A multi-level access control and privacy protection oriented experimental animal ethics review management system

Abstract

The application discloses a multi-level access control and privacy protection-oriented experimental animal ethics review management system and particularly relates to the field of experimental animal ethics review, and comprises a permission access module, a semantic modeling module, an expert matching module, a review closed loop module, a compliance archiving module and a strategy evolution module. By constructing a bidirectional semantic path alignment mechanism between a high-order index tensor driven by semantic modeling and an expert portrait semantic atlas, and in combination with permission hierarchical control, review feedback closed loop and strategy evolution scheduling model, automation and structured management of information extraction, expert allocation and feedback processing in the whole process of experimental animal ethics application are realized, so that problems such as manual processing lag, insufficient matching accuracy and data untraceability in the existing review process are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of laboratory animal ethics review technology, and more specifically, to a laboratory animal ethics review management system for multi-level access control and privacy protection. Background Technology

[0002] In the existing ethical review process for laboratory animals, the submission and review of application materials mainly rely on manual processing. Specifically, applicants submit materials by paper or send electronic materials by email, which are then manually sorted and distributed by staff to experts for offline or email review. Finally, staff manually assign codes and provide feedback on the approval documents.

[0003] In the context of rapidly increasing ethical review volume and highly concentrated application time, this approach has exposed systemic problems such as slow process response, arbitrary expert allocation, insufficient approval consistency, and difficulty in subsequent data traceability and management. Although some systems have attempted to introduce preliminary automation methods through keyword extraction and static text matching based on cosine similarity, their algorithm models lack semantic understanding and context modeling capabilities, resulting in the accuracy and professional fit of expert matching still heavily relying on manual intervention, making it difficult to support ethical review scenarios with large-scale, high-frequency, and high-accuracy requirements.

[0004] Therefore, the current ethical review process for laboratory animals lacks an automated processing mechanism with deep semantic understanding and intelligent expert matching capabilities, which has become a core technical bottleneck restricting the efficiency and fairness of the review. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a laboratory animal ethics review and management system oriented towards multi-level access control and privacy protection. By constructing a bidirectional semantic path alignment mechanism between a high-order index tensor driven by semantic modeling and an expert profile semantic graph, and combining hierarchical access control, closed-loop review feedback, and strategy evolution scheduling model, the system achieves automated and structured management of information extraction, expert allocation, and feedback processing throughout the entire process of laboratory animal ethics application. This effectively solves the problems of slow manual processing, insufficient matching accuracy, and untraceable data in the existing review process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an experimental animal ethics review and management system for multi-level access control and privacy protection, including an access permission module, a semantic modeling module, an expert matching module, a review closed-loop module, a compliance archiving module, and a strategy evolution module;

[0007] The access module is used to identify the access level and control the field display of the identity data, unit information and project attributes of the user applying for ethics, and to perform hierarchical encryption processing on the data filled in according to the access level, generating a multi-level encrypted and encapsulated ethics application data packet;

[0008] The semantic modeling module is used to perform semantic parsing and term annotation operations on ethics application data packets, construct keyword classification graphs and generate semantic vector matrix representations, and construct semantic index tensors in combination with four types of review and judgment indicators;

[0009] The expert matching module is used to construct a semantic graph from the expert behavior vector group, cross-compare the application semantic index tensor with the execution path of the expert semantic graph, and output the sorted expert matching candidate set by combining three types of scheduling factors.

[0010] The closed-loop review module generates review tasks based on the expert matching candidate set and constructs a standardized feedback structure. It then performs consistency comparison and sentiment shift judgment on the applicant's revised response, triggers the expert replacement process, and reconstructs the expert matching list.

[0011] The compliance archiving module is used to merge the application semantic index tensor with the expert path structure to generate an approval structure diagram. After generating and verifying the consistency of Chinese and English approval documents, it assigns a unique number and archives the structure diagram and related factors into the database to establish a searchable semantic index system.

[0012] The strategy evolution module is used to generate evolutionary training samples based on expert behavior trajectories, extract three types of behavioral factor groups and feed them back to the expert profile map to update the matching strategy parameters, and push the updated results to the semantic matching engine to achieve adaptive scheduling optimization.

[0013] In a preferred embodiment, the access module is used to input the ethical application user identification data, the applicant unit registration information and the scientific research project attribute data as an access identity set into the login access system, and obtain the access permission level identifier through the access control judgment process.

[0014] In the control and judgment process, the access permission level identifier is mapped to the sensitive information classification template to perform hierarchical permission judgment operations and generate a data presentation template with display fields. The data presentation template is embedded in the ethics application input interface to protect sensitive fields by permission shielding. Sensitive fields include experimental animal species, experimental operation methods and biological intervention path information. The application data filled in by the user is encrypted according to the field permission level, and the multi-level encrypted encapsulated ethical application data packet is output as an unstructured data source.

[0015] In a preferred embodiment, the semantic modeling module is used to input the ethical application data package into the semantic parsing process to perform text normalization, entity recognition and dependency syntax extraction operations, and construct a grammatical hierarchical tree structure; the subject-verb-object chain, prepositional phrase and additional components in the grammatical hierarchical tree structure are labeled according to the ethical terminology lexicon and classification hierarchical index to form a multidimensional keyword structure set; the multidimensional keyword structure set is processed by the semantic hierarchical mapping function to obtain a keyword classification map including a two-layer structure of main concept and sub-concept;

[0016] The keyword classification map is input into the context embedding model to generate a semantic vector matrix representation. Combined with four categories of indicators—experiment content type, animal subjects, intervention methods, and ethical sensitivity level—a high-order semantic index tensor is constructed.

[0017] In a preferred embodiment, the expert matching module is used to construct an expert profile by combining expert registration information, historical review records, and processed experimental type data into an expert behavior vector group, and extracting its research topic path vector and feedback preference weight factor.

[0018] The expert behavior vector group is constructed into a node-side graph structure and a graph embedding algorithm is executed to form an expert semantic graph with directional path and entity association degree. The application semantic index tensor and the expert semantic graph are input into the semantic path comparison engine. The path overlap score is calculated through the path cross-score function and the semantic consistency level index is labeled.

[0019] The consistency level index is weighted and fused with three scheduling factors: the expert's current workload, the task response frequency, and the acceptance boundary range, to output a sorted set of expert matching candidates.

[0020] In a preferred embodiment, the review closed-loop module combines the expert matching candidate set with the scheduling priority and performs a combination selection based on the current acceptable status and historical conflict factors, generates an expert assignment task sheet and starts the review process, and constructs the review results and opinion text output by the review process into a standardized feedback structure, the structure fields of which include review pass label, rejection reason identifier and revision suggestion factor group;

[0021] The standardized feedback structure is returned to the applicant's interface. The applicant submits the revised content, and the response analysis module parses it to generate a semantic response chain. The semantic response chain is compared with the original revision suggestion factor group to generate a revision consistency score index. If the index is less than the lower limit threshold for consistency judgment, it is determined to be a conflicting revision path.

[0022] An emotional response intensity assessment process is performed on the conflict revision path to calculate the opinion shift tensor between the applicant and the original expert. If the shift tensor exceeds the upper limit threshold for objection judgment, an expert replacement process is triggered. In the expert replacement process, the shift tensor and the original semantic path are reconstructed and fed back to the expert profile graph to filter out expert paths that are highly consistent with the applicant's conflict labels and generate a new matching list.

[0023] In a preferred embodiment, the compliance archiving module is used to merge the approved application semantic index tensor with the expert path and access permission level identifier structure to form the final approval structure diagram.

[0024] The system generates approval documents by executing the approval structure diagram and generating Chinese and English approval document content trees based on language preference parameters. Before output, a consistency verification mechanism is used to check the structural isomorphism relationship between the approval structure diagram and the expert feedback path. After the verification is passed, the system assigns a unique number to each approval structure diagram and generates the final PDF approval document. The approval structure diagram, expert path diagram, and feedback identification group are archived together in the ethical approval structured database, and a semantic reverse index table is established with keyword vectors as the index.

[0025] The database supports structured multi-condition search and semantic question-answering retrieval based on keyword indexing, applicant identity, animal type, and ethical rating tags.

[0026] In a preferred embodiment, the strategy evolution module is used to form an expert behavior trajectory set by combining each expert review path, response content label, rejection reason type, and task completion cycle, and use it as input to form an expert evolution training sample; the expert evolution training sample is input into the task response modeler to extract three types of behavioral factors: domain interest change trend vector, feedback tendency coefficient, and task saturation state factor; the three types of behavioral factors are fed back to the expert profile map and the expert behavior boundary is reconstructed, updating the matching probability distribution, rejection threshold strategy, and acceptance boundary range in the expert map;

[0027] The updated expert graph results are pushed to the semantic matching engine, and an adaptive scheduling strategy is applied in subsequent matching decisions to optimize time sensitivity and matching efficiency.

[0028] In a preferred embodiment, the semantic modeling module is used to perform semantic parsing and term annotation operations on the ethics application data package, construct a keyword classification map and generate a semantic vector matrix expression, and construct a high-order semantic index tensor by combining four categories of review and judgment indicators: experimental content type, animal object, intervention method and ethical sensitivity level.

[0029]

[0030] Where τ ijklv represents the semantic index value of the i-th type of experimental content, the j-th type of animal object, the k-th type of intervention method, and the l-th type of ethical sensitivity level in the higher-order semantic index tensor; M represents the total number of identified terms in the ethical application text; v (m) Let be the context word vector representation of the m-th term; σ(·) is a non-linear activation function, which is used to enhance the context semantic responsiveness; f represents the context modeling function ctx In the input context window x (m) The gradient of f with respect to the parameter θ; θ is the set of trainable parameters of the context modeling function; f ctx (·) is the context modeling function; x (m) The context window text fragment in the original text where the m-th term is located; The structure mapping function indicates whether the term m belongs to the tensor position (i,j,k,l), and takes the value 1 or 0.

[0031] In a preferred embodiment, the expert matching module is used to construct a semantic graph from the expert behavior vector group, and perform semantic path cross-comparison between the application semantic index tensor and the expert semantic graph. The module outputs a sorted set of expert matching candidates by weighted fusion of the path overlap score and three types of expert scheduling factors.

[0032]

[0033] in The overall matching score of the semantic tensor between expert e and the ethics application;

[0034] The set of paths in the expert semantic graph; To request the set of path structures in the semantic index tensor; u e (t) is the semantic embedding vector of the t-th node in the expert path; u app (t) is the semantic embedding vector of the t-th node in the semantic path; <·,·> are the vector dot product operators; ||·|| is the L2 norm of the vector; ξ(t) is the path node importance function; Γ(λ) e ,μ e ,ν e ) is the scheduling harmonic function; λ e This represents the expert's current workload intensity index; μ e ν represents the expert response frequency coefficient. e This is the expert semantic deviation coefficient.

[0035] In a preferred embodiment, the review closed-loop module is used to generate review tasks based on expert matching results, and generate revision consistency scores and sentiment shift tensors based on expert feedback and applicant revision responses. If both the inconsistency and sentiment shift conditions are met simultaneously, the expert replacement process is triggered.

[0036]

[0037] Where S edit The semantic consistency score between the applicant's revised response and the original expert recommendations; N is the total number of revision factor items compared; ω i r is the semantic importance weight of the i-th suggestion factor; i The vector representation in semantic space originally suggested by experts; s i The vector representation of the applicant's response in the semantic space; cos(·,·) is the cosine similarity function; The sentiment shift tensor distance between expert feedback and applicant response; M is the total number of sentiment dimensions analyzed; β j Let be the semantic sensitivity weight of the j-th sentiment dimension; The vector projection of the applicant's response text onto the j-th sentiment dimension; τ is the vector projection of the original expert feedback text onto the j-th sentiment dimension; s τ is the lower limit threshold for semantic consistency judgment; d This is the upper limit threshold for the intensity of emotional shift; ExpertSwitch is the expert replacement decision variable, and a value of 1 triggers the expert replacement process.

[0038] The technical effects and advantages of this invention are as follows:

[0039] 1. By inputting the ethical application text into the semantic modeling module to construct a high-order semantic index tensor, and performing semantic path cross-comparison with the expert profile graph, the manual screening mechanism is effectively replaced, fundamentally improving the accuracy and automation of expert matching, and solving the problem of lagging manual processing under large-scale applications;

[0040] 2. By embedding a field-level permission judgment process in the permission access module, the user access level is logically mapped to the sensitive information field template to generate the minimum displayable data presentation structure, thereby realizing the controllable exposure of sensitive information and improving the data privacy and security protection capabilities.

[0041] 3. By constructing the application corpus into a keyword classification map and generating a semantic vector matrix, and then combining it with four types of judgment indicators to form a multidimensional tensor expression, the ethical text has a clear and structured semantic expression, which facilitates subsequent intelligent reasoning, rapid retrieval and approval generation.

[0042] 4. By using the response consistency scoring and sentiment shift tensor judgment mechanism in the closed-loop review module, intelligent comparison and cross-validation of the applicant's revised content after expert rejection can be achieved, thereby improving the accuracy of human-computer interaction and the rigor of the process in expert feedback processing.

[0043] 5. By performing structural isomorphism verification between the approval path structure and expert feedback results, and then completing the numbering allocation and bilingual document generation process, we ensure that the output documents have consistency and verifiability in logical and semantic structure, thereby strengthening the traceability of subsequent filing and retrieval.

[0044] 6. By introducing a task response behavior training sample generation mechanism, we extract the interest change vector, task saturation factor and feedback tendency parameter, dynamically reconstruct the matching probability distribution in the expert profile map, and realize the adaptive optimization of the review expert recommendation strategy as it evolves with usage behavior. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0046] 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, and 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.

[0047] Refer to the instruction manual appendix Figure 1 An embodiment of the present invention provides an experimental animal ethics review and management system for multi-level access control and privacy protection, comprising an access permission module, a semantic modeling module, an expert matching module, a review closed-loop module, a compliance archiving module, and a strategy evolution module.

[0048] The access module is used to identify the permission level and control the minimum display of fields for the identity data, unit information and project attributes of the ethics application user, and to perform hierarchical encryption processing on the filled data according to the permission level, generating a multi-level encrypted and encapsulated ethics application data packet;

[0049] The semantic modeling module is used to perform semantic parsing and term annotation operations on ethics application data packets, construct keyword classification graphs and generate semantic vector matrix representations, and construct semantic index tensors in combination with four types of review and judgment indicators;

[0050] The expert matching module is used to construct a semantic graph from the expert behavior vector group, cross-compare the application semantic index tensor with the execution path of the expert semantic graph, and output the sorted expert matching candidate set by combining three types of scheduling factors.

[0051] The closed-loop review module generates review tasks based on the expert matching candidate set and constructs a standardized feedback structure. It then performs consistency comparison and sentiment shift judgment on the applicant's revised response, triggers the expert replacement process, and reconstructs the expert matching list.

[0052] The compliance archiving module is used to merge the application semantic index tensor with the expert path structure to generate an approval structure diagram. After generating and verifying the consistency of Chinese and English approval documents, it assigns a unique number and archives the structure diagram and related factors into the database to establish a searchable semantic index system.

[0053] The strategy evolution module is used to generate evolutionary training samples based on expert behavior trajectories, extract three types of behavioral factor groups and feed them back to the expert profile map to update the matching strategy parameters, and push the updated results to the semantic matching engine to achieve adaptive scheduling optimization.

[0054] The access control module is used to input the user identification data of the ethics application, the registration information of the applicant unit and the attribute data of the scientific research project as the access identity set into the login access system, and obtain the access permission level identifier through the access control judgment process;

[0055] In the control and judgment process, the access permission level identifier is mapped to the sensitive information classification template to perform hierarchical permission judgment operations, generating a data presentation template with the minimum displayable fields; the data presentation template is embedded in the ethics application input interface to protect sensitive fields by permission shielding, including experimental animal species, experimental operation methods and biological intervention path information; the application data filled in by the user is encrypted hierarchically according to the field permission level, and the multi-level encrypted encapsulated structure of the ethics application data packet is output as an unstructured data source.

[0056] The semantic modeling module is used to input the ethical application data package into the semantic parsing process to perform text normalization, entity recognition and dependency syntax extraction operations, and construct a grammatical hierarchical tree structure. The subject-verb-object chain, prepositional phrase and additional components in the grammatical hierarchical tree structure are labeled according to the ethical terminology lexicon and classification hierarchical index to form a multi-dimensional keyword structure set. The multi-dimensional keyword structure set is processed by the semantic hierarchical mapping function to obtain a keyword classification map including a two-layer structure of main concept and sub-concept.

[0057] The keyword classification map is input into the context embedding model to generate a semantic vector matrix representation. Combined with four categories of indicators—experiment content type, animal subjects, intervention methods, and ethical sensitivity level—a high-order semantic index tensor is constructed.

[0058] The expert matching module is used to construct expert profiles by combining expert registration information, historical review records, and processed experiment type data into expert behavior vector groups, and extract their research topic path vectors and feedback preference weight factors.

[0059] The expert behavior vector group is constructed into a node-side graph structure and a graph embedding algorithm is executed to form an expert semantic graph with directional path and entity association degree. The application semantic index tensor and the expert semantic graph are input into the semantic path comparison engine. The path overlap score is calculated through the path cross-score function and the semantic consistency level index is labeled.

[0060] The consistency level index is weighted and fused with three scheduling factors: the expert's current workload, the task response frequency, and the acceptance boundary range, to output a sorted set of expert matching candidates.

[0061] The closed-loop review module combines the expert matching candidate set with the scheduling priority and performs a combination selection based on the current acceptable status and historical conflict factors. It generates an expert assignment task sheet and starts the review process. The review results and opinion text output by the review process are constructed into a standardized feedback structure. The structure fields include review pass label, rejection reason identifier and revision suggestion factor group.

[0062] The standardized feedback structure is returned to the applicant's interface. The applicant submits the revised content, and the response analysis module parses it to generate a semantic response chain. The semantic response chain is compared with the original revision suggestion factor group to generate a revision consistency score index. If the index is less than the lower limit threshold for consistency judgment, it is determined to be a conflicting revision path.

[0063] An emotional response intensity assessment process is performed on the conflict revision path to calculate the opinion shift tensor between the applicant and the original expert. If the shift tensor exceeds the upper limit threshold for objection judgment, an expert replacement process is triggered. In the expert replacement process, the shift tensor and the original semantic path are reconstructed and fed back to the expert profile graph to filter out expert paths that are highly consistent with the applicant's conflict labels and generate a new matching list.

[0064] The compliance archiving module is used to merge the semantic index tensor of approved applications with the expert path and access permission level identifier structure to form the final approval structure diagram;

[0065] The system generates approval documents from the approval structure diagram and generates Chinese and English approval document content trees based on language preference parameters. Before output, a consistency verification mechanism checks the structural isomorphism between the approval structure diagram and the expert feedback path. After verification, the system assigns a unique number to each approval structure diagram and generates a final PDF approval document, supporting export in both Chinese and English formats. The approval structure diagram, expert path diagram, and feedback identifier group are archived together in the ethical approval structured database, and a semantic inverse index table is established with keyword vectors as the index.

[0066] The database supports structured multi-condition search and semantic question-answering retrieval based on keyword indexing, applicant identity, animal type, and ethical rating tags.

[0067] The strategy evolution module is used to assemble a set of expert behavior trajectories by combining each expert review path, response content tag, rejection reason type, and task completion cycle, and use this as input to form expert evolution training samples. The expert evolution training samples are then input into the task response modeler to extract three types of behavioral factors: domain interest change trend vector, feedback tendency coefficient, and task saturation state factor. These three types of behavioral factors are then fed back to the expert profile map to reconstruct the expert behavior boundary and update the matching probability distribution, rejection threshold strategy, and acceptance boundary range in the expert map.

[0068] The updated expert graph results are pushed to the semantic matching engine, and an adaptive scheduling strategy is applied in subsequent matching decisions to optimize time sensitivity and matching efficiency.

[0069] It should be noted that in the formula structure involved in this scheme, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system. This combination of "dimensionless terms and terms with units" can be understood as a composite structural expression commonly used in mathematical physics modeling. It conforms to the principle of dimensional consistency and has a clear physical interpretation basis.

[0070] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can form a unified structure through function mapping, ratio combination or normalization adjustment, with clear units and clear meaning. The overall expression conforms to the principle of dimensional consistency and the conventional formula of engineering modeling.

[0071] In this scheme, constants, weights, adjustment factors, threshold parameters, proportional coefficients, etc., are all adjustable control parameters for different application environments. Their values ​​depend on the target equipment configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are converged within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have a unique preset value, they have clear adjustment logic and calculation paths. They belong to the deterministic setting process in engineering implementation. The purpose of this setting is to ensure that the scheme is both universally adaptable and reproducible and operable, without affecting its technical clarity and feasibility.

[0072] The semantic modeling module is used to perform semantic parsing and term annotation operations on ethics application data packets, construct keyword classification maps and generate semantic vector matrix expressions, and construct high-order semantic index tensors by combining four categories of review and judgment indicators: experimental content type, animal subjects, intervention methods and ethical sensitivity level.

[0073]

[0074] Where τ ijkl v represents the semantic index value of the i-th type of experimental content, the j-th type of animal object, the k-th type of intervention method, and the l-th type of ethical sensitivity level in the higher-order semantic index tensor; M represents the total number of identified terms in the ethical application text; v (m) Let be the context word vector representation of the m-th term; σ(·) is a non-linear activation function, which is used to enhance the context semantic response capability, including but not limited to ReLU or GELU; f represents the context modeling function ctx In the input context window x (m) The gradient of the parameter θ is used to reflect the change in semantic sensitivity of the term in the context; θ is the set of trainable parameters of the context modeling function; f ctx (·) represents the context modeling function, which generates a semantic environment vector of a term within the syntactic structure; x (m) The context window text fragment in the original text where the m-th term is located; The structure mapping function indicates whether the term m belongs to the tensor position (i,j,k,l), and takes the value 1 or 0.

[0075] The expert matching module is used to construct a semantic graph from the expert behavior vector group, and to perform semantic path cross-comparison between the application semantic index tensor and the expert semantic graph. The module then outputs a sorted set of expert matching candidates by weighted fusion of path overlap score and three types of expert scheduling factors.

[0076]

[0077] in The overall matching score of the semantic tensor between expert e and the ethics application;

[0078] The path set in the expert semantic graph is used to represent the path structure of its research topic. To request the set of path structures in the semantic index tensor; u e (t) is the semantic embedding vector of the t-th node in the expert path; u app (t) is the semantic embedding vector of the t-th node in the semantic path; <·,·> are the vector dot product operators, which represent semantic similarity; ||·|| is the L2 norm (Euclidean modulus) of the vector, used for normalization; ξ(t) is the path node importance function, which represents the proportion of the t-th node in semantic matching; Γ(λ e ,μ e ,ν e) represents the scheduling harmonic function, which is used to fuse the current state parameters of the experts; λ e This refers to the expert's current workload intensity index, which represents the task density function value of the tasks the expert has received; μ e The expert response frequency coefficient represents the rate at which tasks are processed per unit of time; ν e The expert semantic deviation coefficient is used to measure the overall semantic distance between the current expert semantic topic and the current application.

[0079] The closed-loop review module is used to generate review tasks based on expert matching results, and generate revision consistency scores and sentiment shift tensors based on expert feedback and applicant revision responses. If both the inconsistency and sentiment shift conditions are met, the expert replacement process is triggered.

[0080]

[0081] Where S edit The semantic consistency score between the applicant's revised response and the original expert recommendations; N is the total number of revision factor items compared; ω i r is the semantic importance weight of the i-th suggestion factor; i The vector representation in semantic space originally suggested by experts; s i The vector representation of the applicant's response in the semantic space; cos(·,·) is the cosine similarity function, which is used to quantify the degree of similarity between two semantic vectors; β represents the sentiment shift tensor distance between expert feedback and applicant response; M represents the total number of sentiment dimensions analyzed (e.g., denial, questioning, rebuttal, mitigation, etc.); j Let be the semantic sensitivity weight of the j-th sentiment dimension; The vector projection of the applicant's response text onto the j-th sentiment dimension; τ is the vector projection of the original expert feedback text onto the j-th sentiment dimension; s This is the lower threshold for semantic consistency judgment; values ​​below this value are considered insufficient responses. d This is the upper limit threshold for the intensity of emotional shift; values ​​exceeding this threshold are considered severe opinion conflicts. ExpertSwitch is the variable for determining expert replacement; a value of 1 triggers the expert replacement process.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A laboratory animal ethics review and management system for multi-level access control and privacy protection, comprising a permission access module, a semantic modeling module, an expert matching module, a review closed-loop module, a compliance archiving module, and a policy evolution module, characterized in that: The access module is used to identify the access level and control the field display of the identity data, unit information and project attributes of the user applying for ethics, and to perform hierarchical encryption processing on the data filled in according to the access level, generating a multi-level encrypted and encapsulated ethics application data packet; The semantic modeling module is used to perform semantic parsing and term annotation operations on ethics application data packets, construct keyword classification graphs and generate semantic vector matrix representations, and construct semantic index tensors in combination with four types of review and judgment indicators; The expert matching module is used to construct a semantic graph from the expert behavior vector group, cross-compare the application semantic index tensor with the execution path of the expert semantic graph, and output the sorted expert matching candidate set by combining three types of scheduling factors. The closed-loop review module generates review tasks based on the expert matching candidate set and constructs a standardized feedback structure. It then performs consistency comparison and sentiment shift judgment on the applicant's revised response, triggers the expert replacement process, and reconstructs the expert matching list. The compliance archiving module is used to merge the application semantic index tensor with the expert path structure to generate an approval structure diagram. After generating and verifying the consistency of Chinese and English approval documents, it assigns a unique number and archives the structure diagram and related factors into the database to establish a searchable semantic index system. The strategy evolution module is used to generate evolutionary training samples based on expert behavior trajectories, extract three types of behavioral factor groups and feed them back to the expert profile map to update the matching strategy parameters, and push the updated results to the semantic matching engine to achieve adaptive scheduling optimization.

2. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 1, characterized in that: The access control module is used to input the user identification data of the ethics application, the registration information of the applicant unit and the attribute data of the scientific research project as the access identity set into the login access system, and obtain the access permission level identifier through the access control judgment process; In the control judgment process, the access permission level identifier is mapped to the sensitive information classification template to perform the permission layering logic judgment operation and generate a data presentation template with display fields; The data presentation template is embedded in the ethics application input interface to protect sensitive fields by access control. These sensitive fields include the species of experimental animals, experimental procedures, and biological intervention pathways. The application data submitted by the user is encrypted hierarchically according to the field access level, and the ethics application data package with a multi-level encrypted encapsulation structure is output as an unstructured data source.

3. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 2, characterized in that: The semantic modeling module is used to input the ethical application data package into the semantic parsing process to perform text normalization, entity recognition and dependency syntax extraction operations, and construct a grammatical hierarchical tree structure. The subject-verb-object chain, prepositional phrase and additional components in the grammatical hierarchical tree structure are labeled according to the ethical terminology lexicon and classification hierarchical index to form a multi-dimensional keyword structure set. The multi-dimensional keyword structure set is processed by the semantic hierarchical mapping function to obtain a keyword classification map including a two-layer structure of main concept and sub-concept. The keyword classification map is input into the context embedding model to generate a semantic vector matrix representation. Combined with four categories of indicators—experiment content type, animal subjects, intervention methods, and ethical sensitivity level—a high-order semantic index tensor is constructed.

4. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 3, characterized in that: The expert matching module is used to construct expert profiles by combining expert registration information, historical review records, and processed experiment type data into expert behavior vector groups, and extract their research topic path vectors and feedback preference weight factors. The expert behavior vector group is constructed into a node-side graph structure and a graph embedding algorithm is executed to form an expert semantic graph with directional path and entity association degree. The application semantic index tensor and the expert semantic graph are input into the semantic path comparison engine. The path overlap score is calculated through the path cross-score function and the semantic consistency level index is labeled. The consistency level index is weighted and fused with three scheduling factors: the expert's current workload, the task response frequency, and the acceptance boundary range, to output a sorted set of expert matching candidates.

5. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 4, characterized in that: The closed-loop review module combines the expert matching candidate set with the scheduling priority and performs a combination selection based on the current acceptable status and historical conflict factors. It generates an expert assignment task sheet and starts the review process. The review results and opinion text output by the review process are constructed into a standardized feedback structure. The structure fields include review pass label, rejection reason identifier and revision suggestion factor group. The standardized feedback structure is returned to the applicant's interface. The applicant submits the revised content, and the response analysis module parses it to generate a semantic response chain. The semantic response chain is compared with the original revision suggestion factor group to generate a revision consistency score index. If the index is less than the lower limit threshold for consistency judgment, it is determined to be a conflicting revision path. An emotional response intensity assessment process is performed on the conflict revision path to calculate the opinion shift tensor between the applicant and the original expert. If the shift tensor exceeds the upper limit threshold for objection judgment, an expert replacement process is triggered. In the expert replacement process, the shift tensor and the original semantic path are reconstructed and fed back to the expert profile graph to filter out expert paths that are highly consistent with the applicant's conflict labels and generate a new matching list.

6. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 5, characterized in that: The compliance archiving module is used to merge the semantic index tensor of approved applications with the expert path and access permission level identifier structure to form the final approval structure diagram; The approval structure diagram will be used to generate approval documents, and a Chinese and English approval document content tree will be generated based on language preference parameters. Before output, a consistency verification mechanism checks its structural isomorphism with the expert feedback path; After verification, the system assigns a unique number to each approval structure diagram and generates a final PDF approval document; the approval structure diagram, expert path diagram, and feedback identification group are archived together in the ethical approval structured database, and a semantic reverse index table is established with keyword vectors as the index. The database supports structured multi-condition search and semantic question-answering retrieval based on keyword indexing, applicant identity, animal type, and ethical rating tags.

7. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 6, characterized in that: The strategy evolution module is used to assemble a set of expert behavior trajectories by combining each expert review path, response content tag, rejection reason type, and task completion cycle, and use this as input to form expert evolution training samples. The expert evolution training samples are then input into the task response modeler to extract three types of behavioral factors: domain interest change trend vector, feedback tendency coefficient, and task saturation state factor. These three types of behavioral factors are then fed back to the expert profile map to reconstruct the expert behavior boundary and update the matching probability distribution, rejection threshold strategy, and acceptance boundary range in the expert map. The updated expert graph results are pushed to the semantic matching engine, and an adaptive scheduling strategy is applied in subsequent matching decisions to optimize time sensitivity and matching efficiency.

8. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 7, characterized in that: The semantic modeling module is used to perform semantic parsing and term annotation operations on ethics application data packets, construct keyword classification maps and generate semantic vector matrix expressions, and construct high-order semantic index tensors by combining four categories of review and judgment indicators: experimental content type, animal subjects, intervention methods and ethical sensitivity level. Where τ ijkl v represents the semantic index value of the i-th type of experimental content, the j-th type of animal object, the k-th type of intervention method, and the l-th type of ethical sensitivity level in the higher-order semantic index tensor; M represents the total number of identified terms in the ethical application text; v (m) Let be the context word vector representation of the m-th term; σ(·) is a non-linear activation function, which is used to enhance the context semantic responsiveness. f represents the context modeling function ctx In the input context window x (m) The gradient of the function with respect to the parameter θ; θ is the set of trainable parameters of the context modeling function; f ctx (·) is the context modeling function; x (m) The context window text fragment in the original text where the m-th term is located; The structure mapping function indicates whether the term m belongs to the tensor position (i,j,k,l), and takes the value 1 or 0.

9. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 8, characterized in that: The expert matching module is used to construct a semantic graph from the expert behavior vector group, and to perform semantic path cross-comparison between the application semantic index tensor and the expert semantic graph. The module then outputs a sorted set of expert matching candidates by weighted fusion of path overlap score and three types of expert scheduling factors. in The overall matching score of the semantic tensor between expert e and the ethics application; The set of paths in the expert semantic graph; To request the set of path structures in the semantic index tensor; u e (t) is the semantic embedding vector of the t-th node in the expert path; u app (t) is the semantic embedding vector of the t-th node in the semantic path; <·,·> are the vector dot product operators; ||·|| is the L2 norm of the vector; ξ(t) is the path node importance function; Γ(λ) e ,μ e ,ν e ) is the scheduling harmonic function; λ e This represents the expert's current workload intensity index; μ e ν represents the expert response frequency coefficient. e This is the expert semantic deviation coefficient.

10. The experimental animal ethics review and management system for multi-level access control and privacy protection as described in claim 9, characterized in that: The closed-loop review module is used to generate review tasks based on expert matching results, and generate revision consistency scores and sentiment shift tensors based on expert feedback and applicant revision responses. If both the inconsistency and sentiment shift conditions are met, the expert replacement process is triggered. Where S edit The semantic consistency score between the applicant's revised response and the original expert recommendations; N is the total number of revision factor items compared; ω i Let be the semantic importance weight of the i-th suggestion factor; r i The vector representation in semantic space originally suggested by experts; s i The vector representation of the applicant's response in the semantic space; cos(·,·) is the cosine similarity function; β represents the sentiment shift tensor distance between expert feedback and applicant responses; M represents the total number of sentiment dimensions analyzed; β j Let be the semantic sensitivity weight of the j-th sentiment dimension; The vector projection of the applicant's response text onto the j-th sentiment dimension; τ is the vector projection of the original expert feedback text onto the j-th sentiment dimension; s τ is the lower threshold for semantic consistency judgment; d This is the upper limit threshold for the intensity of emotional shift; ExpertSwitch is the expert replacement decision variable, and a value of 1 triggers the expert replacement process.

Images