Enterprise marketing-oriented digital asset knowledge graph interactive query system

Abstract

The application discloses an enterprise marketing-oriented digital asset knowledge graph interactive query system and relates to the technical field of intelligent retrieval interaction. The application obtains the multi-modal digital asset original data of an enterprise, extracts digital asset entities and their semantic association relations from the multi-modal digital asset original data, further constructs a digital asset knowledge graph, divides multiple asset association sub-domains based on the subject direction of enterprise marketing, sets hierarchical query permissions for access users, establishes a mapping relationship between the hierarchical query permissions and the digital asset knowledge graph, further sets the access reachable range of each access user, obtains user query requirements, positions a starting query anchor point in the digital asset knowledge graph based on the user query requirements, further divides multiple interactive query paths, traverses in the digital asset knowledge graph along the interactive query paths, and integrates the traversal results of each interactive query path.

Description

Technical Field

[0001] This invention relates to the field of intelligent retrieval and interaction technology, specifically to a digital asset knowledge graph interactive query system for enterprise marketing. Background Technology

[0002] Against the backdrop of the rapid development of the digital marketing industry, enterprises have accumulated massive amounts of multimodal digital assets across various marketing campaigns, encompassing copywriting, visual materials, transaction data, and more. Currently, most enterprises adopt a decentralized storage and management model for digital assets, with asset data from different modalities and channels existing independently, forming multiple data silos and lacking a unified semantic association system. Existing asset queries largely rely on keyword searches and manual screening, resulting in low query efficiency and difficulty in uncovering implicit marketing connections between cross-modal data, thus failing to fully unlock the business value of digital assets.

[0003] Meanwhile, some enterprises have introduced knowledge graph technology for asset management. However, existing marketing knowledge graphs mostly remain at the level of general entity associations, lacking marketing-specific entity types and refined relationship definitions. Furthermore, they do not perform subdomain division for different marketing business themes, resulting in high query path redundancy and insufficient accuracy. In addition, most systems lack hierarchical access control mechanisms, failing to adapt to the differentiated query needs of personnel at different positions and levels within an enterprise. This poses a risk of sensitive marketing data leakage and makes it difficult to meet enterprises' needs for refined and secure marketing asset management and interactive querying. Therefore, we provide a digital asset knowledge graph interactive query system for enterprise marketing. Summary of the Invention

[0004] The purpose of this invention is to provide an interactive query system for digital asset knowledge graphs for enterprise marketing, in order to address the shortcomings in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A digital asset knowledge graph interactive query system for enterprise marketing, including a knowledge graph construction module, an interaction permission planning module, a query path planning module, and a query feedback module; The knowledge graph construction module is used to acquire the original data of the enterprise's multimodal digital assets, extract digital asset entities and their semantic relationships from the original data of the multimodal digital assets, and then construct a digital asset knowledge graph, and divide multiple asset-related subdomains based on the enterprise's marketing subject direction. The interactive permission planning module is used to set hierarchical query permissions for accessing users, establish a mapping relationship between hierarchical query permissions and digital asset knowledge graph, and then set the access range of each accessing user. The query path planning module is used to obtain user query requirements, locate the starting query anchor point in the digital asset knowledge graph based on user query requirements, and then divide multiple interactive query paths. The query feedback module is used to traverse the digital asset knowledge graph along the interactive query path, integrate the traversal results of each interactive query path, and feed them back to the corresponding user.

[0006] Furthermore, the interactive query system for digital asset knowledge graphs aimed at enterprise marketing involves the following process for acquiring raw data on an enterprise's multimodal digital assets: The original data of the multimodal digital assets includes text modality, visual modality, and digital and structured modality; Spatiotemporal alignment processing is performed on different modal data under the same marketing event. Taking the marketing event as the core linking anchor, each piece of original multimodal digital asset data is first labeled with event tags and timestamps. At the same time, the scene spatial attributes of the original multimodal digital asset data are matched. For visual modal data, text information, product elements and brand logos in images and videos are extracted in advance through optical character recognition and multimodal content understanding technologies and converted into structured text information for entity extraction.

[0007] Furthermore, the process of extracting digital asset entities and their semantic relationships based on the original data of multimodal digital assets includes: The entity extraction is divided into two categories: regular entity extraction and marketing-specific entity extraction. The conventional entity extraction is combined with industry-specific dictionaries to extract common entity types from text and structured data. The marketing-specific entities are customized for enterprise marketing business scenarios and include exclusive entity types. The relation extraction is divided into two categories: explicit relation extraction and implicit relation extraction. Explicit relations are directly generated from structured and semi-structured data and stored in the form of standardized triples. Implicit relations are obtained through cross-modal semantic reasoning and association mining.

[0008] Furthermore, the construction process of the digital asset knowledge graph includes: The digital asset knowledge graph includes an entity layer, a relationship layer, and a rule and constraint layer. The entity layer contains several entity nodes. The number of nodes is related to the types of digital asset entities extracted. Each entity node comes with complete attribute information. The relationship layer consists of directed edges connecting entity nodes. According to business semantics, it is divided into three categories: causal edges, which are used to identify the causal relationship between marketing actions and results. Relationship edges are used to identify the association relationships between entities, such as benchmarking, substitution, and subordination. Edge attributes include association strength and association scenario. Temporal edges are used to identify the sequential relationships within a marketing process. The rules and constraints layer is used to store marketing business rules and graph reasoning constraints, including time range constraints for marketing activities, data validity period constraints, derivation rules for user segmentation, and calculation rules for marketing effects, providing rule support for subsequent asset-related subdomain division, path reasoning, and result verification.

[0009] Furthermore, the process of dividing multiple asset-related subdomains based on the enterprise's marketing focus includes: The division of the multi-asset related subdomains adopts an input-driven approach, with the enterprise's current marketing focus as the core division target; First, locate the entity nodes directly related to each marketing entity in the digital asset knowledge graph and record them as seed nodes. Starting from the seed nodes, multi-level diffusion is carried out along various relationship edges of the relationship layer. At the same time, a diffusion stopping condition is set: when the semantic relevance between the entity node diffused to and the marketing entity is lower than a preset threshold, the diffusion in that direction is stopped. During the diffusion process, the semantic relevance between adjacent entity nodes is calculated in real time. An aggregation threshold is set. When the semantic correlation between two entity nodes is greater than or equal to the aggregation threshold, the two diffusion nodes are assigned to the same candidate set. Otherwise, they are not assigned to the same candidate set. After all diffusion nodes have been calculated, all candidate sets are aggregated to form multiple independent multi-asset related subdomains. For entity nodes that overlap, their affiliation is determined based on their average semantic correlation with different subdomains. The subdomain with higher correlation is assigned to them.

[0010] Furthermore, the process of setting tiered query permissions for users, establishing a mapping relationship between these permissions and the digital asset knowledge graph, and defining the access range for each user includes: Access users are analyzed through a multi-dimensional role breakdown, encompassing four dimensions: job level, business line affiliation, data security level authorization, and project-specific permissions. Job levels are categorized hierarchically from entry-level staff to middle management and senior decision-makers. Business lines are divided into departments such as marketing, creative, sales, and finance. Data security level authorization is categorized into four levels: public, internal, confidential, and top secret. Project-specific permissions are granted temporarily and can be revoked for specific marketing projects. After this multi-dimensional breakdown, a tiered query permission tag set is generated for each access user, containing the corresponding permission level tag and scope of effectiveness for each dimension. Meanwhile, the interactive permission planning module sets up a hierarchical labeling system in the digital asset knowledge graph. The labels are divided into two categories: entity hierarchical and relationship hierarchical. A direct mapping rule is established between hierarchical query permission tags and hierarchical labels in the digital asset knowledge graph. Next, the reachability range of the four dimensions of user job level, business line, data security level, and project permission is intersected. After level matching and range division, the access reachability range corresponding to each accessing user is finally generated.

[0011] Furthermore, the process of locating the initial query anchor point in the digital asset knowledge graph based on user query requirements includes: Natural language understanding is used to process user queries and identify three key types of information: query intent, core entities, and constraints. The system locates the starting query anchor point in the digital asset knowledge graph, links the core entities in the user's query requirements with entity nodes in the digital asset knowledge graph, and uses semantic relevance calculation to match entity nodes as the starting query anchor point. During the location process, permission pre-verification is performed simultaneously. If the matched starting anchor point is outside the user's access range, an insufficient permission prompt is returned, and similar anchor points within the user's access range are recommended for the user to choose from.

[0012] Furthermore, the process of dividing multiple interactive query paths includes: After locating the starting query anchor point, the query path planning module performs a breadth-first traversal along the relation edges of the digital asset knowledge graph, generating multiple candidate query paths. During the traversal, permission pruning is performed simultaneously to remove entity nodes and relation edges that are beyond the user's access range, ensuring that all candidate paths are within the user's access range. Redundancy pruning is also performed to remove redundant paths that exceed the preset limit or have entity nodes that are repeatedly traversed. For the filtered candidate paths, the query path planning module calculates the query matching degree value of each path, sets a path filtering threshold, and includes candidate paths with query matching degree values ​​greater than or equal to the filtering threshold into the effective path set. The paths are then sorted from high to low according to their query matching degree values, and the top 5 candidate paths are selected as the final interactive query paths. At the same time, the next interactive query path is used as the alternative path to the previous one according to the sorting. If the previous interactive query path becomes unreachable due to missing data or permission changes, the system automatically switches to the corresponding alternative path to continue the query. After generating the interactive query path, the query feedback module performs a deep traversal of the digital asset knowledge graph along each interactive query path, and pulls the attribute information and relational data of all entity nodes passed through the interactive query path. After the traversal is completed, the traversal results of each interactive query path are integrated; first, the results are deduplicated, and entities and relationships that appear repeatedly in different interactive query paths are merged.

[0013] Furthermore, the process of traversing the digital asset knowledge graph along the interactive query paths and integrating the traversal results of each interactive query path includes: After the results are integrated, the query feedback module will use appropriate adaptation methods to provide feedback to the corresponding users based on their roles and query scenarios. These feedback methods include visual knowledge graphs, structured reports, multimodal resource packages, and natural language summaries.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a three-layer digital asset knowledge graph by aligning multimodal digital assets in time and space and extracting entity relationships. This can effectively uncover implicit marketing associations across modalities. At the same time, it divides multiple asset association subdomains based on the marketing subject direction to realize the scenario-based aggregation of assets, which improves the accuracy and relevance of marketing asset queries to a certain extent and provides comprehensive data support for enterprise marketing decisions.

[0015] 2. This invention achieves refined data access control through a hierarchical permission mapping mechanism, adapting to the query needs of different job roles while ensuring the security of sensitive marketing data; combined with intelligent query path planning and multi-form result feedback, it effectively improves query efficiency and result readability, and enhances the reuse efficiency and management value of enterprise digital assets. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, the digital asset knowledge graph interactive query system for enterprise marketing includes a knowledge graph construction module, an interactive permission planning module, a query path planning module, and a query feedback module. The knowledge graph construction module is used to acquire the original data of the enterprise's multimodal digital assets, extract digital asset entities and their semantic relationships from the original data of the multimodal digital assets, and then construct a digital asset knowledge graph, and divide multiple asset-related subdomains based on the enterprise's marketing subject direction. The interactive permission planning module is used to set hierarchical query permissions for accessing users, establish a mapping relationship between hierarchical query permissions and digital asset knowledge graph, and then set the access range of each accessing user. The query path planning module is used to obtain user query requirements, locate the starting query anchor point in the digital asset knowledge graph based on user query requirements, and then divide multiple interactive query paths. The query feedback module is used to traverse the digital asset knowledge graph along the interactive query path, integrate the traversal results of each interactive query path, and feed them back to the corresponding user.

[0020] The working principle of the present invention is illustrated below through examples: The enterprise marketing department synchronizes the original data of multimodal digital assets to the knowledge graph construction module. The original data of multimodal digital assets includes text modality, visual modality, and digital and structured modality. It should be noted that text modal data primarily originates from the enterprise's internal marketing content library, customer service ticket system, social media monitoring platform, market research database, and contract management system. Specifically, it covers official brand marketing copy, customer service dialogue records, social media user comments and reposts, third-party market research reports, channel cooperation and advertising contracts, etc. Visual modal data is mainly stored in the enterprise's digital material management platform and various advertising back-ends, including brand image promotional images, high-definition posters for the entire product line, short video marketing content, information flow advertising materials, and complete screenshots of advertising landing pages, etc. Digital and structured modal data originate from the enterprise's business middle platform and data warehouse, including customer transaction records across the entire value chain, exposure, click-through rate, conversion rate, and return on investment data from various advertising platforms, user membership levels, and consumption tag data, etc. All original multimodal digital asset data undergoes data anonymization and compliance verification before being integrated into the system. Sensitive information involving user privacy is anonymized or encrypted for storage to ensure that data collection and use comply with relevant data security regulations. After receiving the raw data of multimodal digital assets, the knowledge graph construction module first performs spatiotemporal alignment processing on the data of different modalities under the same marketing event: The spatiotemporal alignment process uses marketing events as the core anchor points. First, each piece of original multimodal digital asset data is labeled with an event tag and a timestamp. For example, "618 New Product Launch" is taken as the core marketing event. The new product posters, corresponding advertising copy, transaction data after the launch, and user comments released during the core marketing event period are all marked with the same event tag and matched with the corresponding time nodes. Simultaneously, the scene spatial attributes of the original data of multimodal digital assets are matched. For example, the materials related to offline store activities are spatially aligned with the transaction data of the corresponding region to ensure that cross-modal data are in the same marketing spatiotemporal dimension, providing a unified data foundation for subsequent entity and relationship extraction. For visual modal data, text information, product elements and brand logos in images and videos are extracted in advance through optical character recognition and multimodal content understanding technologies, and converted into structured text information that can be used for entity extraction, so as to realize the semantic connection between visual modality and text modality. After completing the spatiotemporal alignment, the knowledge graph construction module sequentially performs entity extraction and relation extraction. The entity extraction is divided into two categories: regular entity extraction and marketing-specific entity extraction. The conventional entity extraction is combined with industry-specific dictionaries to extract common entity types such as "brand name", "product model", "spokesperson name", "channel name" and "user ID" extracted from text and structured data; The marketing-specific entities are custom-developed for enterprise marketing business scenarios, including exclusive entity types such as "marketing activities", "placement locations", "conversion nodes", "user segmentation groups", "marketing selling points" and "competitor brands". For example, the "marketing activity" entity is identified by the rule combination of "activity name + activity time + activity budget", and user segmentation entities such as "high-value old customers" and "potential new customers" are identified by the aggregation of user consumption behavior tags. The relationship extraction is divided into two categories: explicit relationship extraction and implicit relationship extraction. Explicit relationships are directly generated from structured and semi-structured data and stored in the form of standardized triples. For example, relationships such as "User A-Purchase-Product B" and "Order C-Attribution-Marketing Activity D" are directly mapped from transaction flow data, and relationships such as "Distributor E-Agent-Brand F" are mapped from organizational structure data. These relationships have clear data source support and are marked with the highest confidence level. Implicit relationships are obtained through cross-modal semantic reasoning and association mining. By supplementing explicit relationships, such as by the temporal correlation between the peak of short video material playback volume and subsequent product search volume, a causal implicit relationship of "short video material G - driving - product H search volume increase" is mined. By keyword co-occurrence and semantic sentiment analysis in user comments, an association implicit relationship of "product I - user evaluation association - selling point J" is mined. Implicit relationships need to undergo confidence assessment. Relationships that are cross-validated through multiple data sources are marked as high confidence, while relationships obtained through inference from only a single data source are marked as low confidence and labeled with a verification mark. Subsequent data will be added to continuously verify and update the relationships.

[0021] After the entity and relation extraction is completed, the knowledge graph construction module builds a three-layer digital asset knowledge graph, which consists of the entity layer, relation layer and rule and constraint layer from bottom to top. The entity layer contains several entity nodes. The number of nodes is related to the types of digital asset entities extracted. Each entity node comes with complete attribute information. For example, product entities come with attributes such as specifications, price, launch time, and core selling points, while user entities come with attributes such as consumption level, region, and preference tags. The relationship layer consists of directed edges connecting entity nodes, which are divided into three categories according to business semantics: causal edges, which are used to identify the causal relationship between marketing actions and results, such as "ad placement - pull → sales increase" and "coupon issuance - promotion → user conversion". The edge attributes include causal confidence, influence period, and effect coefficient. Relationship edges are used to identify the relationships between entities, such as benchmarking, substitution, and subordination. For example, "Competitor A - Benchmark → This Product B" and "Material C - Belongs to → Activity D". The edge attributes include the relationship strength and the relationship scenario. Temporal edges are used to identify the sequential relationships in the marketing process, such as "event pre-heating - pre-event → big promotion" and "user interest - follow-up → order conversion". Edge attributes include time interval and conversion probability. The rules and constraints layer is used to store marketing business rules and graph reasoning constraints, including time range constraints for marketing activities, data validity period constraints, derivation rules for user segmentation, and calculation rules for marketing effects, providing rule support for subsequent asset-related subdomain division, path reasoning, and result verification.

[0022] Furthermore, after completing the basic construction of the digital asset knowledge graph, the knowledge graph construction module divides the multi-asset related subdomains based on the enterprise's current marketing focus. The division of the multi-asset related subdomains adopts an input-driven approach, with the enterprise's current marketing focus (such as "new product launch", "competitor interception", "improving repeat purchases of existing customers", "enhancing the efficiency of major promotional activities") as the core division target; First, locate the entity nodes directly related to each marketing entity in the digital asset knowledge graph and record them as seed nodes. For example, when the marketing entity's direction is "new product launch", the new product entity, the new product marketing activity entity, the new product material entity, and the new product target user group entity are recorded as seed nodes. Starting from the seed node, multi-level diffusion is carried out along various relationship edges in the relationship layer. The diffusion order is set to 3 by default. At the same time, a diffusion stopping condition is set: when the semantic relevance between the entity node to which the diffusion is carried and the marketing subject is lower than a preset threshold, the diffusion in that direction is stopped. During the diffusion process, the semantic relevance between adjacent entity nodes is calculated in real time. The semantic relevance is calculated by the overlap of entity attributes and the number of co-occurrence scenarios. The formula for calculating the semantic association degree between the entity nodes is: Where S represents semantic relevance, β represents connection strength weighting coefficient (default value 0.6), C represents the connection strength value between nodes, and R represents the marketing subject relevance correction coefficient. Set an aggregation threshold. When the semantic correlation S between two entity nodes is greater than or equal to the aggregation threshold, the two nodes are assigned to the same candidate set. After all diffusion nodes have been calculated, the community detection algorithm is used to aggregate all candidate sets, and finally form multiple independent multi-asset related subdomains. For entity nodes that overlap, their affiliation is determined according to the average semantic correlation with different subdomains. The subdomain with higher correlation is the final affiliation. It should be noted that the multi-asset associated subdomains are not fixed. When the enterprise's marketing focus changes, or when new entities and relationship data are added to the knowledge graph, the system will automatically trigger the subdomain re-division process to dynamically update the subdomain scope and node composition, ensuring that the subdomains always match the current marketing business needs.

[0023] Furthermore, once the digital asset knowledge graph and multi-asset related subdomains are constructed, the interaction permission planning module initiates the permission configuration process, sets hierarchical query permissions for accessing users, and establishes a mapping relationship between permissions and the digital asset knowledge graph. First, the accessing users are analyzed through a multi-dimensional role breakdown, including four dimensions: job level, business line affiliation, data security level authorization, and project-specific permissions. Job levels are divided into entry-level staff, middle management, and senior decision-makers. Business lines are divided into departments such as marketing, creative, sales, and finance. Data security level authorization is divided into four levels: public, internal, confidential, and top secret. Project-specific permissions are granted temporarily and can be revoked for specific marketing projects. After this multi-dimensional breakdown, a hierarchical query permission tag set is generated for each accessing user, containing the permission level tag and its effective scope for each dimension. Meanwhile, the interactive permission planning module sets up a hierarchical labeling system in the digital asset knowledge graph, with labels divided into two categories: entity hierarchy and relationship hierarchy. Entity classification is divided into four levels, L1 to L4, based on the sensitivity of the entity and the value of the data. L1 is the public level, corresponding to entity information that can be publicly disclosed, such as published product names and publicly announced event names; L2 is the internal level, corresponding to entity information that can be viewed by all employees within the enterprise, such as advertising creative materials and general marketing insights; L3 is the confidential level, corresponding to entity information that can only be viewed by the corresponding business line, such as marketing campaign budgets and customer segmentation data; L4 is the top secret level, corresponding to entity information that can only be viewed by core authorized personnel, such as core user data and undisclosed strategic plans. Furthermore, after completing the hierarchical labeling of the user permission tag set and the digital asset knowledge graph, the interaction permission planning module performs mapping association and hierarchical matching; First, establish a direct mapping rule between hierarchical query permission tags and hierarchical annotations of the digital asset knowledge graph. The data security level in the user permission tag corresponds one-to-one with the L1-L4 levels of the digital asset knowledge graph. L1 permission can only access entities and relationships annotated at the L1 level, L2 permission can access entities and relationships at the L2 level and below, and so on. Secondly, the intersection calculation of multi-dimensional permissions is performed. The reachable range of four dimensions, namely user job level, business line, data security level, and project permission, is processed by intersection. For example, a junior employee of the marketing department with data security level of L2 and belonging to the new product launch project, has the following reachable range: entities and relationships with label level ≤ L2, belonging to the marketing line, and belonging to the new product launch subdomain. After hierarchical matching and range division, the reachable range corresponding to each user is finally generated. The reachable range is stored in the form of a subgraph, which clearly marks the entity nodes, relationship edges and attribute fields that the user can access. It should be noted that when a user belongs to multiple business lines or projects, the union of the permissions of each line is taken as the basic scope, and then the data security level is used to constrain downwards. If a permission conflict occurs, such as the same entity being labeled with different security levels in different lines, the higher security level will be used for permission verification to avoid unauthorized access to sensitive data.

[0024] Furthermore, after a user submits a query request to the query path planning module, the query parsing, anchor point location, and path planning are carried out sequentially. First, natural language processing is used to understand user queries and identify three key types of information: query intent, core entities, and constraints. For example, if a user queries "Which type of material brings the best conversion effect during the new product launch period", the query intent is identified as "comparative analysis of material effects", the core entities are "new product launch activities", "marketing materials" and "conversion effect", and the constraint is "new product launch period". After completing the requirement analysis, the query path planning module locates the starting query anchor point in the digital asset knowledge graph. It links the core entities in the user's query requirements with the entity nodes in the digital asset knowledge graph and matches the most suitable entity node as the starting query anchor point through semantic relevance calculation. During the location process, permission pre-verification is performed simultaneously. If the matched starting anchor point is outside the user's access range, an insufficient permission prompt is returned, and similar anchor points within the user's access range are recommended for the user to choose from. After locating the starting query anchor point, the query path planning module performs a breadth-first traversal along the relation edges of the digital asset knowledge graph, generating multiple candidate query paths. During the traversal, permission pruning is performed simultaneously to remove entity nodes and relation edges that are beyond the user's access range, ensuring that all candidate paths are within the user's access range. Redundancy pruning is also performed to remove redundant paths that exceed the preset limit or have entity nodes that are repeatedly traversed, significantly reducing the size of candidate paths. For the filtered candidate paths, the query path planning module calculates the query matching degree δ for each path to evaluate the degree of fit between the path and the query requirements. The formula for calculating the query matching degree δ is: ,in , , , These are the weight coefficients for four dimensions: semantic relevance, information gain, query efficiency, and permission adaptability. X represents the semantic relevance between the path and the query requirement, Z represents the information gain value of the path, E represents the query efficiency of the path (calculated by the reciprocal of the path length), and k represents the weight coefficient of the marketing subdomain. Set a path filtering threshold to include candidate paths with a query matching degree value δ greater than or equal to the threshold into the effective path set. Sort the candidate paths from high to low according to the query matching degree value and select the top 5 candidate paths as the final interactive query paths. At the same time, the next interactive query path is used as the alternative path of the previous one according to the sorting. If the previous interactive query path becomes unreachable due to missing data or permission changes, the system will automatically switch to the corresponding alternative path to continue the query, ensuring the stability of the query process. After generating the interactive query path, the query feedback module performs a deep traversal of the digital asset knowledge graph along each interactive query path, and pulls the attribute information and relational data of all entity nodes passed through the interactive query path. After the traversal is complete, the traversal results of each interactive query path are integrated and processed. First, the results are deduplicated by merging entities and relationships that appear repeatedly in different interactive query paths, retaining the attribute data with the highest confidence. Second, multimodal data aggregation is performed by associating and attaching text materials, visual materials, and structured data related to the same entity. For example, the entity "new product poster" can simultaneously attach poster image materials, advertising copy text, and advertising click-through rate data, among other multimodal information. Finally, the results are labeled with confidence levels, marked as high confidence, medium confidence, and unverified based on the data source and verification status, making it easier for users to distinguish the reliability of the data.

[0025] Furthermore, after the results are integrated, the query feedback module will use an appropriate adaptation method to feed back the results to the corresponding user based on the user role and query scenario. Feedback formats include four types: First, a visual knowledge graph format, which displays query results in an interactive graph node diagram, supporting clicking on nodes to view details and related data; second, a structured report format, which presents data indicators and comparison results in the form of tables and statistical charts, suitable for data statistics query needs; third, a multimodal material package format, which categorizes and packages matching images, videos, and text materials for material query needs; and fourth, a natural language summary format, which automatically generates a text summary of query conclusions, extracting core conclusions and key data, suitable for quick browsing scenarios. It should be noted that after the query results are provided, users can submit feedback, such as marking the results as irrelevant, adding query constraints, or adjusting the query direction. After receiving user feedback, the query feedback module synchronizes the feedback information to the query path planning module, fine-tunes the weight parameters of path relevance, and re-plans and generates a query path that better suits the user's needs. At the same time, every fixed data update cycle, the entity and relationship data of the digital asset knowledge graph are updated synchronously, the multi-asset related subdomains are re-planned, and the query matching degree of historical query paths is updated in batches to achieve continuous iterative optimization of query capabilities.

[0026] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A digital asset knowledge graph interactive query system for enterprise marketing, characterized in that: It includes a knowledge graph construction module, an interaction permission planning module, a query path planning module, and a query feedback module; The knowledge graph construction module is used to acquire the original data of the enterprise's multimodal digital assets, extract digital asset entities and their semantic relationships from the original data of the multimodal digital assets, and then construct a digital asset knowledge graph, and divide multiple asset-related subdomains based on the enterprise's marketing subject direction. The interactive permission planning module is used to set hierarchical query permissions for accessing users, establish a mapping relationship between hierarchical query permissions and digital asset knowledge graph, and then set the access range of each accessing user. The query path planning module is used to obtain user query requirements, locate the starting query anchor point in the digital asset knowledge graph based on user query requirements, and then divide multiple interactive query paths. The query feedback module is used to traverse the digital asset knowledge graph along the interactive query path, integrate the traversal results of each interactive query path, and feed them back to the corresponding user.

2. The digital asset knowledge graph interactive query system for enterprise marketing according to claim 1, characterized in that, The interactive query system for digital asset knowledge graphs aimed at enterprise marketing involves the following process for acquiring raw data on an enterprise's multimodal digital assets: The original data of the multimodal digital assets includes text modality, visual modality, and digital and structured modality; Spatiotemporal alignment processing is performed on different modal data under the same marketing event. Taking the marketing event as the core linking anchor, each piece of original multimodal digital asset data is first labeled with event tags and timestamps. At the same time, the scene spatial attributes of the original multimodal digital asset data are matched. For visual modal data, text information, product elements and brand logos in images and videos are extracted in advance through optical character recognition and multimodal content understanding technologies and converted into structured text information for entity extraction.

3. The interactive query system for digital asset knowledge graphs for enterprise marketing according to claim 2, characterized in that, The process of extracting digital asset entities and their semantic relationships based on raw multimodal digital asset data includes: The entity extraction is divided into two categories: regular entity extraction and marketing-specific entity extraction. The conventional entity extraction is combined with industry-specific dictionaries to extract common entity types from text and structured data. The marketing-specific entities are customized for enterprise marketing business scenarios. The relation extraction is divided into two categories: explicit relation extraction and implicit relation extraction. Explicit relations are directly generated from structured and semi-structured data and stored in the form of standardized triples. Implicit relations are obtained through cross-modal semantic reasoning and association mining.

4. The interactive query system for digital asset knowledge graphs for enterprise marketing according to claim 3, characterized in that, The process of constructing the digital asset knowledge graph includes: The digital asset knowledge graph includes an entity layer, a relationship layer, and a rule and constraint layer. The entity layer contains several entity nodes, the number of which is related to the types of digital asset entities extracted. Each entity node is accompanied by complete attribute information. The relationship layer consists of directed edges connecting entity nodes, which are divided into causal edges, relational edges, and temporal edges according to business semantics. The rules and constraints layer is used to store marketing business rules and graph reasoning constraints, including time range constraints for marketing activities, data validity period constraints, derivation rules for user segmentation, and calculation rules for marketing effects, providing rule support for subsequent asset-related subdomain division, path reasoning, and result verification.

5. The interactive query system for digital asset knowledge graphs for enterprise marketing according to claim 4, characterized in that, The process of dividing multiple asset-related subdomains based on the enterprise's marketing focus includes: The division of the multi-asset related subdomains adopts an input-driven approach, with the enterprise's current marketing focus as the core division target; First, locate the entity nodes directly related to each marketing entity in the digital asset knowledge graph and record them as seed nodes. Starting from the seed nodes, multi-level diffusion is carried out along various relationship edges of the relationship layer. At the same time, a diffusion stopping condition is set: when the semantic relevance between the entity node diffused to and the marketing entity is lower than a preset threshold, the diffusion in that direction is stopped. During the diffusion process, the semantic relevance between adjacent entity nodes is calculated in real time. An aggregation threshold is set. When the semantic correlation between two entity nodes is greater than or equal to the aggregation threshold, the two diffusion nodes are assigned to the same candidate set. Otherwise, they are not assigned to the same candidate set. After all diffusion nodes have been calculated, all candidate sets are aggregated to form multiple independent multi-asset related subdomains. For entity nodes that overlap, their affiliation is determined based on their average semantic correlation with different subdomains. The subdomain with higher correlation is assigned to them.

6. The digital asset knowledge graph interactive query system for enterprise marketing according to claim 5, characterized in that, The process of setting tiered query permissions for users, establishing a mapping relationship between these permissions and the digital asset knowledge graph, and defining the access range for each user includes: Access users are analyzed through a multi-dimensional role breakdown, encompassing four dimensions: job level, business line affiliation, data security level authorization, and project-specific permissions. Job levels are categorized hierarchically from entry-level staff to middle management and senior decision-makers. Business lines are divided into departments such as marketing, creative, sales, and finance. Data security level authorization is categorized into four levels: public, internal, confidential, and top secret. Project-specific permissions are granted temporarily and can be revoked for specific marketing projects. After this multi-dimensional breakdown, a tiered query permission tag set is generated for each access user, containing the corresponding permission level tag and scope of effectiveness for each dimension. Meanwhile, the interactive permission planning module sets up a hierarchical labeling system in the digital asset knowledge graph. The labels are divided into two categories: entity hierarchical and relationship hierarchical. A direct mapping rule is established between hierarchical query permission tags and hierarchical labels in the digital asset knowledge graph. Next, the reachability range of the four dimensions of user job level, business line, data security level, and project permission is intersected. After level matching and range division, the access reachability range corresponding to each accessing user is finally generated.

7. The digital asset knowledge graph interactive query system for enterprise marketing according to claim 6, characterized in that, The process of locating the initial query anchor point in the digital asset knowledge graph based on user query requirements includes: Natural language understanding is used to process user queries and identify three key types of information: query intent, core entities, and constraints. The system locates the starting query anchor point in the digital asset knowledge graph, links the core entities in the user's query requirements with entity nodes in the digital asset knowledge graph, and uses semantic relevance calculation to match entity nodes as the starting query anchor point. During the location process, permission pre-verification is performed simultaneously. If the matched starting anchor point is outside the user's access range, an insufficient permission prompt is returned, and similar anchor points within the user's access range are recommended for the user to choose from.

8. The digital asset knowledge graph interactive query system for enterprise marketing according to claim 7, characterized in that, The process of dividing into multiple interactive query paths includes: After locating the starting query anchor point, the query path planning module takes the anchor point as the starting point and performs a breadth-first traversal along the relation edges of the digital asset knowledge graph to generate multiple candidate query paths. During the traversal, permission pruning is performed simultaneously to remove entity nodes and relation edges that are beyond the user's access range, ensuring that all candidate paths are within the user's access range. Redundancy pruning is also performed to remove redundant paths. For the filtered candidate paths, the query path planning module calculates the query matching degree value of each path, sets a path filtering threshold, and includes candidate paths with query matching degree values ​​greater than or equal to the filtering threshold into the effective path set. The paths are then sorted from high to low according to their query matching degree values, and the top 5 candidate paths are selected as the final interactive query paths. At the same time, the next interactive query path is used as the alternative path to the previous one according to the sorting. If the previous interactive query path becomes unreachable due to missing data or permission changes, the system automatically switches to the corresponding alternative path to continue the query. A deep traversal is performed along each interactive query path in the digital asset knowledge graph. The attribute information and relational data of all entity nodes traversed by the interactive query path are retrieved. After the traversal is completed, the traversal results of each interactive query path are integrated and deduplicated. Entities and relations that appear repeatedly in different interactive query paths are merged.

9. The digital asset knowledge graph interactive query system for enterprise marketing according to claim 8, characterized in that, The process of traversing the digital asset knowledge graph along interactive query paths and integrating the traversal results of each interactive query path includes: After the results are integrated, the query feedback module will use appropriate adaptation methods to provide feedback to the corresponding users based on their roles and query scenarios. These feedback methods include visual knowledge graphs, structured reports, multimodal resource packages, and natural language summaries.

Images