Data asset intelligent recommendation method based on security classification constraint and multi-factor fusion
By employing an intelligent recommendation method that combines security level constraints with multi-factor fusion, the system addresses the security risks and limited strategy issues of data asset recommendation systems under strict access control, achieving a balance between security and efficiency, and improving the accuracy and adaptability of recommendations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-04-14
AI Technical Summary
Existing data asset recommendation systems suffer from risks of unauthorized access under strict access control environments, have limited recommendation strategies that are detached from business context, and have rigid algorithm architectures that are difficult to adapt to permission constraints, resulting in security vulnerabilities and poor recommendation performance.
An intelligent recommendation method based on security level constraints and multi-factor fusion is adopted. Through real-time permission authentication, multi-path recall strategy and fusion ranking, the recommendation results are ensured to be within the scope of user permissions, and personalized recommendations are made in combination with user identity, organizational attributes and behavioral characteristics.
It achieves a balance between security and efficiency under strict access control, eliminates the risk of unauthorized access, improves the accuracy and practicality of recommendations, adapts to complex security environments, and enhances the discovery efficiency and utilization of data assets.
Smart Images
Figure CN121858634A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of data management and information technology, specifically relating to an intelligent recommendation method for data assets based on security level constraints and multi-factor fusion. Background Technology
[0002] As enterprises deepen their digital transformation, data platforms, as unified data capability platforms, have become core enterprise infrastructure. Within a data platform, governed and integrated data is packaged into reusable data assets and provided externally through an asset portal. However, with the explosive growth in the number of assets, how users can quickly and accurately discover and obtain the high-value data they need from massive amounts of assets has become a significant challenge.
[0003] To address the aforementioned issues, intelligent recommendation technology has been introduced into the field of data asset management. Existing commonly used recommendation schemes primarily draw upon mature algorithms from e-commerce and content information sectors, and their implementation ideas mainly include: Collaborative filtering: By analyzing a large amount of users' historical behavior data (such as browsing, applying, and downloading), it identifies user groups with similar interests to the current user and recommends assets with high preferences from these groups to the current user.
[0004] Content-based recommendation: Analyze the metadata (such as asset type and topic tags) of assets that users have applied for or frequently searched, and then recommend other assets with similar characteristics.
[0005] Popular Ranking Recommendation: Simply sort and recommend assets based on the overall number of applications or recent applications, based on the general principle of "birds of a feather flock together".
[0006] These technical solutions have enabled personalized recommendations to some extent and improved the usability of data assets.
[0007] Traditional recommendation systems have the following problems: 1. The security mechanism is disconnected from the recommendation logic, posing an inherent risk of unauthorized access. Current recommendation algorithms generally fail to consider the security level of data assets and user access permissions as core decision-making factors in their design and operation. Their recommendation engines lack mandatory verification of the fundamental security rule that "user security level ≥ asset security level" when generating candidate lists. This can lead to the system recommending highly sensitive assets that low-security users do not have access to. This not only renders the recommendations invalid but also results in the excessive exposure of sensitive data, posing a serious risk of data leakage. For example, collaborative filtering algorithms might recommend core assets to all users due to frequent access by high-security user groups, which is unacceptable in government, finance, and other similar scenarios.
[0008] 2. The recommendation strategy is too narrow in scope and detached from the business context. The lack of business context leads to recommendation results that are "statistically relevant" but "business-useless." The core problem with existing recommendation technologies in a data platform environment is that their recommendation logic relies excessively on user behavior statistical features (such as click-through rates and collaborative filtering similarity), failing to deeply integrate the rich business semantics and organizational structure information inherent within the data platform. Specifically, the algorithm cannot effectively utilize key business contexts such as the asset's department, business domain, data lineage, and the user's role, position, and department. This results in recommendation results that only reflect superficial statistical relevance (e.g., "other similar users also viewed asset A"), failing to understand the true value of data assets in specific business scenarios.
[0009] Furthermore, traditional collaborative filtering might recommend popular assets across departments, but these assets may have no business value for users in another department. Popularity-based recommendations might allow some general-purpose assets to dominate the rankings for a long time, while some high-value but niche specialized assets will never be discovered by potential users. Current technology lacks a comprehensive recommendation model that can integrate multiple dimensions of factors, including user identity (department, role, seat), group behavior trends (preferences of users in the same department / security level), personal historical preferences (search, view, apply, download), and global popularity.
[0010] 3. The algorithm architecture is rigid and difficult to adapt to environments with permission constraints. The rigidity of algorithm architecture makes it difficult to adapt to the specific constraints of high-security environments. The underlying root cause of the aforementioned accuracy problems lies in the inherent rigidity of traditional recommendation algorithm architecture design. These algorithms are usually based on the ideal assumption of a "global and complete data perspective." However, the strict security level control mechanism of the data platform naturally divides the entire data into multiple isolated "permission subsets." Under this constraint, directly applying existing algorithms faces two major challenges: first, the data perspective is limited, and the algorithm cannot securely and effectively utilize the collective wisdom across security levels (for example, the valuable behavioral patterns of high-security users within the same business line cannot be securely used to guide recommendations for low-security users); second, interest modeling is distorted, and the algorithm struggles to accurately distinguish whether a user is "not interested" in a certain asset or simply "does not have permission to access" it. This mismatch between architecture and environment leads to a significant decrease in the performance of the recommendation system within the security boundary, making it unable to meet the fundamental requirements of the specific and complex scenario of the data platform. Summary of the Invention (a) Technical problems to be solved The technical problem this invention aims to solve is how to provide an intelligent recommendation method for data assets based on security level constraints and multi-factor fusion, so as to address the fundamental defects of existing recommendation technologies in environments with strict security level access control.
[0011] (II) Technical Solution To address the aforementioned technical problems, this invention proposes an intelligent recommendation method for data assets based on security level constraints and multi-factor fusion. This method includes the following steps: Step A: The user logs into the asset portal; Step B: The asset portal initiates a recommendation request; Step C: The intelligent recommendation engine receives the request; Step D: Request parsing and context extraction After receiving a recommendation request from the asset portal, the intelligent recommendation engine first parses the request message and extracts key fields. Step E: Real-time authorization The intelligent recommendation engine initiates a real-time query to the system management subsystem, requesting the security level attribute bound to the user ID and other possible organizational attributes; Step F: Obtain metadata While obtaining the user's security level, the metadata of all data assets is asynchronously pulled from the asset pool; Step G: Rigid density filtration The system executes the core security level adaptive filtering algorithm: using the security rule "user security level ≥ asset security level" as a Boolean condition to filter all data assets; Step H: Generation of a safe candidate set A dynamic and absolutely secure initial set of candidate assets is generated through a rigid security level filtering process as a security candidate set. Step I: After obtaining the safe candidate set, the system adopts a multi-path recall strategy to mine assets that users may be interested in from different dimensions in a parallel manner. Specifically, this includes: semantic recall based on vector retrieval, precise recall based on keywords, relational recall based on knowledge graphs, and recall based on behavioral sequences. Step J: Aggregate Recall Results The results of the above four recalls are deduplicated and aggregated to form a larger candidate pool; Step K: Merge Sort A fusion sorting algorithm is used to uniformly score and sort the assets in the pool; Step L: Result Generation After merging and sorting, the N assets with the highest scores are selected to form the final recommendation list, which is returned to the asset portal via API. Step M: Return to the asset portal interface rendering After receiving the list, the asset portal queries detailed display information based on the asset ID and renders and displays it in the user interface; Step N: User views the results Subsequent user interactions will be recorded and fed back to the system to update the user interest model and continuously optimize the recommendation effect.
[0012] (III) Beneficial Effects This invention proposes an intelligent recommendation method for data assets based on density level constraints and multi-factor fusion, with the main advantages being reflected in the following three aspects: 1. Built-in security level constraints enable deep collaboration between recommendations and security.
[0013] The primary advantage of this invention is that it no longer simply "grafts" a separate recommendation module onto the security system of the data platform, but rather deeply embeds security rules into the underlying layer of the recommendation logic.
[0014] Core Innovation: The system moves the traditional post-event verification principle of "user security level ≥ asset security level" from the data platform to a rigid prerequisite for the recommendation algorithm's operation. Before generating any candidate recommendation list, the system first performs mandatory filtering of all assets based on the user's identity attributes (such as security level), ensuring that the recommendation engine only operates within the "secure asset pool" that the user has access to.
[0015] The value of this solution: It fundamentally eliminates the risk of recommending highly sensitive assets to low-security users, not only avoiding ineffective recommendations but also completely eliminating the potential for excessive exposure and data leakage of sensitive data caused by the recommendation system. This transforms the recommendation system from a potential security risk into a reliable and proactive component of the enterprise's data security management system.
[0016] 2. Multi-dimensional business feature integration enables a leap from "statistically relevant" to "business-useful".
[0017] This invention breaks through the limitations of traditional recommendation algorithms that rely solely on statistical features of user behavior. By introducing and integrating rich business semantics, it greatly improves the accuracy and practicality of recommendations.
[0018] Core Innovation: The recommendation algorithm can deeply leverage the multi-dimensional features inherent in the data platform. This includes the business attributes of data assets (such as department, business domain, and data lineage), the organizational attributes of users (user roles, positions, and departments), and dynamic behavioral data (such as user searches, views, applications, and downloads of assets). This integration enables the system to understand the true value of data in specific business scenarios, rather than just superficial statistical correlations.
[0019] Value of the solution: It effectively addresses the pain point of recommendation results being "statistically relevant" but "useless in business." For example, the system can identify a data asset that is not only frequently used by users in the same department but also has a close relationship with the data in the user's current project, thereby providing highly accurate recommendations with business action value, significantly improving the discovery efficiency and utilization rate of data assets.
[0020] 3. A flexible and scalable recommended architecture that adapts to the complex constraints of high-security environments.
[0021] This invention designs a flexible and adaptable system architecture that can proactively adapt to and effectively address the unique technical challenges arising from classified information control.
[0022] Core Innovation: It adopts a layered, collaborative design philosophy. This architecture can adaptively select or combine various recommendation strategies (such as popularity-based recommendations, collaborative filtering-based recommendations, and knowledge-based recommendations) according to different data environments and user scenarios. It acknowledges and effectively addresses the problems of "limited data visibility" and "distorted interest modeling" caused by access control isolation.
[0023] The value of the solution lies in enabling recommendation systems to maintain high efficiency and intelligence while strictly adhering to security rules. For example, for newly listed high-security assets, the system can utilize their metadata (content-based recommendations) for a cold start; for low-security users, it can safely reference the group trends of other low-security users within the same department, thereby maximizing recommendation effectiveness within security boundaries. Attached Figure Description
[0024] Figure 1 This is a flowchart of the intelligent recommendation method for data assets based on security level constraints and multi-factor fusion, as described in this invention. Figure 2 A knowledge graph for asset entities, user entities, and behavioral data. Detailed Implementation
[0025] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
[0026] This invention belongs to the interdisciplinary field of data management and information technology, specifically relating to intelligent recommendation technology for data assets within a data platform system. In particular, this invention relates to a method and system capable of adaptively and intelligently filtering, sorting, and recommending data assets based on user identity, permissions, and multi-dimensional behavioral characteristics in an environment with strict security level (security grade) access control. Its core lies in addressing the disconnect between traditional recommendation technologies and access control in high-security scenarios. By using the security attributes (security level) of data assets as a rigid prerequisite constraint for the recommendation algorithm, and integrating multiple factors such as user attributes, group behavior, personal preferences, and popularity, it achieves personalized recommendations that balance security and efficiency. Therefore, this invention is a comprehensive application that deeply integrates data asset management, intelligent recommendation algorithms, and information security technologies.
[0027] This invention proposes an intelligent recommendation method and system for data assets based on security level constraints and multi-factor fusion, aiming to overcome the fundamental defects of existing recommendation technologies under strict security level access control environments. Specifically, this invention focuses on solving the following three key technical problems: 1. The security mechanism is disconnected from the recommendation logic, posing an inherent risk of unauthorized access; 2. The recommendation strategy is too narrow in scope and detached from the business context; 3. The algorithm architecture is rigid and difficult to adapt to environments with permission constraints.
[0028] This invention describes an intelligent recommendation method and system for data assets based on security level constraints and multi-factor fusion. Its main advantages are as follows: built-in security level constraints enable deep collaboration between recommendation and security; multi-dimensional business feature fusion achieves a leap from "statistical relevance" to "business usefulness"; and a flexible and scalable recommendation architecture adapts to the complex constraints of high-security environments.
[0029] A data asset intelligent recommendation method and system based on security level constraints and multi-factor fusion, flowchart as follows: Figure 1 As shown, it mainly includes three key time-series stages: security verification and candidate set generation (step DH), multi-path parallel recall (step I), and fusion sorting and result return (step JN). The specific execution steps and technical details of each stage are as follows: Step A: The user logs into the asset portal; Step B: The asset portal initiates a recommendation request; Step C: The data fusion subsystem (i.e., the intelligent recommendation engine) receives the request; 1. Security verification and candidate set generation stage This stage is the security cornerstone of the entire method, ensuring that all subsequent operations are performed under strict security level restrictions, thus fundamentally solving the problem of "disconnect between security mechanisms and recommendation logic". The process is as follows: Step D: Request parsing and context extraction After receiving a recommendation request from the asset portal, the data fusion subsystem (i.e., the intelligent recommendation engine) first parses the request message and extracts key fields. The most crucial of these is the user's unique identifier (User ID). The system then uses this identifier to retrieve the user's basic attribute information from the context of the current session, preparing for subsequent permission verification.
[0030] Step E: Real-time authorization The intelligent recommendation engine initiates a real-time query to the system management subsystem (which serves as a unified permission center), requesting the security level attribute (such as "important") and other possible organizational attributes (such as department and role) bound to the user ID. This step is completed through an internally secure API interface to ensure the authenticity of the user's identity and the accuracy of the permission information.
[0031] Step F: Obtain metadata While obtaining the user's security level, the system asynchronously pulls the metadata of all data assets from the asset pool, which must include the security level attribute of each asset.
[0032] Step G: Rigid density filtration The system executes the core security level adaptive filtering algorithm: using the security rule "user security level ≥ asset security level" as a Boolean condition to filter all data assets.
[0033] Step H: Generation of a safe candidate set A dynamic and absolutely secure initial set of candidate assets is generated through a rigid security-level filtering process, serving as a secure candidate set. This set ensures that no subsequent calculations or recommendations will include assets that the user is not authorized to access, fundamentally eliminating the risk of unauthorized access.
[0034] 2. Multi-path parallel recall phase Step I: After obtaining the safe candidate set, the system adopts a multi-path recall strategy to mine assets that users may be interested in from different dimensions in a parallel manner, balancing recall efficiency and result diversity. It uses rich business semantics and intelligent algorithms to compensate for the shortcomings of traditional methods and overcome the problems of "single-dimensional recommendation strategies" and "rigid algorithm architecture".
[0035] Strategy I1: Semantic Recall Based on Vector Retrieval Pre-trained semantic models (such as Sentence-BERT) are used to transform textual metadata (e.g., names, descriptions, business tags) of assets in the safe candidate set into high-dimensional vectors (Embeddings). Simultaneously, users' implicit or explicit interest preferences (e.g., historical search keywords, long-term interest tags) are also transformed into vectors. Using an approximate nearest neighbor search algorithm (such as Faiss), similarity calculations (commonly cosine similarity) are performed in the vector space to recall the Top-K assets most semantically similar to the user's interests.
[0036] Strategy I2: Keyword-based Precision Recall An inverted index is built for assets, supporting efficient keyword matching of asset metadata. When a user has a specific search term or the system identifies user interest tags, algorithms such as BM25 are used for retrieval. This strategy focuses on exact matching, and is particularly suitable for scenarios involving technical terms, specific codes, or asset numbers.
[0037] Strategy I3: Relationship Retrieval Based on Knowledge Graph (Business Context Retrieval) Based on knowledge graph relationship retrieval, entity associations are retrieved and reasoned to obtain a list of recalled assets; Strategy I4: Recall Based on Behavioral Sequences The analysis is based on behavioral sequences, and interest modeling is performed using CNN to obtain a list of assets to be recalled. 3. Merging, sorting, and result return stage This phase aims to dynamically weight and optimize the results of multi-path recall. This is a flexible architecture design and adaptive learning mechanism that further enhances the accuracy and adaptability of recommendations, consolidating the solutions to technical problems 2 and 3.
[0038] Step J: Aggregate Recall Results The results of the four recalls mentioned above are deduplicated and aggregated to form a larger candidate pool.
[0039] Step K: Merge Sort Subsequently, a fusion ranking algorithm is used to uniformly score and rank the assets in the pool. Preferably, the Reverse Rank Fusion (RRF) algorithm is used: this algorithm assigns a score to the ranking of each asset in different recall lists, and finally sums the scores of each list with weights, thereby comprehensively considering the performance of assets in various strategies and avoiding the dominance of a single strategy in the results. In addition, the ranking model also introduces business rules for weighting, such as increasing the weight of newly listed assets in the user's department, or dynamically adjusting the weights based on the freshness of the assets (such as the listing time).
[0040] Step L: Result Generation After merging and sorting, the N assets with the highest scores are selected to form the final recommendation list, which is then returned to the asset portal via API.
[0041] Step M: Return to the asset portal interface rendering After receiving the list, the asset portal queries detailed display information based on the asset ID (such as asset name, asset type, tags, resources contained under the asset, description, etc.) and renders and displays it in the user interface.
[0042] Step N: User views the results Subsequent user interactions (such as searching, viewing, applying for, and downloading assets) will be recorded and fed back to the system to update the user interest model and continuously optimize the recommendation effect.
[0043] In summary, through the aforementioned strict timing control and multi-level algorithm collaboration, this invention achieves accurate, diverse, and business-valuable intelligent asset recommendation services for users while ensuring data security.
[0044] This invention describes an intelligent recommendation method and system for data assets based on security level constraints and multi-factor fusion. Its main advantages are reflected in the following three aspects: 1. Built-in security level constraints enable deep collaboration between recommendations and security.
[0045] The primary advantage of this invention is that it no longer simply "grafts" a separate recommendation module onto the security system of the data platform, but rather deeply embeds security rules into the underlying layer of the recommendation logic.
[0046] Core Innovation: The system moves the traditional post-event verification principle of "user security level ≥ asset security level" from the data platform to a rigid prerequisite for the recommendation algorithm's operation. Before generating any candidate recommendation list, the system first performs mandatory filtering of all assets based on the user's identity attributes (such as security level), ensuring that the recommendation engine only operates within the "secure asset pool" that the user has access to.
[0047] The value of this solution: It fundamentally eliminates the risk of recommending highly sensitive assets to low-security users, not only avoiding ineffective recommendations but also completely eliminating the potential for excessive exposure and data leakage of sensitive data caused by the recommendation system. This transforms the recommendation system from a potential security risk into a reliable and proactive component of the enterprise's data security management system.
[0048] 2. Multi-dimensional business feature integration enables a leap from "statistically relevant" to "business-useful".
[0049] This invention breaks through the limitations of traditional recommendation algorithms that rely solely on statistical features of user behavior. By introducing and integrating rich business semantics, it greatly improves the accuracy and practicality of recommendations.
[0050] Core Innovation: The recommendation algorithm can deeply leverage the multi-dimensional features inherent in the data platform. This includes the business attributes of data assets (such as department, business domain, and data lineage), the organizational attributes of users (user roles, positions, and departments), and dynamic behavioral data (such as user searches, views, applications, and downloads of assets). This integration enables the system to understand the true value of data in specific business scenarios, rather than just superficial statistical correlations.
[0051] Value of the solution: It effectively addresses the pain point of recommendation results being "statistically relevant" but "useless in business." For example, the system can identify a data asset that is not only frequently used by users in the same department but also has a close relationship with the data in the user's current project, thereby providing highly accurate recommendations with business action value, significantly improving the discovery efficiency and utilization rate of data assets.
[0052] 3. A flexible and scalable recommended architecture that adapts to the complex constraints of high-security environments.
[0053] This invention designs a flexible and adaptable system architecture that can proactively adapt to and effectively address the unique technical challenges arising from classified information control.
[0054] Core Innovation: It adopts a layered, collaborative design philosophy. This architecture can adaptively select or combine various recommendation strategies (such as popularity-based recommendations, collaborative filtering-based recommendations, and knowledge-based recommendations) according to different data environments and user scenarios. It acknowledges and effectively addresses the problems of "limited data visibility" and "distorted interest modeling" caused by access control isolation.
[0055] The value of the solution lies in enabling recommendation systems to maintain high efficiency and intelligence while strictly adhering to security rules. For example, for newly listed high-security assets, the system can utilize their metadata (content-based recommendations) for a cold start; for low-security users, it can safely reference the group trends of other low-security users within the same department, thereby maximizing recommendation effectiveness within security boundaries.
[0056] Comparison of traditional recommendation systems and density-constrained multi-factor fusion recommendation systems The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A data asset intelligent recommendation method based on security level constraints and multi-factor fusion, characterized in that, The method includes the following steps: Step A: The user logs into the asset portal; Step B: The asset portal initiates a recommendation request; Step C: The intelligent recommendation engine receives the request; Step D: Request parsing and context extraction After receiving a recommendation request from the asset portal, the intelligent recommendation engine first parses the request message and extracts key fields. Step E: Real-time authorization The intelligent recommendation engine initiates a real-time query to the system management subsystem, requesting the security level attribute bound to the user ID and other possible organizational attributes; Step F: Obtain metadata While obtaining the user's security level, the metadata of all data assets is asynchronously pulled from the asset pool; Step G: Rigid density filtration The system executes the core security level adaptive filtering algorithm: using the security rule "user security level ≥ asset security level" as a Boolean condition to filter all data assets; Step H: Generation of a safe candidate set A dynamic and absolutely secure initial set of candidate assets is generated through a rigid security level filtering process as a security candidate set. Step I: After obtaining the safe candidate set, the system adopts a multi-path recall strategy to mine assets that users may be interested in from different dimensions in a parallel manner. Specifically, this includes: semantic recall based on vector retrieval, precise recall based on keywords, relational recall based on knowledge graphs, and recall based on behavioral sequences. Step J: Aggregate Recall Results The results of the above four recalls are deduplicated and aggregated to form a larger candidate pool; Step K: Merge Sort A fusion sorting algorithm is used to uniformly score and sort the assets in the pool; Step L: Result Generation After merging and sorting, the N assets with the highest scores are selected to form the final recommendation list, which is returned to the asset portal via API. Step M: Return to the asset portal interface rendering After receiving the list, the asset portal queries detailed display information based on the asset ID and renders and displays it in the user interface; Step N: User views the results Subsequent user interactions will be recorded and fed back to the system to update the user interest model and continuously optimize the recommendation effect.
2. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, In step D, the user's unique identifier, User ID, is extracted, and then the user's basic attribute information is obtained from the context of the current session based on this identifier, in preparation for subsequent permission verification.
3. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, In step E, the organizational attributes include: department and role.
4. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, Step E is completed through an internally secure API interface to ensure the authenticity of the user's identity and the accuracy of the permission information.
5. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, In step F, the metadata includes the security level attribute of each asset.
6. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, The semantic recall based on vector retrieval includes: using a pre-trained semantic model to transform the textual metadata of assets in the safe candidate set into high-dimensional vectors; simultaneously, transforming the user's implicit or explicit interest preferences into vectors; and using an approximate nearest neighbor search algorithm to perform similarity calculation in the vector space to recall the Top-K assets that are most similar to the user's interest semantics.
7. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, The keyword-based precise recall includes: constructing an inverted index for assets to support efficient keyword matching of asset metadata; and using the BM25 algorithm for retrieval when the user has a specific search term or the system identifies user interest tags.
8. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, The knowledge graph-based relationship recall includes: retrieving entity associations based on knowledge graph relationships and reasoning to obtain a list of assets to be recalled.
9. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, The behavior sequence-based recall includes: analyzing behavior sequences and using CNN for interest modeling to obtain a list of assets to be recalled.
10. The intelligent recommendation method for data assets based on security level constraints and multi-factor fusion as described in claim 1, characterized in that, Step K includes: using the Reverse Ranking Fusion RRF algorithm: this algorithm assigns a score to the ranking of each asset in different recall lists, and finally sums the scores of each list in a weighted manner, so as to comprehensively consider the performance of the asset in various strategies and avoid the single strategy dominating the result; in addition, the ranking model will also introduce business rules for weighting, increase the weight of newly listed assets in the user's department, or dynamically adjust the weight according to the freshness of the asset.