Ai-based system and method for data sharing and access control

TWI946232BActive Publication Date: 2026-11-11陈立新
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
TW114149982
Authority / Receiving Office
TW · TW
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-12-18
Publication Date
2026-11-11
Estimated Expiration
2045-12-17

Smart Images

  • Figure TWG2TB001920984_001
    Figure TWG2TB001920984_001
  • Figure TWG2TB001920984_002
    Figure TWG2TB001920984_002
  • Figure TWG2TB001920984_003
    Figure TWG2TB001920984_003
Patent Text Reader

Abstract

This invention provides an artificial intelligence-based data sharing and access control system and method. Its core technologies include: 1. A multimodal intelligent classification engine capable of simultaneously processing text, images, and files, combined with a personalized learning mechanism; 2. A permission management processor supporting natural language commands, allowing for intuitive setting of sharing permissions; 3. A distributed encrypted storage and hierarchical key management mechanism, providing different levels of data decryption capabilities based on user permission levels, achieving fine-grained security control. This system aims to address the shortcomings of traditional data management in classification, authorization, and security, improving the efficiency and security of collaborative environments.
Need to check novelty before this filing date? Find Prior Art

Claims

1. An AI-based data sharing and access control system, comprising: a client application providing a conversational user interface; an API gateway that receives upload requests when a user uploads data items through the conversational user interface, and routes and transmits the data items and user identification information to a backend; an AI classification engine module that receives the data items transmitted by the API gateway, performs semantic feature extraction on the text content of the data items using a text analysis module to generate semantic vectors, performs visual feature recognition on the image content of the data items using an image recognition module to generate visual vectors, and performs structural analysis on the format and metadata of the data items using a file structure analysis module to generate structural vectors; the AI ​​classification engine module performs weighted fusion operations on the semantic vectors, visual vectors, and structural vectors to generate a unified feature vector, and calculates the user's... The classification preference coefficient vector of similar past data items is used to generate a weighted feature vector after matrix operation between the unified feature vector and the classification preference coefficient vector. The AI ​​classification engine module calculates the probability distribution of each category based on the weighted feature vector and the pre-trained classification model. The classification decision and suggestion generation module makes a decision: if the difference between the highest probability and the second highest probability exceeds a preset threshold, a single category suggestion is generated; otherwise, multiple category suggestions with confidence scores are generated. The AI ​​classification engine module sends the category suggestion back to the conversational user interface through the API gateway for the user to confirm. A command parsing processor receives a confirmation signal when the user confirms the category suggestion through the conversational user interface, associates the data item with the confirmed category label, and sets a minimum permission requirement for the data item based on the confirmed category label. The command parsing processor further receives the natural language permission management command input by the user, extracts the user entity, group entity, and data category entity from the natural language permission management command using an entity recognition module, determines whether the natural language permission management command is one of granting, revoking, or modifying operations using an intent classification module, and combines the user entity, group entity, data category entity, and operation type to generate a structured permission rule. The structured permission rule includes a numerical representation of the permission subject, permission object, and permission level. An access control verification processor receives the command, parses the structured permission rules generated by the processor, and stores them in a permission index table. The permission index table establishes a hash mapping relationship between user identification codes and permission levels. When the API gateway receives a data query request from a user, it sends the data query request to the access control verification processor. The access control verification processor locates the corresponding permission level in the permission index table based on the user identification code, and the permission verification module compares the permission level with the minimum permission requirement of the queried data category.When the permission level is greater than or equal to the minimum permission requirement, an access permission signal is generated, and the corresponding data item is retrieved from the encrypted storage array based on the access permission signal. The data item is then returned to the requesting user through the API gateway and the dialogic user interface. If the permission level is lower than the minimum permission requirement, an access denial signal is generated, and the unauthorized access attempt is recorded. An encrypted storage array includes a plurality of storage nodes. After the command parser processor completes the association between the data item and the category tag, it receives the data item, encrypts the data item using an encryption algorithm, and then fragments and stores the encrypted data item in the plurality of storage nodes after redundant encoding. The access control verification processor determines the number of storage nodes and the range of data fragments that can be accessed based on the permission level of the requesting user. The hierarchical encryption key management module provides the requesting user with the access permission signal with a corresponding number of encryption key fragments, so that the requesting user can use the encryption key fragments to reconstruct the data encryption key and decrypt it to obtain the data item content that matches their permission level.

2. The system described in claim 1, wherein the text analysis module performs semantic feature extraction using a deep learning language model, the image recognition module performs visual feature recognition using a neural network model, and the document structure analysis module performs structure analysis using a document layout analysis algorithm.

3. The system as described in claim 1, wherein the time decay weight function is calculated by decreasing the weight according to a time interval, and the artificial intelligence classification engine module uses the time decay weight function to calculate the user's classification preference coefficient vector for similar type data items within a preset time interval.

4. The system as described in claim 1, wherein the weighted fusion operation is performed using an attention-weight-based mechanism that fuses the semantic vector, the visual vector, and the structural vector to produce the unified feature vector.

5. The system as described in claim 1, wherein when the difference between the highest probability and the second highest probability exceeds the preset threshold, the classification decision and suggestion generation module generates the single category suggestion; otherwise, it generates a plurality of category suggestions ranked with confidence scores.

6. The system as described in claim 1, wherein the entity recognition module uses a machine learning model to extract the user entity, the group entity, and the data category entity, and the intent classification module uses a classification model to determine the operation type.

7. The system as described in claim 1, wherein the encryption algorithm is a symmetric encryption algorithm, the redundancy coding employs an error-correcting code, and the access control verification processor determines the number of encryption key fragments provided to the requesting user based on the permission level, wherein the number of encryption key fragments is positively correlated with the permission level.

8. The system as described in Request 1 further includes an anomaly detection module that monitors the query frequency and query content pattern of the requesting user. When the query frequency exceeds a preset frequency or the query content deviates from the historical pattern of the requesting user by more than a set threshold, the anomaly detection module generates a security alarm signal and suspends the access rights of the requesting user until identity re-verification is completed.

9. The system as described in Request 1, wherein the API gateway further includes a load balancer that, when multiple upload requests are received simultaneously, distributes the requests to different processing queues according to the current system load, each processing queue having a preset capacity, and when the capacity is exceeded, returns a system busy signal to the user and suggests retrying later.

10. A data sharing and access control method based on artificial intelligence, comprising the following steps: When a user uploads a data item through a conversational user interface, an API gateway receives the upload request and routes the data item and user identification information to an artificial intelligence classification engine module; The artificial intelligence classification engine module receives the data item, uses a text analysis module to perform semantic feature extraction on the text content of the data item to generate a semantic vector, uses an image recognition module to perform visual feature recognition on the image content of the data item to generate a visual vector, and uses a file structure analysis module to perform structural analysis on the format and metadata of the data item to generate a structural vector; The semantic vector, the visual vector, and the structural vector are weighted and fused to generate a unified feature vector, and a time decay weight function is used to calculate the user's classification preference coefficient vector for similar data items in the past. A weighted feature vector is generated after matrix operation between the unified feature vector and the classification preference coefficient vector. The classification decision and suggestion generation module calculates the probability distribution of each category based on the weighted feature vector and the pre-trained classification model. If the difference between the highest probability and the second highest probability exceeds a preset threshold, a single category suggestion is generated; otherwise, multiple category suggestions with confidence scores are generated and sent back to the conversational user interface for user confirmation via the API gateway. When the user confirms the category suggestion through the conversational user interface, the command parsing processor receives the confirmation signal, associates the data item with the confirmed category label, and sets a minimum permission requirement for the data item based on the confirmed category label. The command parsing processor receives the natural language permission management command input by the user, extracts the user entity, group entity, and data category entity from the natural language permission management command using the entity recognition module, determines that the natural language permission management command is one of granting, revoking, or modifying operations using the intent classification module, and combines the user entity, group entity, data category entity, and operation type to generate structured permission rules. The structured permission rules include numerical representations of the permission subject, permission object, and permission level. The access control verification processor receives the structured permission rules and stores them in a permission index table, which establishes a hash mapping relationship between user identification codes and permission levels. The data item is then transmitted to an encrypted storage array, where it is encrypted using an encryption algorithm. The encrypted data item is then redundantly encoded and fragmented and stored across multiple storage nodes. When the API gateway receives a data query request from a user, it transmits the request to the access control verification processor. The processor locates the corresponding permission level in the permission index table based on the user identification code, and the permission verification module compares the permission level with the minimum permission requirement for the queried data category.When the permission level is greater than or equal to the minimum permission requirement, an access permission signal is generated. The access control verification processor retrieves the corresponding data item from the encrypted storage array based on the access permission signal. It determines the number of storage nodes and the range of data fragments that can be accessed based on the requesting user's permission level, and provides the requesting user with a corresponding number of encryption key fragments. The requesting user uses these encryption key fragments to reconstruct the data encryption key and decrypt it to obtain the data item content corresponding to their permission level. The data item is then returned to the requesting user through the API gateway and the conversational user interface. If the permission level is lower than the minimum permission requirement, an access denial signal is generated, and the unauthorized access attempt is recorded.

11. The method as described in claim 10, wherein the semantic feature extraction step is performed using a deep learning language model, the visual feature recognition step is performed using a neural network model, and the structural analysis step is performed using a document layout analysis algorithm.

12. The method as described in claim 10, wherein the step of calculating the classification preference coefficient vector further comprises using a time decay weight function that calculates weights by decreasing them according to a time interval to generate the classification preference coefficient vector of the user within a preset time interval.

13. The method as described in claim 10, wherein the weighted fusion operation step utilizes an attention-weight-based mechanism to perform feature fusion on the semantic vector, the visual vector, and the structural vector to generate the unified feature vector.

14. The method as described in claim 10, wherein the step of generating a category suggestion further comprises generating a single category suggestion when the difference between the highest probability and the second highest probability exceeds the preset threshold, and otherwise generating a plurality of category suggestions ranked with confidence scores.

15. The method as described in claim 10, wherein the step of extracting the user entity, group entity, and data category entity is performed using a machine learning model, and the step of determining the operation type is performed using a classification model.

16. The method as described in claim 10, wherein the encryption processing step employs a symmetric encryption algorithm, the redundancy coding step employs an error-correcting code, and the step of providing encryption key fragments determines the number provided based on the permission level of the requesting user, wherein the number of encryption key fragments is positively correlated with the permission level.

17. The method as described in claim 10 further includes an anomaly detection step, which monitors the query frequency and query content pattern of the requesting user, and generates a security alarm signal and suspends the access rights of the requesting user until identity re-verification is completed when the query frequency exceeds a preset frequency or the query content deviates from the historical pattern of the requesting user by more than a set threshold.

18. The method as described in claim 10, wherein the routing step further includes a load balancing process that, when multiple upload requests are received simultaneously, distributes the requests to different processing queues according to the current system load, each processing queue having a preset capacity, and when the capacity is exceeded, returns a system busy signal to the user and suggests retrying later.

Citation Information

Patent Citations

  • Intelligent file management method and system

    CN120974275A

  • Method, apparatus and system for secure file management

    TW202443425A

  • Shared pharmacy system

    TWM670637U