A secure file transmission system based on path encryption and dual-role account management

Abstract

This invention discloses a secure file transfer system based on path encryption and dual-role account management, comprising: a system administrator module for generating storage server access parameters; a confidentiality administrator module for generating basic storage server access configurations; a storage server module for providing read and write services for the actual directory structure and file data; an encrypted path component for constructing a path mapping table; a client module for initiating file access requests and displaying the returned file list and file data; a proxy transmission risk control module for generating a path risk scoring model based on an improved MetaPath2Vec model; a proxy transmission execution module for recording transmission audit data and updating the risk model; and a transmission administrator module for reviewing and returning approval results. This invention employs path encryption and dual-role management to achieve risk identification and security control throughout the entire file transfer process.

Description

Technical Field

[0001] This invention relates to the field of information security and data transmission technology, and in particular to a secure file transfer system based on path encryption and dual-role account management. Background Technology

[0002] In existing file transfer systems, clients typically send access requests directly to the server based on the actual path. The server then verifies the request according to access permissions and returns the file content. In this design, the actual path information is directly exposed to the client, allowing attackers to infer the server's directory structure based on path characteristics, increasing the risk of sensitive file leakage.

[0003] Meanwhile, access control mechanisms are mostly based on single-role permission judgments, lacking correlation analysis of access behaviors themselves, and unable to identify potential risky access. In addition, traditional systems generally rely on static permission policies, and cannot dynamically adjust security policies according to changes in access logs or historical transmission records.

[0004] Existing technologies typically employ rule-based detection methods for risk identification, which cannot express the relationships between complex access behaviors, resulting in insufficient ability to identify abnormal path combinations, unauthorized access links, and spoofed access behaviors.

[0005] Access credential management for storage servers is mostly maintained by a single role, lacking a role-based permission isolation mechanism, which poses management risks. At the same time, the session key and token verification mechanism during file transfer lacks a dynamic generation method based on path behavior, making it difficult to achieve node-by-node verification of the access link. With constantly changing access behavior, the existing system lacks a model update mechanism based on access logs and transmission records, and cannot adaptively adjust the risk assessment model, making it difficult to continuously improve the ability to identify access risks.

[0006] Therefore, how to provide a secure file transfer system based on path encryption and dual-role account management is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a secure file transfer system based on path encryption and dual-role account management. This invention completely isolates the real path from the client logic, constructs an access interface that exposes only the encrypted path identifier using a path encryption mapping method, and introduces a dual-role account management mechanism to hierarchically control access credentials to the storage server, thereby improving the security of credential management. This invention combines access log data to construct a heterogeneous path behavior graph structure, and utilizes an improved MetaPath2Vec model and a path policy automaton to construct a path risk scoring model, achieving risk identification and dynamic access control of the file access link. Simultaneously, the model is updated based on actual transmission events to maintain evaluation capabilities. This invention achieves high-strength path hiding, fine-grained access control, and intelligent risk identification, possessing the advantages of unpredictable paths, quantifiable access behavior analysis, and adaptively updated security policies.

[0008] A secure file transfer system based on path encryption and dual-role account management according to an embodiment of the present invention includes: The system administrator module is used to generate storage server access parameters; The confidentiality administrator module is used to enter access passwords or keys and store them encrypted, forming encrypted credential data and access control rules, which are combined with storage server access parameters to generate the basic configuration for storage server access. The storage server module is used to provide read and write services for the actual directory structure and file data based on the storage server access base configuration; The encrypted path component is used to obtain the real path set based on the storage server access basic configuration and build a path mapping table; The client module is used to initiate file access requests based on encrypted path identifiers to the proxy transmission module and display the returned file list and file data; The proxy transmission risk control module is used to collect access log data, construct a heterogeneous graph structure of path behavior, generate a path risk scoring model based on the improved MetaPath2Vec model, and calculate the path risk score. The proxy transfer execution module is used to generate session keys and path tokens for allowed requests, and to proxy access to the real path to complete file transfer, record transfer audit data and update the risk model; The Transmission Administrator module is used to review medium-risk access requests submitted by the proxy transmission execution module and return the approval results.

[0009] Optionally, modules can be integrated using the following methods: Configure the storage server access parameters and generate the basic storage server access configuration using dual-role accounts; Based on the storage server access basic configuration, obtain the real path set, perform path encryption mapping on the real path set, and generate a path mapping table; Collect file access behavior data and generate access log records; Construct a heterogeneous graph structure of path behavior based on access log data; An improved MetaPath2Vec model is constructed, and constrained meta-path sampling is performed on the heterogeneous graph structure of path behavior to obtain the positive sample path sequence set and the negative sample path sequence set, respectively. These are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. When the client initiates a file access request, a path sequence to be evaluated is constructed, the path risk scoring model is called to obtain the path risk score, and access control and path encryption proxy transmission are executed to generate transmission audit data. The path behavior heterogeneous graph structure and path policy automaton are updated based on access log data and transmission audit data, and the path risk scoring model is retrained or incrementally updated.

[0010] Optionally, the storage server access basic configuration includes storage server access parameters, encrypted credential data, and access control rules.

[0011] Optionally, the step of obtaining the real path set based on the storage server access basic configuration, performing path encryption mapping on the real path set, and generating a path mapping table specifically includes: Based on the secure transmission server and storage server access basic configuration, generate storage server connection credentials; Based on the storage server connection credentials, the secure transmission server initiates a session establishment request to the storage server, forming a set of real paths; For each real path in the set of real paths, a random salt value is selected based on the storage server access base configuration; The real path and the random salt value are concatenated and then processed by a key-based message authentication algorithm to obtain the encrypted path identifier. The secure transmission server reads the file attribute information and sensitivity level label returned by the storage server, and combines the real path, the corresponding encrypted path identifier, the file attribute information and the sensitivity level label into a path mapping item. All path mapping items are then merged to generate a path mapping table.

[0012] Optionally, the process of collecting file access behavior data and generating access log records specifically includes: The secure transmission server receives file access requests initiated by clients and reads the original record data of this access behavior from the access requests. The data in the original record data are concatenated sequentially according to a fixed field arrangement order to form an access behavior record vector; After the access processing is completed, the access result of the access request is obtained and appended to the end of the access behavior record vector to form an extended access behavior record vector; The extended access behavior record vector is formatted and written into the access log database according to the structure of the access log record data to generate access log record data.

[0013] Optionally, the construction of the heterogeneous path behavior graph structure based on access log data specifically includes: Parse the format of each field in the access log data and generate the field parsing results; Based on the field parsing results, corresponding node type tags are established, and an initial heterogeneous graph node set is created according to the field values; Perform deduplication on the initial heterogeneous graph node set and assign a unique node index value to each node to obtain the heterogeneous graph node set; Based on the relationships between fields in the access log data, each relationship is represented as an edge pair, generating an initial edge set; The initial edge set is deduplicated, and each edge is labeled with its corresponding edge type to obtain the heterogeneous graph edge set. Based on the set of nodes and edges of the heterogeneous graph, a heterogeneous graph structure for path behavior is generated.

[0014] Optionally, the construction of the improved MetaPath2Vec model involves performing constrained meta-path sampling on the heterogeneous graph structure of path behavior to obtain sets of positive and negative sample path sequences, which are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. Specifically, this includes: Based on the heterogeneous graph structure of path behavior, read the set of nodes and edges of the heterogeneous graph, and set the graph embedding training parameters in the secure transmission server. Initialize the node embedding matrix and the improved MetaPath2Vec model structure to obtain the improved MetaPath2Vec basic model; A finite state automaton is constructed based on file transfer security policies and dual-role control rules. The path policy automaton includes a set of states, a set of input symbols, a state transition function, an initial state, and a set of accepting states. Applying the path policy automaton to a heterogeneous path behavior graph structure generates a set of positive sample path sequences. Perform node replacement, node insertion, or node deletion operations on each positive sample path sequence in the positive sample path sequence set to obtain the perturbed path sequence; The input symbol sequence corresponding to the perturbated path sequence is input into the path policy automaton for state transition judgment, forming a set of negative sample path sequences; The nodes in each path sequence in the positive sample path sequence set and the negative sample path sequence set are converted into corresponding embedding vectors in turn, and the average operation is performed on all embedding vectors in the same path sequence to obtain the path sequence representation vector. Each path sequence in the set of positive sample path sequences is labeled as a positive class label, and each path sequence in the set of negative sample path sequences is labeled as a negative class label. Based on the path sequence representation vector, a sequence-level loss function is constructed, and the sequence-level loss value is calculated. Based on the co-occurrence relationship of nodes in the positive sample path sequence set and the negative sample path sequence set, a positive sample node pair set and a negative sample node pair set are generated. Each positive sample node pair and each negative sample node pair are converted into a corresponding node embedding vector combination, a node-level loss function is constructed, and the node-level loss value is calculated. The sequence-level loss function and the node-level loss function are weighted to obtain the total loss function. The node embedding matrix and the parameters of the sequence discrimination layer are iteratively updated until the total loss function meets the convergence condition. The trained improved MetaPath2Vec model is then used as the path risk scoring model.

[0015] Optionally, when the client initiates a file access request, constructing a path sequence to be evaluated, calling a path risk scoring model to obtain a path risk score, and performing access control and path encryption proxy transmission to generate transmission audit data specifically includes: When a client initiates a file access request based on an encrypted path identifier, a path sequence to be evaluated is formed based on the data recorded in the access log. Each node in the path sequence to be evaluated is converted into a corresponding node embedding vector in turn, and the embedding vectors of all nodes in the path sequence are averaged to generate the path sequence representation vector of the current request. Input the path sequence representation vector into the sequence discrimination layer of the path risk scoring model to obtain the path risk score for the current request; The path risk score is compared with a preset risk threshold to determine the risk level of the current request; When the risk level is low, execution is authorized directly, and the corresponding file access request is written to the authorization queue. When the risk level is medium, a review task is generated and assigned to a review role account. The review role account makes an approval decision and executes the task when the approval is granted. When the risk level is high, the file access request is rejected. In each allowed file access request, the order of node access by the client during the actual transmission process is recorded, and the node access order is combined to form a path identifier; Input the path identifier, user identifier, role identifier, and time parameter into the key derivation algorithm to generate the corresponding session key; A path token is generated based on the session key. The path token is then attached to the access request sent by the client in the transport session. The path token is then checked for consistency. If the check is inconsistent, the access request is terminated. When the path token verification is successful, the secure transmission server uses the session key to access the real path corresponding to the encrypted path identifier in the storage server and proxies the file data transmission to form transmission audit data.

[0016] Optionally, the step of updating the heterogeneous graph structure of path behavior and the path policy automaton based on access log data and transmission audit data, and retraining or incrementally updating the path risk scoring model specifically includes: New access record entries are generated based on newly added access log data, and new audit record entries are generated based on newly added transmission audit data. Extract node fields and node relationship fields from newly added access record entries and newly added audit record entries, and update them to form a new heterogeneous path behavior graph structure; Based on the updated heterogeneous graph structure of path behavior, the path policy automaton is synchronously adjusted to generate an updated path policy automaton. Based on the updated heterogeneous graph structure of path behavior and the updated path policy automaton, the meta-path sampling process constrained by the automaton is re-executed to obtain a new set of positive sample path sequences and a new set of negative sample path sequences. Based on the new set of positive sample path sequences and the new set of negative sample path sequences, each path sequence is converted into a node embedding vector and a path sequence representation vector is generated. Based on the total loss function, the node embedding matrix and sequence discrimination layer parameters are retrained or incrementally updated until the total loss value meets the preset convergence condition, thus obtaining the updated path risk scoring model.

[0017] The beneficial effects of this invention are: This invention constructs a path-based encryption-based access mechanism, ensuring that clients only encounter encrypted path identifiers during file browsing and access, without exposing the actual directory structure. This reduces the risk of sensitive information leakage based on path inference. The resolution and access to the actual path are entirely handled by a secure transmission server, keeping the directory structure and file organization invisible during transmission, thus enhancing the security of the server-side directory.

[0018] This invention employs a dual-role account management approach, where the configuration of access credentials and the entry of keys are completed by different roles, creating a controllable hierarchical structure for the basic configuration of storage server access. In the access control phase, a path risk scoring model is incorporated to achieve dynamic access decisions, providing differentiated processing methods for requests with different user roles, operation types, and sensitivity levels. This allows the security of the access process to adaptively adjust as request behavior changes.

[0019] This invention constructs a heterogeneous path behavior graph structure and introduces an improved MetaPath2Vec model and path policy automaton, enabling effective modeling of the relationships between access behaviors. The system continuously updates based on access log data and transmission audit data, giving the risk assessment model incremental learning capabilities. This allows it to adapt to changes in access patterns over long-term operation, thereby improving the system's ability to identify abnormal access behavior and the accuracy of access link control. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a secure file transfer system based on path encryption and dual-role account management proposed in this invention; Figure 2 This is a flowchart of a secure file transfer system based on path encryption and dual-role account management proposed in this invention. Figure 3 This is a schematic diagram of the training process for an improved MetaPath2Vec model in a secure file transfer system based on path encryption and dual-role account management, as proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A secure file transfer system based on path encryption and dual-role account management includes: The system administrator module is used to generate storage server access parameters; The confidentiality administrator module is used to enter access passwords or keys and store them encrypted, forming encrypted credential data and access control rules, which are combined with storage server access parameters to generate the basic configuration for storage server access. The storage server module is used to provide read and write services for the actual directory structure and file data based on the storage server access base configuration; The encrypted path component is used to obtain the real path set based on the storage server access basic configuration and build a path mapping table; The client module is used to initiate file access requests based on encrypted path identifiers to the proxy transmission module and display the returned file list and file data; The proxy transmission risk control module is used to collect access log data, construct a heterogeneous graph structure of path behavior, generate a path risk scoring model based on the improved MetaPath2Vec model, and calculate the path risk score. The proxy transfer execution module is used to generate session keys and path tokens for allowed requests, and to proxy access to the real path to complete file transfer, record transfer audit data and update the risk model; The Transmission Administrator module is used to review medium-risk access requests submitted by the proxy transmission execution module and return the approval results.

[0023] In this embodiment, the modules are interconnected using the following method: Configure the storage server access parameters and generate the basic storage server access configuration using dual-role accounts; Based on the storage server access basic configuration, obtain the real path set, perform path encryption mapping on the real path set, and generate a path mapping table; Collect file access behavior data and generate access log records; Construct a heterogeneous graph structure of path behavior based on access log data; An improved MetaPath2Vec model is constructed, and constrained meta-path sampling is performed on the heterogeneous graph structure of path behavior to obtain the positive sample path sequence set and the negative sample path sequence set, respectively. These are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. When the client initiates a file access request, a path sequence to be evaluated is constructed, the path risk scoring model is called to obtain the path risk score, and access control and path encryption proxy transmission are executed to generate transmission audit data. The path behavior heterogeneous graph structure and path policy automaton are updated based on access log data and transmission audit data, and the path risk scoring model is retrained or incrementally updated.

[0024] In this embodiment, configuring the storage server access parameters and generating the storage server access basic configuration through dual-role accounts specifically includes: In the account management module of the secure transmission server, create system administrator account, confidentiality administrator account, review role account and audit administrator account respectively, and assign non-overlapping permission sets to each account; The system administrator logs into the secure transmission server, enters the network address of the storage server, access protocol type, access username, and allowed root directory range, and establishes a correspondence between the client identifier and the root directory range to generate storage server access parameters; Log in to the secure transmission server with the confidentiality administrator account, enter the password or key corresponding to the access username, write the password or key into the credential storage area of ​​the secure transmission server, perform encryption storage processing on the password or key, and generate encrypted credential data. In the network access control module of the secure transmission server, access control rules are configured based on the storage server access parameters and encrypted credential data. The access control rules restrict clients from directly accessing the storage server and allow the secure transmission server to access the storage server. The access control rules include client access blocking rules that prohibit clients from directly accessing the storage service server, proxy access allowing rules that allow secure transmission servers to access the storage service server, access source verification rules that verify the access source, session link restriction rules that limit the access link, access range constraint rules that limit the access directory range, and unauthorized behavior blocking rules for blocking unauthorized access behavior. The storage server access parameters, encrypted credential data, and access control rules are combined and stored to form the basic configuration for storage server access.

[0025] In this embodiment, the step of obtaining the real path set based on the storage server access basic configuration, performing path encryption mapping on the real path set, and generating a path mapping table specifically includes: The secure transmission server reads the storage server's network address, access protocol type, access username, and encrypted credential data from the storage server's access base configuration, and generates storage server connection credentials for establishing an access session. Based on the storage server connection credentials, the secure transmission server initiates a session establishment request to the storage server. After the session is successfully established, a directory traversal operation is performed on the preset root directory range to obtain the directory nodes, file nodes and their hierarchical relationships in sequence, forming a set of real paths, where each real path is represented by a string sequence. For each real path in the set of real paths, select the path encryption key and the corresponding random salt value associated with the key parameters in the storage server access basic configuration. Concatenate the real path and the random salt value in order and input them into the key-based message authentication algorithm for calculation to obtain the encrypted path identifier that corresponds one-to-one with the real path. The encrypted path identifier is then used as the unique identifier of the real path in the path mapping table. The random salt value is generated in real time by the secure transmission server based on the built-in random number generation module when generating the encrypted path identifier. The random number generation module outputs a one-time random bit sequence according to a preset bit length and uses the random bit sequence as the random salt value corresponding to the real path. The message authentication algorithm includes a key-related hash calculation process that takes the path encryption key as the key input and the concatenation result of the real path and random salt value as the message input, and performs the calculation process according to a fixed compression function and iterative structure. The calculation process generates a fixed-length authentication code based on a preset hash function family. The authentication code serves as an encrypted path identifier that corresponds one-to-one with the real path. The secure transmission server reads the file attribute information and sensitivity level label returned by the storage server, and combines the real path, the corresponding encrypted path identifier, the file attribute information and the sensitivity level label into a path mapping item according to a one-to-one correspondence. All path mapping items are then merged to generate a path mapping table. The path mapping table is stored in the mapping index area of ​​the secure transmission server. When a client requests file display information, the encrypted path identifier and corresponding display field are retrieved from the path mapping table, and a file display result containing only the encrypted path identifier is generated and returned to the client.

[0026] In this embodiment, the process of collecting file access behavior data and generating access log records specifically includes: The secure transmission server receives file access requests initiated by clients, and sequentially reads user identifier, user role, operation type, target encrypted path identifier, target sensitivity level, access time, time window, and client identifier from the access request, and uses the above fields as the original record data for this access behavior; The user identifier, user role, operation type, target encrypted path identifier, target sensitivity level, access time, time window, and client identifier are concatenated in a fixed field order to form an access behavior record vector. After the access processing is completed, the access result of the access request is obtained, and the access result is appended to the end of the access behavior record vector to form an extended access behavior record vector; The extended access behavior record vector is formatted and written into the access log database according to the structure of the access log record data to generate access log record data.

[0027] In this embodiment, the construction of the heterogeneous path behavior graph structure based on access log record data specifically includes: Read user identifier, user role, operation type, target encrypted path identifier, target sensitivity level, access time window, and client identifier one by one from the access log data, parse the format of each field and generate the field parsing results; Based on the field parsing results, corresponding node type tags are established for user identifier, user role, operation type, target encrypted path identifier, target sensitivity level, access time window and client identifier, respectively. User nodes, role nodes, operation nodes, encrypted path nodes, sensitivity level nodes, time window nodes and intermediate nodes are created according to the field values ​​to form an initial heterogeneous graph node set. Perform deduplication on the initial heterogeneous graph node set and assign a unique node index value to each node to obtain the heterogeneous graph node set; Based on the relationships between fields in the access log data, user-role association, role-operation association, operation-encrypted path association, encrypted path-sensitivity level association, operation-time window association, and operation-intermediate node association are established respectively. Each association is represented as an edge pair, and an initial edge set is generated. The initial edge set is deduplicated, and each edge is labeled with its corresponding edge type to obtain the heterogeneous graph edge set. Based on the set of nodes and edges of the heterogeneous graph, a heterogeneous graph structure of path behavior is generated and stored in the graph storage area.

[0028] In this embodiment, the construction of the improved MetaPath2Vec model involves performing constrained meta-path sampling on the heterogeneous graph structure of path behavior to obtain sets of positive and negative sample path sequences, which are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. Specifically, this includes: Based on the heterogeneous graph structure of path behavior, the set of nodes and edges of the heterogeneous graph are read. Graph embedding training parameters are set in the secure transmission server. The graph embedding training parameters include embedding dimension, random walk step size, number of random walks, and context window size. The node embedding matrix and the improved MetaPath2Vec model structure are initialized to obtain the improved MetaPath2Vec basic model. The node embedding matrix is ​​initialized as a two-dimensional matrix composed of the number of nodes in the heterogeneous graph and the embedding dimension, where the number of rows corresponds to the number of nodes in the heterogeneous graph node set, and the number of columns corresponds to the embedding dimension. Construct a finite state automaton to represent legitimate transmission path sequences based on file transfer security policies and dual-role control rules; The file transfer security policy includes restrictions on the order of allowed node types, node combination methods, and optional jump methods in the transfer path; The dual-role control rules include restrictions on the role conditions, approval conditions, and permission conditions required for access operations involving sensitive nodes or sensitive paths. Based on the file transfer security policy, the set of states and the set of input symbols used to describe the path traversal process are determined. Based on the dual-role control rule, the state transition function is determined. The initial state and the set of accepting states are specified in a fixed way. The set of states, the set of input symbols, the state transition function, the initial state and the set of accepting states are combined to form a path policy automaton. The state transition function is used to map the current state and the input symbols to the next state according to the rule-defined correspondence. The path strategy automaton is applied to the heterogeneous graph structure of path behavior. The starting node is selected from the set of nodes in the heterogeneous graph. The adjacent nodes are randomly walked according to the preset meta-path type. At each step of the walk, the input symbol is determined according to the current state and the candidate node type and the automaton state is updated through the state transition function. Only when the node sequence starts from the initial state and finally reaches the set of accepting states is the corresponding node sequence recorded as a positive sample path sequence. The positive sample path sequence set is accumulated. Perform node replacement, node insertion, or node deletion operations on each positive sample path sequence in the positive sample path sequence set to obtain the perturbed path sequence. Then, input the input symbol sequence corresponding to the perturbed path sequence into the path policy automaton for state transition judgment. If the set of accepting states has not been reached, the perturbed path sequence is recorded as a negative sample path sequence, and the perturbed path sequence set is accumulated. For each path sequence in the set of positive sample path sequences and the set of negative sample path sequences, read the nodes in the path sequence one by one in node order, convert each node into a corresponding embedding vector in turn, and perform an average operation on all embedding vectors in the same path sequence to obtain the path sequence representation vector. Each path sequence in the set of positive sample path sequences is labeled as a positive class label, and each path sequence in the set of negative sample path sequences is labeled as a negative class label. Based on the path sequence representation vector, a sequence-level loss function is constructed, and the sequence-level loss value is calculated. ; in, For sequence-level loss function, The index is used to represent any path sequence in the set of positive sample path sequences and the set of negative sample path sequences. These are path sequence labels, including positive labels (value 1) and negative labels (value 0). For the first A path sequence representation vector of a path sequence. For the Sigmoid function, The weight parameter vector of the sequence discrimination layer is used to perform a linear transformation on the path sequence representation vector. These are the bias parameters for the sequence discrimination layer; Based on the co-occurrence relationships of nodes in the positive sample path sequence set and the negative sample path sequence set, a positive sample node pair set and a negative sample node pair set are generated. Each positive sample node pair and each negative sample node pair are then converted into a corresponding node embedding vector combination. A node-level loss function is constructed, and the node-level loss value is calculated. ; in, For node-level loss function, For the set of positive sample node pairs, For the set of negative sample node pairs, This is a node pair in the set of positive sample node pairs. A node pair in the set of negative sample node pairs. For the target node, For nodes in a heterogeneous graph, Negative sample nodes that should not appear simultaneously with the target node. , , Representing nodes respectively ,node and nodes The node embedding vector, For nodes Node embedding vectors and nodes The result of the inner product operation between the node embedding vectors; The sequence-level loss function and the node-level loss function are weighted to obtain the total loss function. The node embedding matrix and the sequence discrimination layer parameters, including the weight parameter vector and the bias parameter of the sequence discrimination layer, are iteratively updated using the stochastic gradient descent algorithm or the mini-batch gradient descent algorithm until the total loss function meets the convergence condition. The improved MetaPath2Vec model trained at this time is used as the path risk scoring model. The convergence conditions include any one or more of the following: the change in the total loss value over several consecutive iterations is less than a preset change threshold, the total loss value is lower than a preset loss threshold, and the number of iterations reaches a preset maximum number of iterations.

[0029] In this embodiment, the steps of constructing a path sequence to be evaluated, calling a path risk scoring model to obtain a path risk score, and performing access control and path encryption proxy transmission to generate transmission audit data when the client initiates a file access request specifically include: When the client initiates a file access request based on the encrypted path identifier, the user identifier, user role, operation type, target encrypted path identifier, target sensitivity level, access time window and client identifier corresponding to the access request are extracted from the access log data, and the fields are connected in a preset order to form a path sequence to be evaluated. Each node in the path sequence to be evaluated is converted into a corresponding node embedding vector in turn, and the embedding vectors of all nodes in the path sequence are averaged to generate the path sequence representation vector of the current request. The path sequence representation vector is input into the sequence discrimination layer of the path risk scoring model. The sequence discrimination layer performs linear calculation on the path sequence representation vector according to the internal weight parameters and bias parameters, and performs nonlinear transformation on the linear calculation result to obtain the path risk score of the current request. The path risk score is compared with a preset risk threshold to determine the risk level of the current request; The risk levels include low risk, medium risk and high risk, where the path risk score of the low risk level is lower than the first risk threshold, the path risk score of the medium risk level is between the first risk threshold and the second risk threshold, and the path risk score of the high risk level is higher than the second risk threshold. When the risk level is low, execution is authorized directly, and the corresponding file access request is written to the authorization queue. When the risk level is medium risk, a review task is generated and assigned to a review role account. The review role account makes the approval decision and executes the task when the approval is approved. The file access request will be rejected when the risk level is high. In each allowed file access request, the node access order of the client during the actual transmission process is recorded. The node access order is combined to form a path identifier. The path identifier, user identifier, role label and time parameter are input into the key derivation algorithm, which generates the corresponding session key. A path token is generated based on the session key. The path token is attached to the access request sent by the client in the transmission session. Each intermediate node performs a consistency check on the path token. If the check is consistent, the access request is allowed to be passed to the secure transmission server. If the check is inconsistent, the access request is terminated. If the path token verification is successful, the secure transmission server uses the session key to access the real path corresponding to the encrypted path identifier in the storage server, and proxies the completion of file data transmission. After the file transmission is completed, the transmission event information is recorded to form transmission audit data.

[0030] In this embodiment, the step of updating the heterogeneous graph structure of path behavior and the path policy automaton based on access log recording data and transmission audit data, and retraining or incrementally updating the path risk scoring model specifically includes: Based on the newly added access log data, extract user identifier, user role, operation type, encrypted path identifier, sensitivity level, access time window and client identifier, write the fields into the access log database and generate new access record entries; Based on the newly added transmission audit data, the node access order, transmission time parameters, session key derivation parameters and path token verification results during the file transmission process are read, the transmission audit data is written into the audit database and a new audit record entry is generated. Extract node fields and node relationship fields from newly added access record entries and newly added audit record entries, add node fields to the original heterogeneous graph node set, add node relationship fields to the original heterogeneous graph edge set, and update to form a new path behavior heterogeneous graph structure; Based on the updated heterogeneous graph structure of path behavior, the state set, input symbol set and state transition relationship describing the legal path sequence in the path policy automaton are synchronously adjusted, and the updated path policy automaton is generated. Based on the updated heterogeneous graph structure of path behavior and the updated path policy automaton, the meta-path sampling process constrained by the automaton is re-executed to obtain a new set of positive sample path sequences and a new set of negative sample path sequences. Based on the new set of positive sample path sequences and the new set of negative sample path sequences, each path sequence is converted into a node embedding vector and a path sequence representation vector is generated. Based on the total loss function, the node embedding matrix and sequence discrimination layer parameters are retrained or incrementally updated until the total loss value meets the preset convergence condition, thus obtaining the updated path risk scoring model.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a confidential document transmission scenario within a large research institution. This institution has multiple research departments that routinely need to transmit experimental data, design documents, and highly sensitive files over the internal network. In traditional systems, clients can directly see the actual directory path, access control relies on static permission configurations, and it is impossible to identify unauthorized access routes or dynamically adjust security policies based on changes in access behavior. This has led to long-term issues such as the risk of real path leakage and significant risks associated with centralized management of access credentials.

[0032] In this embodiment, the system consists of a client module, a proxy transmission server, a storage server, and a dual-role management terminal. First, the system administrator configures the storage server's network address, access protocol, root directory range, and access username. Then, the security administrator enters the corresponding access key and encrypts it, thus forming the basic configuration for storage server access. During the file browsing phase, the client only receives the encrypted path identifier generated by the encrypted path component and cannot directly see the actual directory structure. When researchers select an encrypted path for access, the proxy transmission server constructs a path sequence to be evaluated based on access log data, calculates a path risk score using the improved MetaPath2Vec model, and then decides whether to allow the access request based on the score result.

[0033] During the several-week trial operation, the system was configured with approximately 350 real path nodes and over 100 valid user accounts, generating about 20,000 access log records daily. After constructing a heterogeneous path behavior graph, the system, through continuous training, can identify structural changes and potential risk characteristics in access links. During the trial operation, the system identified over 30 abnormal access combinations by ordinary users, most of which were judged as medium risk and blocked after review, preventing unauthorized access to highly sensitive files. Compared with the original access control method that relies on static rules, this invention improves the abnormal access identification rate by approximately 25% and reduces the number of unauthorized access triggers by approximately 30%, indicating that this invention has higher effectiveness in identifying access behavior risks.

[0034] Regarding transmission security, thanks to the path encryption mechanism, no directory probing behavior based on path inference occurred on the client side during the trial operation. Furthermore, through a node-by-node path token verification mechanism, this invention achieved a 100% forgery link rejection rate in simulated attacks. All intercepted and forged access requests were rejected by the system due to inconsistent path tokens, effectively improving the security of the transmission link.

[0035] Regarding system load, the path risk scoring model of this invention employs an incremental training mechanism, which can maintain low inference latency even with continuously increasing access logs. During the trial operation, the average calculation time for risk scores was approximately 40–60 ms, which did not significantly impact file access latency. During high-concurrency access phases, the system's file transfer success rate remained around 97%, demonstrating a significant improvement in stability compared to the original system, thus verifying the engineering feasibility of this invention in terms of real-time performance and stability.

[0036] Table 1 Performance Verification Data of Path Encryption and Risk Scoring Mechanism

[0037] As can be seen from the table above, this invention significantly improves upon existing file transfer systems in terms of path hiding, access risk identification, transmission security, and overall system stability. Firstly, regarding access behavior coverage, this invention's system, by constructing a heterogeneous path behavior graph structure and continuously collecting access logs, increases the number of valid logs per day from 12,000 to 20,000, representing a data coverage improvement of approximately 66%. This higher data coverage provides a more complete foundation for subsequent risk modeling and behavior analysis, enabling the model to more accurately capture changing trends in access patterns.

[0038] Secondly, regarding the ability to identify abnormal access, the path risk scoring model constructed in this invention combines a heterogeneous graph structure with the improved MetaPath2Vec embedding learning method, increasing the abnormal access identification rate from 56% in the original system to 81%, an improvement of 25 percentage points. Simultaneously, in terms of unauthorized access blocking capabilities, the system of this invention can make real-time judgments based on risk scores when an access request is triggered, increasing the number of blocking attempts from 14 to 19, an improvement of 35%. These data demonstrate that this invention significantly outperforms traditional solutions that rely solely on static permission checks in identifying unauthorized access links and discovering abnormal access combinations.

[0039] Regarding path security, this invention employs a path encryption mechanism to completely isolate the real path from the client, reducing the path exposure rate from 100% to 0% and completely eliminating the risk of inference attacks based on the path structure. Furthermore, through node-by-node verification of session keys and path tokens, this invention achieved a 100% rejection rate in simulated forged link tests, demonstrating that the system can effectively block access requests that are tampered with or forged mid-transmission, improving the integrity and reliability of the transmission link.

[0040] Finally, regarding system stability and efficiency, the average time for risk scoring inference in this invention remains stable within the range of 40–60 ms, meeting the real-time requirements in file transfer scenarios. Under high-concurrency access conditions, the file transfer success rate of this invention's system reaches 97%, a significant improvement over the original system's 92%, indicating that the system can maintain stable operation even when handling a large number of transfer tasks. In summary, this invention demonstrates superior engineering application value in terms of refined access control, path concealment capabilities, secure link verification, and model adaptive capabilities.

[0041] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A secure file transfer system based on path encryption and dual-role account management, characterized in that, include: The system administrator module is used to generate storage server access parameters; The confidentiality administrator module is used to enter access passwords or keys and store them encrypted, forming encrypted credential data and access control rules, which are combined with storage server access parameters to generate the basic configuration for storage server access. The storage server module is used to provide read and write services for the actual directory structure and file data based on the storage server access base configuration; The encrypted path component is used to obtain the real path set based on the storage server access basic configuration and build a path mapping table; The client module is used to initiate file access requests based on encrypted path identifiers to the proxy transmission module and display the returned file list and file data; The proxy transmission risk control module is used to collect access log data, construct a heterogeneous graph structure of path behavior, generate a path risk scoring model based on the improved MetaPath2Vec model, and calculate the path risk score. The proxy transfer execution module is used to generate session keys and path tokens for allowed requests, and to proxy access to the real path to complete file transfer, record transfer audit data and update the risk model; The Transmission Administrator module is used to review medium-risk access requests submitted by the proxy transmission execution module and return the approval results.

2. The secure file transfer system based on path encryption and dual-role account management according to claim 1, characterized in that, The modules are connected in the following way: Configure the storage server access parameters and generate the basic storage server access configuration using dual-role accounts; Based on the storage server access basic configuration, obtain the real path set, perform path encryption mapping on the real path set, and generate a path mapping table; Collect file access behavior data and generate access log records; Construct a heterogeneous graph structure of path behavior based on access log data; An improved MetaPath2Vec model is constructed, and constrained meta-path sampling is performed on the heterogeneous graph structure of path behavior to obtain the positive sample path sequence set and the negative sample path sequence set, respectively. These are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. When the client initiates a file access request, a path sequence to be evaluated is constructed, the path risk scoring model is called to obtain the path risk score, and access control and path encryption proxy transmission are executed to generate transmission audit data. The path behavior heterogeneous graph structure and path policy automaton are updated based on access log data and transmission audit data, and the path risk scoring model is retrained or incrementally updated.

3. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The basic configuration for accessing the storage server includes storage server access parameters, encrypted credential data, and access control rules.

4. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The process of obtaining the real path set based on the storage server access configuration, performing path encryption mapping on the real path set, and generating a path mapping table specifically includes: Based on the secure transmission server and storage server access basic configuration, generate storage server connection credentials; Based on the storage server connection credentials, the secure transmission server initiates a session establishment request to the storage server, forming a set of real paths; For each real path in the set of real paths, a random salt value is selected based on the storage server access base configuration; The real path and the random salt value are concatenated and then processed by a key-based message authentication algorithm to obtain the encrypted path identifier. The secure transmission server reads the file attribute information and sensitivity level label returned by the storage server, and combines the real path, the corresponding encrypted path identifier, the file attribute information and the sensitivity level label into a path mapping item. All path mapping items are then merged to generate a path mapping table.

5. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The specific data collection and generation of access log records for file access behavior includes: The secure transmission server receives file access requests initiated by clients and reads the original record data of this access behavior from the access requests. The data in the original record data are concatenated sequentially according to a fixed field arrangement order to form an access behavior record vector; After the access processing is completed, the access result of the access request is obtained and appended to the end of the access behavior record vector to form an extended access behavior record vector; The extended access behavior record vector is formatted and written into the access log database according to the structure of the access log record data to generate access log record data.

6. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The construction of the heterogeneous path behavior graph structure based on access log data specifically includes: Parse the format of each field in the access log data and generate the field parsing results; Based on the field parsing results, corresponding node type tags are established, and an initial heterogeneous graph node set is created according to the field values; Perform deduplication on the initial heterogeneous graph node set and assign a unique node index value to each node to obtain the heterogeneous graph node set; Based on the relationships between fields in the access log data, each relationship is represented as an edge pair, generating an initial edge set; The initial edge set is deduplicated, and each edge is labeled with its corresponding edge type to obtain the heterogeneous graph edge set. Based on the set of nodes and edges of the heterogeneous graph, a heterogeneous graph structure for path behavior is generated.

7. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The construction of the improved MetaPath2Vec model involves performing constrained meta-path sampling on the heterogeneous graph structure of path behavior to obtain sets of positive and negative sample path sequences, which are then input into the improved MetaPath2Vec model for training to obtain the path risk scoring model. Specifically, this includes: Based on the heterogeneous graph structure of path behavior, read the set of nodes and edges of the heterogeneous graph, and set the graph embedding training parameters in the secure transmission server. Initialize the node embedding matrix and the improved MetaPath2Vec model structure to obtain the improved MetaPath2Vec basic model; A finite state automaton is constructed based on file transfer security policies and dual-role control rules. The path policy automaton includes a set of states, a set of input symbols, a state transition function, an initial state, and a set of accepting states. Applying the path policy automaton to a heterogeneous path behavior graph structure generates a set of positive sample path sequences. Perform node replacement, node insertion, or node deletion operations on each positive sample path sequence in the positive sample path sequence set to obtain the perturbed path sequence; The input symbol sequence corresponding to the perturbated path sequence is input into the path policy automaton for state transition judgment, forming a set of negative sample path sequences; The nodes in each path sequence in the positive sample path sequence set and the negative sample path sequence set are converted into corresponding embedding vectors in turn, and the average operation is performed on all embedding vectors in the same path sequence to obtain the path sequence representation vector. Each path sequence in the set of positive sample path sequences is labeled as a positive class label, and each path sequence in the set of negative sample path sequences is labeled as a negative class label. Based on the path sequence representation vector, a sequence-level loss function is constructed, and the sequence-level loss value is calculated. Based on the co-occurrence relationship of nodes in the positive sample path sequence set and the negative sample path sequence set, a positive sample node pair set and a negative sample node pair set are generated. Each positive sample node pair and each negative sample node pair are converted into a corresponding node embedding vector combination, a node-level loss function is constructed, and the node-level loss value is calculated. The sequence-level loss function and the node-level loss function are weighted to obtain the total loss function. The node embedding matrix and the parameters of the sequence discrimination layer are iteratively updated until the total loss function meets the convergence condition. The trained improved MetaPath2Vec model is then used as the path risk scoring model.

8. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, When the client initiates a file access request, the process of constructing a path sequence to be evaluated, calling a path risk scoring model to obtain a path risk score, and executing access control and path encryption proxy transmission to generate transmission audit data specifically includes: When a client initiates a file access request based on an encrypted path identifier, a path sequence to be evaluated is formed based on the data recorded in the access log. Each node in the path sequence to be evaluated is converted into a corresponding node embedding vector in turn, and the embedding vectors of all nodes in the path sequence are averaged to generate the path sequence representation vector of the current request. Input the path sequence representation vector into the sequence discrimination layer of the path risk scoring model to obtain the path risk score for the current request; The path risk score is compared with a preset risk threshold to determine the risk level of the current request; When the risk level is low, execution is authorized directly, and the corresponding file access request is written to the authorization queue. When the risk level is medium, a review task is generated and assigned to a review role account. The review role account makes an approval decision and executes the task when the approval is granted. When the risk level is high, the file access request is rejected. In each allowed file access request, the order of node access by the client during the actual transmission process is recorded, and the node access order is combined to form a path identifier; Input the path identifier, user identifier, role identifier, and time parameter into the key derivation algorithm to generate the corresponding session key; A path token is generated based on the session key. The path token is then attached to the access request sent by the client in the transport session. The path token is then checked for consistency. If the check is inconsistent, the access request is terminated. When the path token verification is successful, the secure transmission server uses the session key to access the real path corresponding to the encrypted path identifier in the storage server and proxies the file data transmission to form transmission audit data.

9. A secure file transfer system based on path encryption and dual-role account management according to claim 2, characterized in that, The process of updating the heterogeneous graph structure of path behavior and the path policy automaton based on access log data and transmission audit data, and retraining or incrementally updating the path risk scoring model specifically includes: New access record entries are generated based on newly added access log data, and new audit record entries are generated based on newly added transmission audit data. Extract node fields and node relationship fields from newly added access record entries and newly added audit record entries, and update them to form a new heterogeneous path behavior graph structure; Based on the updated heterogeneous graph structure of path behavior, the path policy automaton is synchronously adjusted to generate an updated path policy automaton. Based on the updated heterogeneous graph structure of path behavior and the updated path policy automaton, the meta-path sampling process constrained by the automaton is re-executed to obtain a new set of positive sample path sequences and a new set of negative sample path sequences. Based on the new set of positive sample path sequences and the new set of negative sample path sequences, each path sequence is converted into a node embedding vector and a path sequence representation vector is generated. Based on the total loss function, the node embedding matrix and sequence discrimination layer parameters are retrained or incrementally updated until the total loss value meets the preset convergence condition, thus obtaining the updated path risk scoring model.

Images