Non-residual knowledge graph construction method and system based on multiple modes

Abstract

The invention discloses a non-residual knowledge graph construction method and system based on multiple modes, and relates to the technical field of data processing, and the method comprises the steps: building a cloud database; establishing a first knowledge graph; configuring a trust channel, performing independent data authentication on entities in the first knowledge graph, and establishing a data authentication result; activating the entity similar authentication network to perform entity association calculation, generating an authentication calling instruction, performing auxiliary authentication of an authentication failure result, and generating an auxiliary authentication result; establishing a second knowledge graph, embedding entities and relationships of the second knowledge graph into a low-dimensional vector space, updating a relationship weight of the second knowledge graph, and establishing a third knowledge graph; and generating activation representation according to the third knowledge graph, and performing digital display and interaction. The technical problem that in the prior art, a non-abandoned digital protection and display method cannot comprehensively reflect non-abandoned richness is solved, and the technical effect of improving the expression ability and the display effect of the non-abandoned knowledge graph is achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for constructing a multimodal intangible cultural heritage knowledge graph. Background Art

[0002] With the rapid development of information technology and digital technology, the protection and inheritance of intangible cultural heritage has received more and more attention. However, with the rapid growth of information and the diversification of data forms, the digital protection and management of intangible cultural heritage are also facing new challenges. Traditional intangible cultural heritage digitization methods often focus on the collection and display of a single data modality, such as text records, picture preservation or video recording, but these methods are powerless in the face of increasingly complex intangible cultural heritage information. As a comprehensive cultural phenomenon, intangible cultural heritage has rich and diverse connotations, covering multiple data forms such as text, images, audio, video and three-dimensional models. It is difficult to fully express its value and significance through a single data modality. Summary of the invention

[0003] The present application provides a method and system for constructing a multimodal intangible cultural heritage knowledge graph, which is used to solve the technical problem that the existing intangible cultural heritage digital protection and display methods cannot fully reflect the richness of intangible cultural heritage.

[0004] In view of the above problems, this application provides a method and system for constructing a multimodal intangible cultural heritage knowledge graph.

[0005] The first aspect of the present application provides a method for constructing a multimodal intangible cultural heritage knowledge graph, the method comprising:

[0006] Establish a cloud database, the cloud database is used to store digital data of intangible cultural heritage, the digital data includes multimodal data source data of text, picture, audio, video, VR, AR and three-dimensional model; describe the digital data in the cloud database based on the ontology, and use each ontology as an entity of the knowledge graph, and establish a first knowledge graph with the entity; configure a trust channel, perform independent data authentication on the entities in the first knowledge graph based on the trust channel, and establish data authentication results, wherein the data authentication results include authentication pass results, authentication fail results and fuzzy authentication results; activate the entity similarity authentication network, perform entity association calculation through the entity similarity authentication network, generate authentication call instructions based on the entity association calculation results, perform auxiliary authentication of the authentication fail result through the authentication call instructions, and generate auxiliary authentication results; update the first knowledge graph according to the data authentication results and the auxiliary authentication results, establish a second knowledge graph, and establish an objective function for embedding vectors, embed the entities and relationships of the second knowledge graph into a low-dimensional vector space based on the objective function, update the relationship weights of the second knowledge graph according to the embedding results, and establish a third knowledge graph; generate an activated representation according to the third knowledge graph for digital display and interaction of intangible cultural heritage.

[0007] The second aspect of the present application provides a system for constructing a multimodal intangible cultural heritage knowledge graph, the system comprising:

[0008] A database establishment module, wherein the database establishment module establishes a cloud database, wherein the cloud database is used to store digital data of intangible cultural heritage, wherein the digital data includes multimodal data source data of text, picture, audio, video, VR, AR and three-dimensional model; a first knowledge graph establishment module, wherein the first knowledge graph establishment module describes the digital data in the cloud database based on the ontology, and uses each ontology as an entity of the knowledge graph to establish the first knowledge graph with the entities; an independent data authentication module, wherein the independent data authentication module configures a trust channel, performs independent data authentication on entities in the first knowledge graph based on the trust channel, and establishes data authentication results, wherein the data authentication results include authentication pass results, authentication fail results and fuzzy authentication results; an auxiliary authentication module, wherein the auxiliary The auxiliary authentication module activates the entity similarity authentication network, performs entity association calculation through the entity similarity authentication network, generates an authentication call instruction based on the entity association calculation result, performs auxiliary authentication of the authentication failure result through the authentication call instruction, and generates an auxiliary authentication result; the knowledge graph update module, the knowledge graph update module updates the first knowledge graph according to the data authentication result and the auxiliary authentication result, establishes the second knowledge graph, and establishes the objective function of the embedding vector, embeds the entities and relationships of the second knowledge graph into the low-dimensional vector space based on the objective function, updates the relationship weights of the second knowledge graph according to the embedding result, and establishes the third knowledge graph; the activation representation generation module, the activation representation generation module generates an activation representation according to the third knowledge graph for the digital display and interaction of intangible cultural heritage.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] The present application establishes a cloud database, which is used to store digital data of intangible cultural heritage, and the digital data includes multimodal data source data of text, pictures, audio, video, VR, AR and three-dimensional models; describes the digital data in the cloud database based on the ontology, and uses each ontology as an entity of the knowledge graph to establish a first knowledge graph with entities; configures a trust channel, and performs independent data authentication on the entities in the first knowledge graph based on the trust channel, and establishes data authentication results, wherein the data authentication results include authentication pass results, authentication fail results and fuzzy authentication results; activates the entity similarity authentication network, performs entity association calculation through the entity similarity authentication network, generates authentication call instructions based on the entity association calculation results, performs auxiliary authentication of authentication fail results through the authentication call instructions, and generates auxiliary authentication results; updates the first knowledge graph according to the data authentication results and the auxiliary authentication results, establishes a second knowledge graph, and establishes an objective function for embedding vectors, embeds the entities and relationships of the second knowledge graph into a low-dimensional vector space based on the objective function, updates the relationship weights of the second knowledge graph according to the embedding results, and establishes a third knowledge graph; generates an activated representation according to the third knowledge graph for digital display and interaction of intangible cultural heritage. The present invention solves the technical problem that the digital protection and display methods of intangible cultural heritage in the prior art cannot fully reflect the richness of intangible cultural heritage. It establishes a cloud database to store multimodal intangible cultural heritage data, constructs a first knowledge graph based on the ontology and performs data authentication, activates the entity similarity authentication network to perform association calculation and auxiliary authentication, and uses embedded vectors to optimize the knowledge graph relationship weights. Finally, it generates a third knowledge graph for digital display and interaction of intangible cultural heritage, thereby achieving the technical effect of improving the expression ability and display effect of the intangible cultural heritage knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of the process of constructing a multimodal intangible cultural heritage knowledge graph provided in an embodiment of the present application;

[0013] Figure 2 A schematic diagram of the structure of a system for constructing a multimodal intangible cultural heritage knowledge graph provided in an embodiment of the present application.

[0014] Explanation of the accompanying drawings: database establishment module 11, first knowledge graph establishment module 12, independent data authentication module 13, auxiliary authentication module 14, knowledge graph update module 15, activation representation generation module 16. DETAILED DESCRIPTION

[0015] The present application provides a method and system for constructing a multimodal intangible cultural heritage knowledge graph, aiming to solve the technical problem that the intangible cultural heritage digital protection and display methods in the prior art cannot fully reflect the richness of the intangible cultural heritage. By establishing a cloud database to store multimodal intangible cultural heritage data, a first knowledge graph is constructed based on the ontology and data authentication is performed, an entity similarity authentication network is activated to perform association calculation and auxiliary authentication, and embedded vectors are used to optimize the knowledge graph relationship weights. Finally, a third knowledge graph is generated for digital display and interaction of intangible cultural heritage, thereby achieving the technical effect of improving the expression ability and display effect of the intangible cultural heritage knowledge graph.

[0016] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0017] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0018] Embodiment 1, as Figure 1 As shown, the present application provides a method for constructing a multimodal intangible cultural heritage knowledge graph, the method comprising:

[0019] Step S100: Establish a cloud database, which is used to store digital data of intangible cultural heritage, and the digital data includes multimodal data source data of text, pictures, audio, video, VR, AR and three-dimensional models.

[0020] In the embodiment of the present application, the user uploads the multimodal data source of intangible cultural heritage to the cloud. The multimodal data source includes text, pictures, audio, video, VR, AR and three-dimensional model multimodal data source data. The text includes traditional stories and historical documents, the pictures include photos of handicrafts and traditional costumes, the audio includes folk music and oral history recordings, the video includes dance and ceremony recordings, and the three-dimensional model is a 3D scanned model of a building or cultural relic.

[0021] Through the above process, the construction of the cloud database is completed.

[0022] Step S200: Describe the digital data in the cloud database based on the ontology, and use each ontology as an entity of the knowledge graph to establish a first knowledge graph with the entities.

[0023] In the embodiment of the present application, after the data is uploaded and stored in the cloud database, ontology construction is performed by using the Web Ontology Language to define key concepts and their attributes in the field of intangible cultural heritage, such as "traditional dance", "folk literature" and "skills", as well as the semantic relationships between them, such as "belongs to", "includes" or "originated from".

[0024] Next, data mapping is performed, and the digital data in the cloud database is mapped to the constructed ontology concepts using the semantic annotation tool Protégé. For example, the text data is matched with the corresponding ontology concepts through the manual annotation function in Protégé, while the image data is manually annotated to the corresponding ontology concepts after human review. This step ensures that the multimodal data can be correctly associated with the corresponding ontology to form a semantic connection. After completing the data mapping, the next step is the entity process. Using the graph database tool Neo4j, each mapped data object is converted into an entity in the knowledge graph. The entity is the core unit of the knowledge graph, representing a specific object or concept in the field of intangible cultural heritage. Through Neo4j, an ontology concept such as "traditional dance" is entityized as a node in the graph, and it is connected to other related entities such as "accompanied music" through the relationship defined in the ontology to form a preliminary knowledge network.

[0025] Finally, all entities and their relationships are integrated into a complete knowledge graph through the knowledge graph construction tool Neo4j to obtain the first knowledge graph. In this step, Neo4j uses its powerful graph database function to organize these entities and relationships into a structured network, so that the multimodal data of intangible cultural heritage and its relationships can be effectively displayed and retrieved.

[0026] Step S300: configure a trust channel, perform independent data authentication on entities within the first knowledge graph based on the trust channel, and establish data authentication results, wherein the data authentication results include authentication pass results, authentication fail results, and fuzzy authentication results.

[0027] In the embodiment of the present application, the trust channel is first configured. Specifically, in this process, an enterprise-level blockchain platform such as Hyperledger Fabric is used to build the trust channel. Whenever a new data is uploaded to the cloud database, the blockchain network records its detailed source information, including the data generation time, upload time, data provider, etc. The configuration of the trust channel is completed through this process.

[0028] Next, the entities in the first knowledge graph are independently authenticated through the trust channel. Specifically, the data source of each entity in the knowledge graph is read by configuring the trust channel. For each data source, a source trust identifier is generated, which is based on factors such as the reliability, authority, and historical performance of the source. Then, based on these source trust identifiers, contradictory relationships are identified within the entity. This process aims to detect whether there is a conflict between different data sources of the same entity. For example, if the information of two data sources of an entity is inconsistent on a certain attribute, this contradiction is identified. These contradictory relationships are then weighted for authentication. In the weighted authentication process, the importance of each data source is weighted according to its trust identifier. Sources with higher trust will have a greater weight in the authentication process, thereby affecting the final authentication result to a greater extent. Finally, based on the results of the weighted authentication, the data authentication result is established. The data authentication results are divided into authentication passed, authentication failed, and fuzzy authentication. If no obvious contradiction is found between the data sources, or the contradiction is small and the trust of the main source is high, the entity will be marked as "authentication passed". If the contradiction is significant and the trust of the main source is low, it will be marked as "authentication failed". When the contradictions between data cannot be clearly resolved, or the trust scores are close, a "fuzzy authentication" result is generated.

[0029] Furthermore, in the method provided in the embodiment of the application, the independent data authentication of entities in the first knowledge graph based on the trust channel and the establishment of data authentication results also include:

[0030] The data source in the entity is read by using the trust channel, and a source trust mark is generated based on the data source; the contradictory relationship of the data in the entity is identified based on the trust channel, and weighted authentication of the contradictory relationship identification is performed based on the source trust mark, and a data authentication result is established according to the weighted authentication result.

[0031] In an embodiment of the present application, the data source of each entity in the knowledge graph is first read through a trust channel. The trust channel is based on blockchain technology, such as Hyperledger Fabric, which is a distributed ledger technology that records the source and change history of all data. When new data is uploaded to the cloud database, a timestamp is automatically generated, and the source of the data, provider information, upload time, etc. are recorded in detail in the blockchain. At the same time, a unique identifier, such as a hash value, is generated for each data to ensure the immutability and traceability of the data. For example, when a video data about "traditional dance" is uploaded, the blockchain will record the source and upload time of the video, and generate a unique hash value as an identifier.

[0032] Next, a source trust mark is generated based on the data source information. This process relies on trust assessment algorithms, such as the Bayesian trust model or the PageRank-based trust assessment algorithm. A trust score is calculated by analyzing factors such as the historical performance of the data source, the reputation of the data provider, and the number of citations of the data. Specifically, when calculating the trust score, a historical performance score, an authority score, and a citation score are performed. The historical performance score reflects the reliability of a data source in past records and is calculated by counting the proportion of past data from the source that has been authenticated. For example, if a data source has 100 records, of which 80 records have been judged to be credible in past authentications, the historical performance score is calculated by dividing the number of authenticated records by the total number of records to obtain a historical performance score of 0.8. The authority score is based on the institutional background and industry reputation of the data source. Different types of sources have different pre-set basic scores. The authority score of academic institutions is 0.9 to 1.0, the authority score of industry associations is 0.7 to 0.8, and the authority score of individuals or unknown sources is 0.4 to 0.6. The citation score reflects how often the data of the source is cited in other authoritative literature and is calculated by analyzing the citation count of the data source. For example, if a data source is cited 50 times in an authoritative database and the total citation count is between 0 and 100, then its citation score is normalized to

[0033] The highest citation is removed from the citation count to obtain a citation score of 0.5. Different factors have different importance, so a weight is assigned to each factor. The weight is set based on specific application requirements. In this application, historical performance is considered more important, so it is assigned a weight of 0.6, while authority and citation scores are assigned weights of 0.3 and 0.1, respectively. After obtaining the score and weight of each factor, the weighted average method is used to calculate the final trust score, and the source trust mark is generated through this process.

[0034] After generating the trust mark, contradictory relationship identification is performed. The purpose of contradictory relationship identification is to detect whether there is a conflict in the information about the same entity from different sources. This step relies on natural language processing technology or data consistency check algorithms. For example, the nuances in the text description are parsed through NLP technology to identify the contradictions between different descriptions of the same entity. If the descriptions of an entity by two data sources are inconsistent in key details, it is marked as contradictory data. For example, one source describes "traditional dance" as originating in the 19th century, and another source describes it as the 18th century, and it is marked as a potential conflict point.

[0035] After marking the conflicting data, compare the trust scores of the two data sources. If the trust score of one data source is significantly higher than that of the other data source, mark the data with higher trust as a certified result, and mark the data with lower trust as a failed result. For example, if the trust score of data source A describing that "traditional dance" originated in the 19th century is 0.95, and the trust score of data source B describing it in the 18th century is 0.7, then select data source A with a higher trust score and mark it as a certified result, while mark data source B as a failed result. If the trust scores of two data sources are the same or very close, mark both data as fuzzy certification results. In this case, it is impossible to clearly decide which data is more credible, so further verification or manual review is required. For example, if the trust scores of two data sources A and B describing that "traditional dance" originated in the 19th century are 0.85 and 0.84 respectively, it is considered that the credibility difference between the two data sources is not large, so both are marked as fuzzy certification results.

[0036] When it is identified that there are multiple data sources that have contradictory descriptions of the same entity, and the trust scores of these data sources are not significantly different, the weighted authentication stage is entered. For example, when there are two data sources describing that "traditional dance" originated in the 19th century and the other source describes it in the 18th century, the size of the trust score is compared. If the two data sources describing that "traditional dance" originated in the 19th century are both less than the trust score describing that traditional dance is in the 18th century, weighted authentication is performed. At this time, since there are two data sources describing that "traditional dance" originated in the 19th century, the description content is the same, and the growth coefficient is introduced to enhance the influence of the same description. For the trust scores of these two data, a growth coefficient is multiplied by each of them. The growth coefficient is pre-set. In this application, the growth coefficient is set to 1.1. After that, the two adjusted trust scores are weighted to obtain a comprehensive trust, and then the comprehensive trust is compared with the trust score describing that "traditional dance" originated in the 18th century. The data with a high trust score is used as the authentication result, and the data with a low trust score is used as the authentication result. If the trust scores of the two data are the same, both data are used as fuzzy authentication results.

[0037] Through the above process, a data authentication result is established.

[0038] Step S400: activating an entity similarity authentication network, performing entity association calculation through the entity similarity authentication network, generating an authentication call instruction based on the entity association calculation result, performing auxiliary authentication of the authentication failure result through the authentication call instruction, and generating an auxiliary authentication result.

[0039] In the embodiment of the present application, in the process of constructing the intangible cultural heritage knowledge graph, it is necessary to ensure the accuracy and relevance of each entity in the knowledge graph. For example, an intangible cultural heritage project may involve multiple data types, such as text, pictures, audio and video. These data are provided by different sources and stored in a cloud database. However, during the preliminary authentication process, some entities may not pass the authentication due to insufficient data or conflicting descriptions. In order to improve the accuracy of the knowledge graph, an entity similarity authentication network is introduced for further verification. In the case of failure of the preliminary authentication, the entity similarity authentication network is activated to perform entity association calculations on the relevant entities in the knowledge graph, and an authentication call instruction is generated based on the entity association calculation results. After receiving the authentication call instruction, the auxiliary authentication stage is entered. At this time, a deeper analysis of "Qiang embroidery" is conducted based on the contribution. Evaluate its influence in cultural communication and tourism, or its unique value in intangible cultural heritage inheritance. Assuming that "Qiang embroidery" is extremely attractive in some cultural exhibitions, even if its descriptions are conflicting, it is retained according to its high contribution and marked as "doubtful" for subsequent more refined verification or manual review. On the contrary, if "Qiang embroidery" contributes less in these evaluations and fails multiple verifications, it may be finally decided to eliminate the entity and remove it from the knowledge graph.

[0040] Through the above process, an auxiliary authentication result is generated.

[0041] Furthermore, in the method provided in the embodiment of the application, the entity association calculation is performed through the entity similarity authentication network, and further includes:

[0042] The frequency statistics layer of the entity similarity authentication network is called to perform co-occurrence frequency and independent frequency statistics of entities, and a first degree of association is generated based on the co-occurrence frequency and independent frequency; the path identification layer of the entity similarity authentication network is called to perform shortest path length calculation between entities, and a second degree of association is generated according to the shortest path length calculation result; the cosine similarity calculation layer of the entity similarity authentication network is called to perform similarity calculation between entities and generate a third degree of association; and an entity association calculation result is generated according to the first degree of association, the second degree of association, and the third degree of association.

[0043] In an embodiment of the present application, the frequency statistics layer of the entity similarity authentication network is first called to count the co-occurrence frequencies and independent frequencies of the entities in the knowledge graph. Co-occurrence frequency refers to the number of times two entities appear in the same context or relationship, and independent frequency refers to the number of times each entity appears independently. By counting these frequencies, the strength of the association between two entities is measured. The higher the co-occurrence frequency, the stronger the semantic association between the two entities is generally. Based on these statistical data, the first degree of association is calculated to quantify the basic association between the two entities. For example, if "Qiang embroidery" and "Qiang culture" frequently appear together in the descriptions of multiple intangible cultural heritage items, their first degree of association will be higher, indicating that they have a strong semantic association. Specifically:

[0044]

[0045] Next, the path identification layer of the entity similarity authentication network is called to perform the shortest path length calculation between entities. The shortest path length refers to the length of the shortest relationship chain from one entity to another in the knowledge graph. The shorter the path, the more direct the relationship between the two entities and the stronger the association. By calculating the shortest path length, the second degree of association is generated. For example, in the knowledge graph, if "Qiang embroidery" and "Qiang dance" are directly related through a few intermediary entities, then their second degree of association will be higher, reflecting the strong association between them. Specifically, the second degree of association = 1 / shortest path length.

[0046] Then the cosine similarity calculation layer of the entity similarity authentication network is called to perform similarity calculations between entities. Cosine similarity is a commonly used similarity measurement method used to calculate the similarity between two vectors in a multidimensional space. By representing the entities in the knowledge graph as vectors, and then calculating the cosine similarity between these vectors. The higher the cosine similarity value, the closer the directions of the two entities in the vector space are, indicating that they are more semantically similar. Through this calculation, the third degree of association is generated, that is, the calculated cosine similarity is the third degree of association.

[0047] After obtaining the first degree of association, the second degree of association and the third degree of association, a weighted fusion algorithm is used to combine the degrees of association from different sources to generate the final entity association calculation result. The weights of the first degree of association, the second degree of association and the third degree of association are pre-set, and the entity association calculation result is obtained through calculation.

[0048] Furthermore, in the method provided in the embodiment of the application, the generating of the auxiliary authentication result further includes:

[0049] Call the abnormal authentication result in the auxiliary authentication result, and perform a threshold trigger determination of a negative impact on the abnormal authentication result; if the threshold of the negative impact is not triggered, perform a contribution evaluation based on the corresponding abnormal authentication result; generate reserved data based on the contribution evaluation result, and identify the reserved data as untrusted data; and save the reserved data in the auxiliary authentication result.

[0050] In an embodiment of the present application, the abnormal authentication result in the auxiliary authentication result is first called. For example, when processing the intangible cultural heritage item "Qiang Embroidery", it may be found that the descriptions from certain sources are significantly different from most data in terms of the origin or age of the craft. To identify these abnormal data, anomaly detection algorithms such as Isolation Forest or DBSCAN are used to analyze the distribution of the data. Through these algorithms, entities that deviate significantly from other data in semantics or structure are marked and identified as "abnormal authentication results".

[0051] Next, the negative impact of the abnormal data is determined. Specifically, the risk value of each abnormal data is first calculated. The calculation of the risk value is based on the degree of relationship rupture or information transmission interference that the abnormal data may cause. Evaluate whether abnormal data, such as incorrect time information, will cause important relationships between key entities to be broken, or whether it will cause incorrect information propagation in the knowledge graph. The calculation of the risk value is determined by the number and importance of the affected relationships. Specifically, all entities and relationships related to the abnormal data are first identified. For example, if the time information of "Qiang Embroidery" is wrong, identify which cultural events or historical entities depend on this time information. For each identified affected relationship, evaluate its importance in the knowledge graph. Importance is measured by the connectivity of the relationship in the knowledge graph, that is, how many other entities depend on this relationship. For example, a relationship connecting multiple core entities is considered to be of high importance. Then a risk value is calculated for each abnormal data based on the number and importance of the affected relationships. Risk value = ∑ (number of affected relationships × importance of the relationship), which reflects the potential destructiveness of abnormal data to the overall structure of the knowledge graph. This risk value is compared with the pre-set threshold. If the calculated risk value exceeds the threshold, the data is marked as high-risk data and will not be retained. If it does not exceed the threshold, it is considered that the negative impact of the data is small and proceeds to the next step.

[0052] If the threshold of negative impact is not triggered, the contribution evaluation is performed based on the corresponding abnormal authentication results. Specifically, in the contribution evaluation process, the key indicators of contribution evaluation include academic value, cultural influence and historical significance. Each indicator is pre-assigned a weight to indicate its relative importance in the total score. Total score = (academic value × 0.3) + (cultural influence × 0.5) + (historical significance × 0.2), and cultural influence is given the highest weight because it may be the most important in the knowledge graph of intangible cultural heritage. Based on these weights and the specific scores of each indicator, the abnormal data is comprehensively scored. For example, if a description of "Qiang embroidery" has unique value in cultural research despite the conflict in time, it is given a higher score in the cultural influence indicator. Then, the contribution of the data is judged based on the total score. If the score is higher than the preset threshold, the data is retained; if it is lower than the preset threshold, it is not retained.

[0053] For abnormal data with high scores and that have not triggered the negative impact threshold, they are generated as retained data and marked as "untrusted data". This mark indicates that although these data are retained, they have certain uncertainties, reminding users to be careful when using them. For example, a controversial description of "Qiang embroidery" is marked as "doubtful data". Finally, the retained data marked as untrusted is saved in the auxiliary authentication results.

[0054] Step S500: Update the first knowledge graph according to the data authentication result and the auxiliary authentication result, establish a second knowledge graph, and establish an objective function for embedding the vector. Based on the objective function, embed the entities and relationships of the second knowledge graph into a low-dimensional vector space, update the relationship weights of the second knowledge graph according to the embedding results, and establish a third knowledge graph.

[0055] In the embodiment of the present application, the first knowledge graph is first updated according to the data authentication results and the auxiliary authentication results, including removing the data that has not passed the authentication and integrating the authenticated data, so as to ensure that the knowledge graph contains only the most accurate and reliable information. For example, if a description in the "Qiang Embroidery" project passes the authentication, it is associated with the existing related entities, such as "Qiang Culture", and its relationship information is updated. Through this process, a second knowledge graph containing the latest and most reliable data is generated.

[0056] Next, in order to further optimize and utilize the knowledge graph, the entities and relations in the knowledge graph are embedded into a low-dimensional vector space for subsequent calculation and analysis. To this end, an objective function of embedding vectors is established, and the entities and relations in the second knowledge graph are mapped to the low-dimensional vector space using knowledge graph embedding technology.

[0057] Finally, the relationship weights of the second knowledge graph are updated according to the embedding results, and the third knowledge graph is established. Specifically, by traversing all triples in the knowledge graph, the distances of the head entity, relationship, and tail entity in each triple in the low-dimensional vector space are calculated by Euclidean distance or cosine similarity. Next, based on the distance values ​​of all triples containing the same relationship, the average distance or weighted average distance of the relationship is calculated to determine the weight of the relationship. The weight of the relationship is obtained by calculating the inverse of the average distance for all triples involving the relationship. The smaller the distance, the greater the weight, indicating that the relationship shows higher consistency and stability in multiple triples. In order to avoid excessive differences in weights between different relationships, the calculated weights are standardized so that the weight values ​​of all relationships are in the same range, such as between 0 and 1, which facilitates subsequent graph adjustments.

[0058] Based on the calculated weights, the second knowledge graph is further adjusted to generate the third knowledge graph. First, identify those relationships with higher weights and enhance their connection strength in the knowledge graph. For example, for a high-weight relationship, increase the connection strength between entities related to the relationship in the graph, or place the relationship in the core position of the graph structure for subsequent analysis and retrieval. For relationships with lower weights, adopt a strategy of weakening or removing them.

[0059] Through the above process, the establishment of the third knowledge graph is completed.

[0060] Furthermore, in the method provided in the embodiment of the application, the objective function of establishing the embedding vector, embedding the entities and relationships of the second knowledge graph into the low-dimensional vector space based on the objective function, further includes:

[0061] Configure the objective function as follows:

[0062]

[0063]

[0064] in, Represents positive triples, is the set of all positive triples, h is the head entity, r describes the association between the head entity and the tail entity, t is the tail entity, represents the negative example triple, h' represents the head entity after being replaced, r' describes the association after being replaced, t' represents the tail entity after being replaced, and γ represents the marginal parameter, which is used to control the function's ability to distinguish between positive and negative examples. Characterize the hierarchical embedding distance, d hier is the hierarchical distance function, are the embedding vectors of the head entity, the associated entity, and the tail entity, respectively. Characterizes context-dependent embedding distance, and are the hierarchical embedding distance and context-dependent embedding distance of negative examples respectively, and + is a positive function. The embedding vector is learned by minimizing the objective function, and the entities and relations of the second knowledge graph are embedded into a low-dimensional vector space.

[0065] In the embodiment of the present application, the objective function is used to optimize the representation of entities and relations in the low-dimensional vector space. The main purpose of the objective function is to ensure that semantically similar entities and relations are as close as possible in the vector space by minimizing the distance difference between positive triples and negative triples. This objective function consists of two distance functions, which are used to measure the hierarchical embedding distance d hier and context-dependent embedding distance d context The (h, r, t) in the formula represents the positive triple in the knowledge graph, that is, the real entity-relationship-entity combination, while (h', r', t') is the negative triple, representing the false combination formed by the replaced entity or relationship.

[0066] In the objective function, the hierarchical embedding distance It is used to measure the distance between entities and relationships in the hierarchy of the knowledge graph. The hierarchy refers to the organizational structure of entities and relationships in the knowledge graph. For example, there may be a parent-child relationship or a category relationship between some entities. By calculating the hierarchical embedding distance, it ensures that entities and relationships with semantic hierarchical relationships maintain this hierarchy in the low-dimensional vector space. For example, if "Qiang embroidery" is a subcategory of "Qiang culture", the system will calculate d hier to ensure that their distance in vector space is short.

[0067] Context-dependent embedding distance The distance between entities connected by a certain relationship in a specific context is calculated. Context can be understood as the way entities interact with other entities under a specific relationship. For example, the definition of "Qiang embroidery" may be different in different cultural backgrounds. context Calculate how these changes affect the distance between entities. Context-dependent embedding distance helps capture more complex semantic relationships, for example, considering the impact of different cultural environments on the definition of "Qiang embroidery" and ensuring that these changes are reasonably expressed in the vector space.

[0068] The optimal embedding vector is learned by minimizing the objective function. The core of the objective function is to establish sufficient discrimination between positive triples and negative triples, so that the positive triples, that is, the real relationship, are more closely represented in the vector space, while the negative triples, that is, the false or unreasonable relationship, are relatively scattered in the vector space. Specifically, by adjusting the embedding vector, the hierarchical embedding distance dhier and context-dependent embedding distance d context The difference between them is maximized, thereby optimizing the representation of entities and relations in the knowledge graph in the low-dimensional vector space. Through this process, the entities and relations in the second knowledge graph are successfully embedded in the low-dimensional vector space.

[0069] Furthermore, the method provided in the application embodiment also includes:

[0070] Establish a user data set, wherein the user data set includes a user direct feedback data set and a user interaction data set, wherein the user interaction data set includes query logs, click behaviors, and search preferences; perform feedback analysis based on the user data set and establish feedback analysis results; establish adaptive weight response feedback based on the feedback analysis results, and optimize the third knowledge graph based on the adaptive weight response feedback.

[0071] In an embodiment of the present application, a user data set is first established through data collection and classification technology, which includes a user direct feedback data set and a user interaction data set. The user direct feedback data set directly collects users' opinions and suggestions on the content of the knowledge graph through user interaction interfaces, such as evaluation forms and feedback buttons, and stores these feedback information in a database for preliminary sorting. The user interaction data set tracks the user's behavioral data when using the knowledge graph platform through log recording technology. These data include the user's query log, click behavior, and search preferences. The query log records all search queries entered by the user on the knowledge graph platform, the click behavior records every entity or relationship clicked by the user on the platform, and the search preference summarizes the user's interest tendencies by analyzing the user's multiple search and click behavior patterns.

[0072] After establishing the user data set, use text analysis and behavior analysis techniques to conduct feedback analysis on this data. First, use natural language processing technology to perform text analysis on the user's direct feedback data to extract keywords, topics, or sentiment tendencies in the user feedback. For example, use TF-IDF to analyze user opinions and suggestions to identify which content is a problem or concern that users repeatedly mention. For user interaction data, use behavioral analysis methods such as click-through rate analysis, heat map analysis, or path analysis to identify user interest patterns. For example, the click-through rate can be used to identify the entities or relationships that users click most frequently, while path analysis can be used to reveal the user's common navigation paths in the knowledge graph, thereby helping the system optimize the way content is displayed.

[0073] Based on the results of feedback analysis, an adaptive weight response feedback mechanism was further established through machine learning and weight adjustment algorithms. The feedback analysis results are used as input data and input into a machine learning model, such as a regression-based weight adjustment model or a reinforcement learning model. The purpose of this model is to dynamically adjust the weights of entities and relationships in the knowledge graph based on user feedback. During the model training process, historical user data and feedback results are used to train the model so that it learns to recognize user preferences. For example, if the model finds that "Qiang embroidery" is a topic that users frequently query, it will automatically increase the weights of entities and relationships related to "Qiang embroidery", making these contents easier for users to find in subsequent searches and browsing. After training, the model can adjust the weights in the knowledge graph in real time in actual applications to ensure that the graph content always meets user needs.

[0074] Under the adaptive weight response feedback mechanism, the third knowledge graph is optimized. First, the weights adjusted by the adaptive weight response feedback mechanism are applied to the third knowledge graph. That is, the priority of each entity and relationship in the knowledge graph is recalculated, and the structure of the graph is rearranged according to the new weights. Entities and relationships with higher weights will be more prominent in the graph, and users will find it easier to find this information when searching. Next, a graph reconstruction algorithm, such as a graph optimization algorithm or a community detection algorithm, is used to optimize the structure of the knowledge graph to eliminate redundant connections, improve the structural efficiency of the graph, and enable users to find the required information more quickly. Finally, A / B testing or user experiments are used to verify whether the optimized knowledge graph has improved user satisfaction and interaction efficiency. Compare the performance of different versions of the knowledge graph among users, and further adjust and optimize based on feedback to ensure that the final knowledge graph can better meet user needs and provide a high-quality user experience.

[0075] Through this series of steps, we finally get the third knowledge graph. This third knowledge graph is optimized by user data, which is not only more reasonable in structure, but also better reflects the interests and needs of users.

[0076] Furthermore, the method provided in the application embodiment also includes:

[0077] A time decay factor for the user is established, and an intangible cultural heritage tour evaluation of the user is performed based on the time decay factor to establish the user's importance; a shared response model is configured according to the user's importance, and the third knowledge graph is called based on the shared response model to perform dynamic content matching.

[0078] In an embodiment of the present application, the interaction data of users when using the knowledge graph platform is collected through server log records. These data include user queries, clicks, browsing, feedback and other activities, and are accompanied by precise timestamps to record the time when the user behavior occurs. When calculating the attenuation factor, it is calculated by 1 / (1+t), where t is the number of days from the time the behavior occurs to the current time. Through this formula, a decay factor is assigned to each user behavior, which directly reflects the attenuation effect of time on user interest.

[0079] Next, the user's interactive behavior is weighted and scored based on the calculated temporal decay factor. First, each behavior score of the user is adjusted, that is, the original score of the behavior is multiplied by its corresponding temporal decay factor to obtain a weighted behavior score. For example, a user clicked on a page related to "Qiang Embroidery" a week ago. The original behavior score was 1 point. The temporal decay factor was applied for adjustment. If the decay factor is 0.125 points. Then, these weighted scores are added up to obtain the user's total weighted score for specific intangible cultural heritage content. In this way, the user's interest in different content can be identified, especially the impact of recent behavior can be emphasized.

[0080] Based on the user's browsing evaluation results, the weighted average method is used to determine the user's importance. The calculation of user importance depends on several key factors, including user activity, browsing breadth, and interaction depth. When calculating the scores of user activity, browsing breadth, and interaction depth, the user's activity score is calculated by weighted average of access frequency and duration. Access frequency reflects the number of times a user accesses the knowledge graph within a certain period of time, while access duration reflects the total length of time the user stays on the platform. The two scores are weighted averaged to obtain the user's activity score. Next, the types of intangible cultural heritage content visited by the user within a certain period of time are counted to calculate the user's browsing breadth score. The browsing breadth score is the ratio of the number of content types that the user actually contacts to the number of all possible content types in the system. This score reflects the breadth of the user's exploration on the platform, that is, whether the user is interested in diverse intangible cultural heritage content. If the user visits more content types, it means that his browsing breadth is large and the score is correspondingly high. Finally, the user's interaction depth score on specific content is calculated by weighted average of clicks and stay time. The click count score reflects the user's interest in a certain content, while the dwell time score indicates how deeply the user engages with the content. The weighted average of the click count and dwell time scores is used to obtain the user's interaction depth score for the content.

[0081] Each factor is then assigned a fixed weight, and the scores of each factor are multiplied by their weights and added up to get the user's importance score. For example, assuming that the weight of activity is 0.5, and the weights of breadth and depth are 0.3 and 0.2 respectively, if a user scores 0.8, 0.6, and 0.7 on these factors respectively, then their importance score is 0.8×0.5+0.6×0.3+0.7×0.2=0.71. This simple calculation method ensures that high-importance and low-importance users can be quickly identified and classified into different importance levels, with users with importance scores of 0.51-1.0 as high-importance users and users with importance scores of 0-0.5 as low-importance users.

[0082] After determining the importance of the user, the shared response model is configured by using the collaborative filtering method. The core of the collaborative filtering method is to recommend content that the current user may be interested in based on the user's historical behavior and the behavior of similar users. First, a user-content matrix is ​​constructed, in which each row represents a user and each column represents a content. The values ​​in the matrix represent the interaction intensity between the user and the content, such as the number of clicks or ratings. Then, the similarity between users is calculated by cosine similarity. In this way, which users are similar to the current user in behavior are identified, and based on the behavioral data of similar users, the content that the current user may be interested in is recommended. Under the guidance of the shared response model, the content that best suits the current user is dynamically matched from the third knowledge graph. For high-importance users, content that is highly relevant to their behavior is recommended first, while for low-importance users, some more extensive and exploratory content is recommended to encourage them to explore more intangible cultural heritage projects.

[0083] Through this series of steps, personalized dynamic content matching is achieved.

[0084] Step S600: Generate an activated representation based on the third knowledge graph for digital display and interaction of intangible cultural heritage.

[0085] In an embodiment of the present application, entities and relationships related to intangible cultural heritage items are extracted from the third knowledge graph using the SPARQL query language. Through SPARQL queries, key entities and relationships related to the user's current needs or interests are located from the knowledge graph. Next, the extracted entities and relationships are converted into visual graphs using a knowledge graph visualization tool. Knowledge graph visualization is to intuitively represent the graph structure through nodes and edges, so that users can clearly see the relationship between various entities. For example, "Qiang embroidery" is displayed as a node in the graph, and is connected to related nodes such as "historical origins" and "skill inheritance" through edges to show its position in the cultural context. Activated representations are further generated based on the visual graph. For example, HTML5 and JavaScript technology can be used to add interactive functions to each node and edge, such as displaying detailed information when clicking a node, or displaying the evolution and inheritance relationship between entities in the knowledge graph through animation effects.

[0086] In order to enhance the vividness of the intangible cultural heritage display, multimedia elements are combined with activated representation. For example, when a user clicks on the "Qiang Embroidery" node, relevant embroidery pictures, introduction videos or cultural music can be displayed. This multimedia integration is achieved through embedded code, making the user experience richer and more diverse. Then, through user interface design technology, the activated representation and multimedia content are integrated into a unified display platform. In order to achieve real-time interaction between users and knowledge graphs, a series of interactive functions are developed. These functions include clicking nodes to view details, dragging nodes to adjust the map layout, and zooming in and out of the map view. The development of interactive functions is implemented using JavaScript and CSS. Through these simple technologies, the map is given more dynamics and operability. For example, users can click on the "Qiang Embroidery" node to view detailed cultural background and skill inheritance information, and adjust its position in the map by dragging the node to better understand its relationship with other cultural elements.

[0087] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects:

[0088] The present application establishes a cloud database, which is used to store digital data of intangible cultural heritage, and the digital data includes multimodal data source data of text, pictures, audio, video, VR, AR and three-dimensional models; describes the digital data in the cloud database based on the ontology, and uses each ontology as an entity of the knowledge graph to establish a first knowledge graph with entities; configures a trust channel, and performs independent data authentication on the entities in the first knowledge graph based on the trust channel, and establishes data authentication results, wherein the data authentication results include authentication pass results, authentication fail results and fuzzy authentication results; activates the entity similarity authentication network, performs entity association calculation through the entity similarity authentication network, generates authentication call instructions based on the entity association calculation results, performs auxiliary authentication of authentication fail results through the authentication call instructions, and generates auxiliary authentication results; updates the first knowledge graph according to the data authentication results and the auxiliary authentication results, establishes a second knowledge graph, and establishes an objective function for embedding vectors, embeds the entities and relationships of the second knowledge graph into a low-dimensional vector space based on the objective function, updates the relationship weights of the second knowledge graph according to the embedding results, and establishes a third knowledge graph; generates an activated representation according to the third knowledge graph for digital display and interaction of intangible cultural heritage. The present invention solves the technical problem that the digital protection and display methods of intangible cultural heritage in the prior art cannot fully reflect the richness of intangible cultural heritage. It establishes a cloud database to store multimodal intangible cultural heritage data, constructs a first knowledge graph based on the ontology and performs data authentication, activates the entity similarity authentication network to perform association calculation and auxiliary authentication, and uses embedded vectors to optimize the knowledge graph relationship weights. Finally, it generates a third knowledge graph for digital display and interaction of intangible cultural heritage, thereby achieving the technical effect of improving the expression ability and display effect of the intangible cultural heritage knowledge graph.

[0089] Embodiment 2 is based on the same inventive concept as the method for constructing a multimodal intangible cultural heritage knowledge graph in the aforementioned embodiment. Figure 2 As shown, the present application provides a system for constructing a multimodal intangible cultural heritage knowledge graph, and the system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0090] A database establishment module 11, wherein the database establishment module 11 establishes a cloud database, wherein the cloud database is used to store digital data of intangible cultural heritage, wherein the digital data includes multimodal data source data of text, picture, audio, video, VR, AR and three-dimensional model; a first knowledge graph establishment module 12, wherein the first knowledge graph establishment module 12 describes the digital data in the cloud database based on the ontology, and uses each ontology as an entity of the knowledge graph to establish the first knowledge graph with the entities; an independent data authentication module 13, wherein the independent data authentication module 13 configures a trust channel, performs independent data authentication on the entities in the first knowledge graph based on the trust channel, and establishes a data authentication result, wherein the data authentication result includes an authentication pass result, an authentication fail result and a fuzzy authentication result; an auxiliary authentication module 14, wherein the The auxiliary authentication module 14 activates the entity similarity authentication network, performs entity association calculation through the entity similarity authentication network, generates an authentication call instruction based on the entity association calculation result, performs auxiliary authentication of the authentication failure result through the authentication call instruction, and generates an auxiliary authentication result; the knowledge graph update module 15, the knowledge graph update module 15 updates the first knowledge graph according to the data authentication result and the auxiliary authentication result, establishes the second knowledge graph, and establishes the objective function of the embedding vector, embeds the entities and relationships of the second knowledge graph into the low-dimensional vector space based on the objective function, updates the relationship weights of the second knowledge graph according to the embedding result, and establishes the third knowledge graph; the activation representation generation module 16, the activation representation generation module 16 generates an activation representation according to the third knowledge graph for the digital display and interaction of intangible cultural heritage.

[0091] Furthermore, the system is also used to implement the following functions:

[0092] Configure the objective function as follows:

[0093]

[0094] in, Represents positive triples, is the set of all positive triples, h is the head entity, r describes the association between the head entity and the tail entity, t is the tail entity, represents the negative example triple, h' represents the head entity after being replaced, r' describes the association after being replaced, t' represents the tail entity after being replaced, and γ represents the marginal parameter, which is used to control the function's ability to distinguish between positive and negative examples. Characterize the hierarchical embedding distance, d hier is the hierarchical distance function, are the embedding vectors of the head entity, the associated entity, and the tail entity, respectively. Characterizes context-dependent embedding distance, and are the hierarchical embedding distance and context-dependent embedding distance of negative examples respectively, and + is a positive function. The embedding vector is learned by minimizing the objective function, and the entities and relations of the second knowledge graph are embedded into a low-dimensional vector space.

[0095] Furthermore, the system is also used to implement the following functions:

[0096] The frequency statistics layer of the entity similarity authentication network is called to perform co-occurrence frequency and independent frequency statistics of entities, and a first degree of association is generated based on the co-occurrence frequency and independent frequency; the path identification layer of the entity similarity authentication network is called to perform shortest path length calculation between entities, and a second degree of association is generated according to the shortest path length calculation result; the cosine similarity calculation layer of the entity similarity authentication network is called to perform similarity calculation between entities and generate a third degree of association; and an entity association calculation result is generated according to the first degree of association, the second degree of association, and the third degree of association.

[0097] Furthermore, the system is also used to implement the following functions:

[0098] The data source in the entity is read by using the trust channel, and a source trust mark is generated based on the data source; the contradictory relationship of the data in the entity is identified based on the trust channel, and weighted authentication of the contradictory relationship identification is performed based on the source trust mark, and a data authentication result is established according to the weighted authentication result.

[0099] Furthermore, the system is also used to implement the following functions:

[0100] Call the abnormal authentication result in the auxiliary authentication result, and perform a threshold trigger determination of a negative impact on the abnormal authentication result; if the threshold of the negative impact is not triggered, perform a contribution evaluation based on the corresponding abnormal authentication result; generate reserved data based on the contribution evaluation result, and identify the reserved data as untrusted data; and save the reserved data in the auxiliary authentication result.

[0101] Furthermore, the system is also used to implement the following functions:

[0102] Establish a user data set, wherein the user data set includes a user direct feedback data set and a user interaction data set, wherein the user interaction data set includes query logs, click behaviors, and search preferences; perform feedback analysis based on the user data set and establish feedback analysis results; establish adaptive weight response feedback based on the feedback analysis results, and optimize the third knowledge graph based on the adaptive weight response feedback.

[0103] Furthermore, the system is also used to implement the following functions:

[0104] A time decay factor for the user is established, and an intangible cultural heritage tour evaluation of the user is performed based on the time decay factor to establish the user's importance; a shared response model is configured according to the user's importance, and the third knowledge graph is called based on the shared response model to perform dynamic content matching.

[0105] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0106] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0107] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for constructing a multimodal intangible cultural heritage knowledge graph, characterized in that: The method comprises: Establishing a cloud database, the cloud database is used to store digital data of intangible cultural heritage, the digital data includes multimodal data source data of text, pictures, audio, video, VR, AR and three-dimensional model; Describing digital data in a cloud database based on the ontology, and using each ontology as an entity of a knowledge graph, and establishing a first knowledge graph with the entities; Configure a trust channel, perform independent data authentication on entities in the first knowledge graph based on the trust channel, and establish a data authentication result, wherein the data authentication result includes an authentication pass result, an authentication fail result, and a fuzzy authentication result; Activate an entity similarity authentication network, perform entity association calculation through the entity similarity authentication network, generate an authentication call instruction based on the entity association calculation result, perform auxiliary authentication of the authentication failure result through the authentication call instruction, and generate an auxiliary authentication result; According to the data authentication result and the auxiliary authentication result, the first knowledge graph is updated, a second knowledge graph is established, and an objective function of embedding vectors is established. Based on the objective function, entities and relationships of the second knowledge graph are embedded into a low-dimensional vector space, and the relationship weights of the second knowledge graph are updated according to the embedding results to establish a third knowledge graph; An activated representation is generated based on the third knowledge graph for digital display and interaction of intangible cultural heritage.

2. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The objective function of establishing the embedding vector, embedding the entities and relations of the second knowledge graph into the low-dimensional vector space based on the objective function, further includes: Configure the objective function as follows: in, Represents positive triples, is the set of all positive triples, h is the head entity, r describes the association between the head entity and the tail entity, t is the tail entity, represents the negative example triple, h' represents the head entity after being replaced, r' describes the association after being replaced, t' represents the tail entity after being replaced, and γ represents the marginal parameter, which is used to control the function's ability to distinguish between positive and negative examples. Characterize the hierarchical embedding distance, d hier is the hierarchical distance function, are the embedding vectors of the head entity, the associated entity, and the tail entity, respectively. Characterizes context-dependent embedding distance, and They are the hierarchical embedding distance and context-dependent embedding distance of negative examples, respectively, and + is a positive function; The embedding vector is learned by minimizing the objective function, and the entities and relations of the second knowledge graph are embedded into a low-dimensional vector space.

3. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The performing entity association calculation through the entity similarity authentication network also includes: Calling the frequency statistics layer of the entity similarity authentication network to perform co-occurrence frequency and independent frequency statistics of the entity, and generating a first correlation degree based on the co-occurrence frequency and the independent frequency; Calling the path identification layer of the entity similarity authentication network to perform shortest path length calculation between entities, and generating a second degree of association according to the shortest path length calculation result; Calling the cosine similarity calculation layer of the entity similarity authentication network to perform similarity calculation between entities and generate a third degree of association; An entity association calculation result is generated according to the first association degree, the second association degree, and the third association degree.

4. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The performing independent data authentication on entities in the first knowledge graph based on the trust channel and establishing a data authentication result also includes: Using the trust channel to read the data source in the entity, and generating a source trust level identifier based on the data source; Based on the trust channel, contradictory relationships in the data within the entity are identified, weighted authentication of the contradictory relationship identification is performed based on the source trust identifier, and a data authentication result is established according to the weighted authentication result.

5. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The generating of the auxiliary authentication result further includes: Calling the abnormal authentication result in the auxiliary authentication result, and performing a threshold triggering determination of a negative impact on the abnormal authentication result; If the threshold of negative impact is not triggered, the contribution evaluation is performed based on the corresponding abnormal authentication results; Generate reserved data according to the contribution evaluation result, and identify the reserved data as untrusted data; The reserved data is stored in the auxiliary authentication result.

6. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The method further comprises: Establishing a user data set, wherein the user data set includes a user direct feedback data set and a user interaction data set, wherein the user interaction data set includes query logs, click behaviors, and search preferences; Perform feedback analysis based on the user data set and establish feedback analysis results; An adaptive weight response feedback is established based on the feedback analysis result, and the third knowledge graph is optimized based on the adaptive weight response feedback.

7. The method for constructing a multimodal intangible cultural heritage knowledge graph according to claim 1, characterized in that: The method further comprises: Establishing a time decay factor for the user, and evaluating the user's intangible cultural heritage tour based on the time decay factor to establish the user's importance; A sharing response model is configured according to the user importance, and the third knowledge graph is called based on the sharing response model to perform dynamic content matching.

8. A multimodal intangible cultural heritage knowledge graph construction system, characterized by: The system comprises: A database establishment module, wherein the database establishment module establishes a cloud database, wherein the cloud database is used to store digital data of intangible cultural heritage, wherein the digital data includes multimodal data source data of text, pictures, audio, video, VR, AR and three-dimensional models; A first knowledge graph building module, which describes digital data in a cloud database based on an ontology, and uses each ontology as an entity of a knowledge graph to build a first knowledge graph with the entities; An independent data authentication module, wherein the independent data authentication module configures a trust channel, performs independent data authentication on entities in the first knowledge graph based on the trust channel, and establishes a data authentication result, wherein the data authentication result includes an authentication pass result, an authentication fail result, and a fuzzy authentication result; An auxiliary authentication module, wherein the auxiliary authentication module activates an entity similarity authentication network, performs entity association calculation through the entity similarity authentication network, generates an authentication call instruction based on the entity association calculation result, performs auxiliary authentication of the authentication failure result through the authentication call instruction, and generates an auxiliary authentication result; A knowledge graph updating module, wherein the knowledge graph updating module updates the first knowledge graph according to the data authentication result and the auxiliary authentication result, establishes a second knowledge graph, and establishes an objective function of embedding vectors, embeds entities and relationships of the second knowledge graph into a low-dimensional vector space based on the objective function, updates the relationship weights of the second knowledge graph according to the embedding result, and establishes a third knowledge graph; An activation representation generation module, wherein the activation representation generation module generates an activation representation based on the third knowledge graph for digital display and interaction of intangible cultural heritage.

Images