an event-based knowledge update algorithm. Update the entity attributes and relationships in the relevant knowledge subgraph based on the event information: ; in, represents the updated knowledge graph, represents the original knowledge graph, represents a detected cultural event, Representation and Events A collection of related entities, Represents the knowledge graph update function, which performs different update operations based on the event type, including: adding new spatiotemporal tags, updating entity attribute values, adjusting relationship weights, and adding new relationship connections. Step 4.3: Build a new content knowledge graph mapping model. Map the selected emerging cultural content to entities and relationships in the knowledge graph: ; ; in, represents the set of entities extracted from the new content, Represents a set of filtered results. represents the set of relations extracted from new content, and Represent the mapping functions from content to entities and relations respectively.

[0084] Step 4.4: Implement the automatic connection algorithm for graph structure prediction. For newly added entities and relationships, predict their connection relationships with entities in the existing knowledge graph: ; in, Representing an entity and There is a relationship between The probability of Represents the source entity (head entity) in the knowledge graph, represents the target entity (tail entity) in the knowledge graph, Represents the graph structure prediction function, which can be implemented based on graph neural network or knowledge graph embedding model. , add the relationship to the knowledge graph.

[0085] The graph structure prediction function can use triple scoring method: ; in, As a scoring function, you can use knowledge graph embedding models such as TransE and DistMult. 、 and Represent the embedding representations of head entity, relation and tail entity respectively. Represents the header entity With tail entity There is a relationship between The probability of represents the head entity in the knowledge graph, Indicates the relationship type, Represents the tail entity in the knowledge graph.

[0086] Step 4.5: Update the spatiotemporal evolution model. Based on the latest knowledge graph structure, update the spatiotemporal propagation rules and evolution laws: ; in, represents the updated space-time propagation function, represents the original space-time propagation function, represents the propagation model update function, Represents the expanded knowledge graph. The spatiotemporal propagation function is used to describe the propagation patterns of cultural content in different time and space dimensions. These patterns are updated by analyzing the knowledge graph structure.

[0087] Therefore, through the above steps, the dual-driven dynamic update and expansion of the knowledge graph will be realized, which can not only respond to the triggering of cultural events and update relevant knowledge in a timely manner, but also automatically integrate emerging cultural content, expand the coverage of the knowledge graph, and provide a more comprehensive, timely and rich knowledge base for the recommendation system.

[0088] Step 5: Based on the updated knowledge graph, construct a timeliness-aware adaptive recommendation algorithm, combine the hotspot decay model and the exploration and utilization balance strategy to generate a recommendation list.

[0089] The specific implementation is as follows: Step 5.1: Construct a hotspot attenuation model.

[0090] For content related to detected cultural events, the hotspot decay weight is calculated based on the difference between the event occurrence time and the current time: ; in, Display content In time The hotspot weight, Representation and content The time when the relevant event occurred, represents the attenuation coefficient, which controls the duration of the hotspot effect. Indicates the current time, Represents the base of the natural logarithm. The decay coefficient can be dynamically adjusted based on the event type: ; in, Representing an event expected duration.

[0091] Step 5.2: Implement a content relevance calculation model based on the knowledge graph.

[0092] Utilize the entity relationship structure in the knowledge graph to calculate the relevance of content and user interests: ; in, Display content With users The correlation, Display content The corresponding entity set, Represents a user The set of entities of interest, Represents the relevance calculation function based on the knowledge graph, Represents a knowledge graph.

[0093] Knowledge graph relevance calculation can be based on the path structure between entities: ; in, It represents the scoring function of the path between entities, which can take into account factors such as path length, path type, and relationship strength. Display content The corresponding entity set, Represents a user The set of entities of interest, Represents a single entity in a collection of content entities, Represents a single entity in the user's interest entity set, represents the knowledge graph, Represents the relevance calculation function based on the knowledge graph, Represents a collection of content entities Each entity in and user interest entity set Each entity in The sum of all possible combinations of is calculated, that is, the sum of the path scores between all entity pairs in the content entity set and the user interest entity set is calculated.

[0094] Step 5.3: Construct an exploration-exploitation balance strategy model.

[0095] While ensuring the relevance of recommended content, rationally introduce novel content and balance exploration and utilization: ; in, Indicates to the user Recommended content The probability of represents the exploration probability, Represents an indicator function (the value is 1 when the condition in the brackets is met, otherwise it is 0). Display content For users The overall recommendation score of represents a probability distribution sampled uniformly from all content, Indicates Get the maximum value , Indicates candidate content.

[0096] In order to adapt to the needs of different users, explore the probability Can be dynamically adjusted based on user behavior: ; in, represents the basic exploration probability, represents the user adaptation function, which dynamically adjusts the exploration ratio based on the user's historical behavior (such as click diversity, content consumption pattern, etc.).

[0097] Step 5.4: Implement the comprehensive recommendation score calculation model.

[0098] Combined with relevance, hot spot weight and novelty, the comprehensive recommendation score of the content is calculated: ; in, Display content For users The overall recommendation score of Display content With users The correlation, Display content In time The hotspot weight, Display content The novelty score of 、 and Represent the weight coefficients of relevance, hotness and novelty respectively, satisfying , which can be dynamically adjusted according to different recommendation scenarios.

[0099] Step 5.5: Construct the final recommendation list generation algorithm. Generate the final recommendation list based on the comprehensive recommendation score and the exploration and utilization balance strategy: ; in, Indicates to the user Recommended content list, represents the recommendation function, Indicates to the user Recommended content The probability of Display content For users The comprehensive recommendation score of Indicates the length of the recommendation list, Represents the exploration probability parameter. The recommendation function is based on the exploration and utilization balance strategy. The probability of selecting the content with the highest score is The probability of randomly selecting from all content ensures that the recommendation results are relevant, diverse, and novel.

[0100] It can be seen that through the above steps, an adaptive recommendation algorithm with timeliness perception ability will be realized. This algorithm can respond to the occurrence of cultural events in a timely manner and recommend content related to hot topics, while maintaining the ability to explore emerging content and providing users with cultural content recommendations that are both relevant and novel.

[0101] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A method for recommending digital cultural communication content based on knowledge graph, characterized in that: The following steps are involved: Build a spatiotemporal knowledge representation model, introduce spatiotemporal tags to construct a quadruple knowledge representation structure, and realize the spatiotemporal awareness of the knowledge graph; Based on the knowledge representation model, a cultural event detection model integrating multi-source data is constructed to collect multi-source data and extract time series features, calculate the suddenness index, and detect cultural events; Based on the detected cultural events, a new cultural content discovery and value assessment model is constructed to extract content features for multi-dimensional evaluation, calculate novelty and value scores, and screen high-value content; Based on the detected cultural events and high-value content, the knowledge graph is dynamically updated and expanded by the algorithm to update the knowledge graph; Based on the updated knowledge graph, a timeliness-aware adaptive recommendation algorithm is constructed, which combines the hotspot decay model and the exploration and utilization balance strategy to generate a recommendation list.

2. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The construction of a spatiotemporal knowledge representation model includes: Construct a spatiotemporal labeled quadruple knowledge representation structure; Define the structure of spatiotemporal markers; Constructing spatiotemporal-aware entity and relationship representations; Implement space-time propagation rules.

3. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The cultural event detection model constructed by multi-source data fusion includes: Multi-source data collection and preprocessing; Construct a temporal feature extraction model for cultural events; Implement the burst index calculation model; Build a multi-source fusion event detection model.

4. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The construction of a new cultural content discovery and value assessment model includes: Construct a model to represent cultural content characteristics; Implementing a novelty measurement model; Construct a cultural value assessment model; Construct a comprehensive screening model.

5. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The algorithm for dynamically updating and expanding the knowledge graph includes the following steps: Construct a knowledge subgraph recognition algorithm for event association; Implement event-based knowledge update algorithm; Build a new content knowledge graph mapping model; Implement an automatic connection algorithm for graph structure prediction; Update the spatiotemporal evolution model.

6. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The timeliness-aware adaptive recommendation algorithm comprises the following steps: Construct hotspot attenuation model; Implement a content relevance calculation model based on knowledge graph; Construct an exploration-exploitation balance strategy model; Implement comprehensive recommendation score calculation model; Construct the final recommendation list generation algorithm.

7. The method for recommending digital cultural content based on knowledge graph according to claim 4, characterized in that: The construction of the cultural value assessment model includes the following steps: Establish cultural content quality assessment indicators that combine expert evaluation and user feedback; Build a cultural content influence assessment model to analyze the scope and speed of dissemination; Implement a cultural content sentiment analysis model to assess sentiment tendency and intensity; Establish a historical value assessment model for cultural content and analyze its relevance to historical culture.

8. The method for recommending digital cultural content based on knowledge graph according to claim 5, characterized in that: The automatic connection algorithm for graph structure prediction includes the following steps: Learn the association patterns between entities based on the existing knowledge graph structure; Use graph neural networks to predict potential relationships between new entities and existing entities; Calculate relationship confidence and set thresholds to filter high-confidence relationships; Verify the consistency of the newly added relations through knowledge reasoning.

9. The method for recommending digital cultural content based on knowledge graph according to claim 1, characterized in that: The construction of the exploration-exploitation balance strategy model includes the following steps: Design a dynamic exploration rate adjustment mechanism to adaptively adjust based on user feedback and content characteristics; Implement a multi-armed bandit algorithm based on Thompson sampling to balance the recommendation ratio of hot content and novel content; Build a user interest drift detection model and adjust the exploration and utilization strategy in a timely manner; Design a diversity protection mechanism to ensure the content diversity of the recommendation list.

10. A digital cultural communication content recommendation system based on knowledge graph, characterized in that: A method for recommending digital cultural communication content based on a knowledge graph according to any one of claims 1 to 9, comprising: The spatiotemporal knowledge representation module is used to construct a spatiotemporal labeled quadruple knowledge representation structure, define the structure of spatiotemporal labels, build spatiotemporal aware entity and relationship representations, and implement spatiotemporal propagation rules; The cultural event detection module of multi-source data fusion is used to collect and preprocess multi-source data, build a temporal feature extraction model for cultural events, implement a suddenness index calculation model, and build a multi-source fusion event detection model; New cultural content discovery and value assessment module, used to build a cultural content feature representation model, implement a novelty measurement model, build a cultural value assessment model, and build a comprehensive screening model; The knowledge graph dynamic update and expansion module is used to build an event-related knowledge subgraph recognition algorithm, implement an event-based knowledge update algorithm, build a new content knowledge graph mapping model, implement an automatic connection algorithm for graph structure prediction, and update the spatiotemporal evolution model; The timeliness-aware adaptive recommendation module is used to build a hotspot decay model, implement a content relevance calculation model based on the knowledge graph, build an exploration and utilization balance strategy model, implement a comprehensive recommendation score calculation model, and build a final recommendation list generation algorithm.

Images