A method and system for migrating unstructured media asset data for converged media platforms

Abstract

This invention proposes a method and system for migrating unstructured media asset data for converged media platforms. By integrating multimodal deep learning and knowledge graph technologies, it achieves intelligent and semantic migration of media asset data. First, a pre-trained multimodal deep learning model is used to perform semantic parsing on the source media assets, generating high-dimensional embedding vectors. Then, a knowledge graph is constructed based on the target platform's standardized metadata model, and entity linking technology is used to dynamically associate media assets with graph entities. Next, cosine similarity calculation and threshold rules are used for intelligent classification mapping, combined with a storage strategy recommendation model to optimize data storage. Finally, media asset files, metadata, semantic vectors, and knowledge associations are encapsulated into a knowledge-enhanced data package to complete the migration. This invention significantly improves the intelligence level and data integrity of cross-platform media asset migration.

Description

Technical Field

[0001] This invention relates to the field of data migration, and in particular to a method and system for migrating unstructured media asset data for converged media platforms. Background Technology

[0002] As media convergence deepens, converged media platforms need to aggregate and manage massive amounts of unstructured media asset data (such as images, videos, audio, and documents) from diverse channels including traditional media, websites, mobile devices, and social media. This data is typically stored in heterogeneous and isolated legacy systems with outdated management models and fragmented semantic information, creating significant data barriers.

[0003] Existing migration technologies mostly rely on simple file copying and basic metadata transfer, which has the following drawbacks: 1. Semantic Fragmentation and Information Degradation: Existing technologies can only process surface metadata such as filenames, sizes, and creation dates, and cannot understand and extract deep semantic information at the content level of media assets (such as scene content, involved characters and events). This results in migrated resources becoming "information silos" lacking contextual relevance, greatly hindering subsequent intelligent retrieval, content recommendation, and knowledge discovery.

[0004] 2. Rigid Mapping and Poor Adaptability: Traditional classification mapping mainly relies on manually preset keyword rules, resulting in high rule base maintenance costs and an inability to handle new concepts, ambiguities, and synonyms. It lacks a mapping mechanism that can understand the essence of the content and automatically adapt to the dynamic data model of the target platform.

[0005] 3. Storage strategy is disconnected from business value: The allocation of target storage space is often based solely on file size or type, failing to correlate with the business value of media assets (such as access popularity and content importance), resulting in low storage resource utilization efficiency and high-value assets failing to receive priority performance guarantees. Summary of the Invention

[0006] To address the aforementioned problems, the present invention aims to provide a method and system for migrating unstructured media asset data for converged media platforms, thereby improving data migration efficiency and reliability.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for migrating unstructured media asset data for converged media platforms includes the following steps: S1: Perform multimodal parsing on the source media asset file. First, extract basic technical metadata, and then use a pre-trained multimodal deep learning model to map the media asset content to a high-dimensional vector space to generate its semantic embedding vector. S2: Extract entities and relationships from the target platform's standardized metadata model (such as classification system, people database, location database, event database) and the metadata of the source data to construct an initial knowledge graph. During the migration process, using the semantic embedding vectors generated in step S1, entity linking and relationship extraction techniques are used to dynamically associate the media assets to be migrated with entities in the graph. S3: Intelligent mapping based on cosine similarity. During mapping, the cosine similarity between the semantic vector of the media asset to be migrated and the vector of each target node is calculated. Combined with the preset threshold rules and the association relationship in the knowledge graph, it is automatically assigned to the most relevant target classification node. S4: Build a storage strategy recommendation model to store the categorized data; S5: Encapsulate the media asset file, basic metadata, its semantic vector, and the association ID in the knowledge graph into a knowledge-enhanced data package. Then, call the target platform API to migrate this data package and the media file entity to the specified storage location. S6: After migration, perform file integrity verification, and verify the semantic vector similarity between the source and target media assets, and verify whether their association in the target platform's knowledge graph is correct, to ensure that the migration process does not cause the loss or distortion of semantic information.

[0008] Furthermore, multimodal parsing is performed on the source media asset files. First, basic technical metadata is extracted, and then a pre-trained multimodal deep learning model is used to map the media asset content to a high-dimensional vector space, generating its semantic embedding vector, as follows: In the initial stage of multimodal parsing, a comprehensive technical analysis of the source media asset files is first performed, automatically extracting basic technical metadata including file format, resolution, duration, frame rate, encoding method, file size, and creation time. Pre-trained multimodal deep learning models (such as CLIP and ALIGN) are then used to perform deep semantic understanding of the actual content of the media assets. For image content, the model extracts visual features through convolutional neural networks to identify semantic elements such as objects, scenes, and facial expressions. For audio content, recurrent neural networks are used to analyze audio waveforms and extract speech content, music style, and ambient sound auditory features. For video content, temporal modeling techniques are combined to capture inter-frame relationships and dynamic change information. After deep processing by the multimodal model, the extracted multi-level feature information is mapped to a unified high-dimensional vector space to generate a fixed-length semantic embedding vector (usually 512-dimensional, 768-dimensional, or 1024-dimensional).

[0009] Furthermore, entities and relationships are extracted from the standardized metadata model of the target platform and the metadata of the source data to construct an initial knowledge graph, as follows: We conduct in-depth analysis of the standardized metadata model of the target platform, and obtain the platform's existing structured knowledge resources, including classification systems, people databases, location databases, and event databases, through API calls. For each knowledge base, we use named entity recognition and relation extraction algorithms to automatically identify core entities (such as people's names, geographical locations, event names, category tags, etc.) and their attribute information (such as people's occupations, location coordinates, event times, etc.). At the same time, we analyze the existing metadata fields in the source data, including titles, descriptions, tags, and category text information, and use natural language processing technology to extract the entities and semantic relationships contained therein, laying the data foundation for building a unified knowledge graph. Based on the extracted entity and relation information, an initial knowledge graph is constructed using graph database technology (such as Neo4j or a self-developed graph engine). The graph uses entities as nodes and relations as edges to form a network-like knowledge structure. During the construction process, a unified ontology model is established to define the hierarchical structure of entity types (such as the inheritance relationship of character-actor-director) and the semantic constraints of relation types (such as the "acting" relationship connecting characters and works).

[0010] Furthermore, during the migration process, the semantic embedding vectors generated in step S1 are used to dynamically associate the media assets to be migrated with entities in the knowledge graph through entity linking and relation extraction techniques. Specifically, during the media asset migration process, the system utilizes the semantic embedding vectors generated in step S1 to achieve intelligent association between media asset content and knowledge graph entities through advanced entity linking technology. First, each entity in the knowledge graph is also converted into a corresponding semantic vector representation to construct an entity embedding space. Then, the similarity between the semantic vector of the media asset to be migrated and the entity vector of the knowledge graph is calculated to identify potential associated entities. Combining the cosine similarity method and introducing a context-aware mechanism, the accuracy of the linking is ensured. At the same time, the system uses relation extraction technology to discover implicit relationships between media assets and linked entities based on semantic vector analysis, dynamically expanding the relation network of the knowledge graph.

[0011] Furthermore, the storage strategy recommendation model comprehensively considers multiple factors, as follows: Storage Tier=F(Access Frequency,Semantic Centrality,File Size) Here, Semantic Centrality is the semantic centrality, which measures the global importance of the media asset by calculating its PageRank value in the knowledge graph; File Size is the file size; Access Frequency is the access frequency; based on this model, media assets with Storage Tie values ​​greater than the threshold are allocated to high-performance storage, while those with lower values ​​are allocated to lower-cost cold storage.

[0012] A system for migrating unstructured media asset data for converged media platforms includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for migrating unstructured media asset data for converged media platforms as described above.

[0013] The present invention has the following beneficial effects: This invention. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a method for migrating unstructured media asset data for converged media platforms is provided, including the following steps: S1: Perform multimodal parsing on the source media asset file. First, extract basic technical metadata, and then use a pre-trained multimodal deep learning model to map the media asset content to a high-dimensional vector space to generate its semantic embedding vector. S2: Extract entities and relationships from the target platform's standardized metadata model (such as classification system, people database, location database, event database) and the metadata of the source data to construct an initial knowledge graph. During the migration process, using the semantic embedding vectors generated in step S1, entity linking and relationship extraction techniques are used to dynamically associate the media assets to be migrated with entities in the graph. S3: Intelligent mapping based on cosine similarity. During mapping, the cosine similarity between the semantic vector of the media asset to be migrated and the vector of each target node is calculated. Combined with the preset threshold rules and the association relationship in the knowledge graph, it is automatically assigned to the most relevant target classification node. S4: Build a storage strategy recommendation model to store the categorized data; S5: Encapsulate the media asset file, basic metadata, its semantic vector, and the association ID in the knowledge graph into a knowledge-enhanced data package. Then, call the target platform API to migrate this data package and the media file entity to the specified storage location. S6: After migration, perform file integrity verification, and verify the semantic vector similarity between the source and target media assets, and verify whether their association in the target platform's knowledge graph is correct, to ensure that the migration process does not cause the loss or distortion of semantic information.

[0016] In this embodiment, multimodal parsing is performed on the source media asset file. First, basic technical metadata is extracted, and then a pre-trained multimodal deep learning model is used to map the media asset content to a high-dimensional vector space to generate its semantic embedding vector, as follows: In the initial stage of multimodal parsing, a comprehensive technical analysis of the source media asset files is first performed, automatically extracting basic technical metadata including file format, resolution, duration, frame rate, encoding method, file size, and creation time. Pre-trained multimodal deep learning models (such as CLIP and ALIGN) are then used to perform deep semantic understanding of the actual content of the media assets. For image content, the model extracts visual features through convolutional neural networks to identify semantic elements such as objects, scenes, and facial expressions. For audio content, recurrent neural networks are used to analyze audio waveforms and extract speech content, music style, and ambient sound auditory features. For video content, temporal modeling techniques are combined to capture inter-frame relationships and dynamic change information. After deep processing by the multimodal model, the extracted multi-level feature information is mapped to a unified high-dimensional vector space to generate a fixed-length semantic embedding vector (usually 512-dimensional, 768-dimensional, or 1024-dimensional).

[0017] In this embodiment, entities and relationships are extracted from the standardized metadata model of the target platform and the metadata of the source data to construct an initial knowledge graph, as detailed below: We conduct in-depth analysis of the standardized metadata model of the target platform, and obtain the platform's existing structured knowledge resources, including classification systems, people databases, location databases, and event databases, through API calls. For each knowledge base, we use named entity recognition and relation extraction algorithms to automatically identify core entities (such as people's names, geographical locations, event names, category tags, etc.) and their attribute information (such as people's occupations, location coordinates, event times, etc.). At the same time, we analyze the existing metadata fields in the source data, including titles, descriptions, tags, and category text information, and use natural language processing technology to extract the entities and semantic relationships contained therein, laying the data foundation for building a unified knowledge graph. Based on the extracted entity and relation information, an initial knowledge graph is constructed using graph database technology (such as Neo4j or a self-developed graph engine). The graph uses entities as nodes and relations as edges to form a network-like knowledge structure. During the construction process, a unified ontology model is established to define the hierarchical structure of entity types (such as the inheritance relationship of character-actor-director) and the semantic constraints of relation types (such as the "acting" relationship connecting characters and works).

[0018] In this embodiment, during the migration process, the semantic embedding vectors generated in step S1 are used to dynamically associate the media assets to be migrated with entities in the knowledge graph through entity linking and relation extraction techniques. Specifically, during the media asset migration process, the system uses the semantic embedding vectors generated in step S1 to achieve intelligent association between media asset content and knowledge graph entities through advanced entity linking technology. First, each entity in the knowledge graph is also converted into a corresponding semantic vector representation to construct an entity embedding space. Then, the similarity between the semantic vector of the media assets to be migrated and the entity vector of the knowledge graph is calculated to identify potential associated entities. The cosine similarity method is combined with a context-aware mechanism to ensure the accuracy of the linking. At the same time, the system uses relation extraction technology to discover implicit relationships between media assets and linked entities based on semantic vector analysis, dynamically expanding the relation network of the knowledge graph.

[0019] In this embodiment, the storage strategy recommendation model comprehensively considers multiple factors, as follows: Storage Tier=F(Access Frequency,Semantic Centrality,File Size) Here, Semantic Centrality is the semantic centrality, which measures the global importance of the media asset by calculating its PageRank value in the knowledge graph; File Size is the file size; Access Frequency is the access frequency; based on this model, media assets with Storage Tie values ​​greater than the threshold are allocated to high-performance storage, while those with lower values ​​are allocated to lower-cost cold storage.

[0020] A system for migrating unstructured media asset data for converged media platforms includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for migrating unstructured media asset data for converged media platforms as described above.

[0021] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0022] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0023] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0024] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for migrating unstructured media asset data for converged media platforms, characterized in that, Includes the following steps: S1: Perform multimodal parsing on the source media asset file. First, extract basic technical metadata, and then use a pre-trained multimodal deep learning model to map the media asset content to a high-dimensional vector space to generate its semantic embedding vector. S2: Extract entities and relationships from the standardized metadata model of the target platform and the metadata of the source data to construct a knowledge graph. During the migration process, using the semantic embedding vectors generated in step S1, the media asset content to be migrated is dynamically associated with entities in the knowledge graph through entity linking and relationship extraction techniques. S3: Intelligent mapping based on cosine similarity. During mapping, the cosine similarity between the semantic embedding vector of the media asset content to be migrated and the vector of each target node is calculated. Combined with the preset threshold rules and the association relationship in the knowledge graph, it is automatically assigned to the most relevant target classification node. S4: Build a storage strategy recommendation model to store the categorized data; S5: Encapsulate the source media asset file, basic technical metadata, semantic embedding vector, and the relationship ID in the knowledge graph into a knowledge-enhanced data package. Then, call the target platform API to migrate this knowledge-enhanced data package and the source media asset file entity to the specified storage location. S6: After migration, perform source media asset file integrity verification, and verify the semantic embedding vector similarity between the source and target media asset content, and verify whether their association in the target platform knowledge graph is correct, to ensure that the migration process does not cause loss or distortion of semantic information. The storage strategy recommendation model comprehensively considers multiple factors, as follows: Storage Tier=F(Access Frequency,Semantic Centrality,File Size) Among them, Semantic Centrality is the global importance of the media asset content to be migrated, which is measured by calculating the PageRank value of the content in the knowledge graph; File Size is the size of the source media asset file to be migrated; Access Frequency is the access frequency; based on this, the storage strategy recommendation model allocates source media asset content with a Storage Tier value greater than the threshold to high-performance storage, and otherwise allocates it to lower-cost cold storage.

2. The method for migrating unstructured media asset data for converged media platforms according to claim 1, characterized in that, The multimodal parsing of the source media asset file first extracts basic technical metadata, and then uses a pre-trained multimodal deep learning model to map the media asset content to a high-dimensional vector space to generate its semantic embedding vector, as detailed below: In the initial stage of multimodal parsing, a comprehensive technical analysis of the source media asset files is first performed, automatically extracting basic technical metadata including file format, resolution, duration, frame rate, encoding method, file size, and creation time. A pre-trained multimodal deep learning model is then used to perform deep semantic understanding of the actual content of the source media asset files. For image content, the multimodal deep learning model extracts visual features through convolutional neural networks to identify semantic elements such as objects, scenes, and facial expressions in the images. For audio content, recurrent neural networks are used to analyze audio waveforms and extract speech content, music style, and ambient sound auditory features. For video content, temporal modeling techniques are combined to capture inter-frame relationships and dynamic change information. After deep processing by a multimodal deep learning model, the extracted multi-level feature information is mapped onto a unified high-dimensional vector space to generate a fixed-length semantic embedding vector.

3. The method for migrating unstructured media asset data for converged media platforms according to claim 1, characterized in that, The process of extracting entities and relationships from the standardized metadata model of the target platform and the metadata of the source data to construct a knowledge graph is as follows: We conduct in-depth analysis of the standardized metadata model of the target platform, and obtain the existing structured knowledge resources of the target platform's classification system library, people library, location library, and event library through API calls. For each library, we use named entity recognition and relation extraction algorithms to automatically identify the core entities and their attribute information. At the same time, we analyze the existing metadata fields in the source data, including titles, descriptions, tags, and classification text information, and use natural language processing technology to extract the entities and semantic relationships contained therein, laying the data foundation for building a unified knowledge graph. Based on the extracted entity and relation information, a knowledge graph is constructed using graph database technology. The knowledge graph uses entities as nodes and relations as edges to form a network-like knowledge structure. During the construction process, a unified ontology model is established, defining the hierarchical structure of entity types and the semantic constraints of relation types.

4. The method for migrating unstructured media asset data for converged media platforms according to claim 3, characterized in that, During the migration process, the semantic embedding vectors generated in step S1 are used to dynamically associate the media asset content to be migrated with entities in the knowledge graph through entity linking and relation extraction techniques. Specifically, during the media asset content migration process, the semantic embedding vectors generated in step S1 are used to achieve intelligent association between the media asset content and entities in the knowledge graph through advanced entity linking techniques. First, each entity in the knowledge graph is also converted into a corresponding semantic vector representation to construct an entity embedding space. Then, the similarity between the semantic embedding vector of the media asset content to be migrated and the entity vector of the knowledge graph is calculated to identify potential associated entities. The cosine similarity method is combined with a context-aware mechanism to ensure the accuracy of the link. At the same time, relation extraction techniques are used to discover the implicit relationships between the media asset content and the linked entities based on semantic vector analysis, dynamically expanding the relation network of the knowledge graph.

5. A data migration system for unstructured media assets for converged media platforms, characterized in that, It includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps in the method for migrating unstructured media asset data for a converged media platform as described in any one of claims 1-4.

Images