Dynamic permission adaptation navigation method and system based on natural language processing

Abstract

The invention relates to the technical field of human-computer interaction, in particular to a dynamic permission adaptation navigation method based on natural language processing. The method comprises the following steps: acquiring a demand instruction text of a user, converting a word sequence corresponding to the demand instruction text into a continuous vector sequence, analyzing a dependency relationship between the continuous vector sequence and a preset permission topic word, and generating a word vector representation of the demand instruction text based on the dependency relationship; inputting the word vector representation into the BERT-TextCNN hybrid model, and outputting a demand category corresponding to the demand instruction text of the user; activating a dynamic permission engine based on the demand category to call a permission matrix, generating a weight optimal path in combination with a DAG algorithm, and determining an access path corresponding to the demand instruction of the user based on the weight optimal path; and adaptive navigation is performed on the user based on a multi-modal interaction mechanism and the access path, so that the authority distribution efficiency and the navigation interaction efficiency can be improved.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction technology, and in particular to a dynamic permission adaptation navigation method based on natural language processing. Background Art

[0002] As digital systems become increasingly complex (such as enterprise applications, e-commerce platforms, and smart terminals), users need to complete operational tasks in multiple functional modules. However, traditional navigation methods have the following technical drawbacks: Limitations of static permissions and fixed navigation: Existing systems often use fixed permissions, such as role-based access control. User permissions are determined at registration and cannot be dynamically adjusted based on real-time needs. For example, ordinary users can only access basic functions, but if they temporarily need advanced permissions, they must submit an application, resulting in operational interruptions.

[0003] Rigid navigation paths: System menus and function portals are statically designed, requiring users to manually click through each layer, unable to jump directly to specific tasks. This results in low operational efficiency, especially in complex processes (such as multi-step approvals). The disconnect between natural language interaction and permission adaptation: Although natural language processing technologies (such as voice assistants and chatbots) have been applied to system interactions, most of them only support shallow command parsing and cannot understand the deep semantics of user intent.

[0004] Therefore, there is an urgent need for a dynamic permission adaptation navigation method based on natural language processing to solve the problems in the existing technology. Summary of the Invention

[0005] The purpose of the present invention is to provide a dynamic permission adaptation navigation method and system based on natural language processing: it aims to achieve personalized adaptation of system function navigation and dynamic allocation of permissions by integrating natural language understanding, user behavior analysis and permission management technology, thereby improving user experience and system security.

[0006] The purpose of the present invention can be achieved through the following technical solutions: On the one hand, a dynamic permission adaptation navigation method based on natural language processing includes: Obtain the user's demand instruction text, convert the word sequence corresponding to the demand instruction text into a continuous vector sequence, analyze the dependency relationship between the continuous vector sequence and the preset permission topic words, and generate a word vector representation of the demand instruction text based on the dependency relationship; Input the word vector representation into the BERT-TextCNN hybrid model to output the demand category corresponding to the user's demand instruction text; Based on the demand category, the dynamic permission engine is activated to retrieve the permission matrix, and the weighted optimal path is generated by combining the DAG algorithm. Based on the weighted optimal path, the access path corresponding to the user's demand instruction is determined; Adapt navigation for users based on multimodal interaction mechanisms and access paths.

[0007] Furthermore, converting the word sequence corresponding to the demand instruction text into a continuous vector sequence specifically includes the following process: The parts of speech of words in the demand instruction text are determined based on Stanford NLP Parser, and the text nouns representing the demand instructions are screened out based on the clustering algorithm. The text nouns are converted into continuous vector sequences through the embedding layer of the encoder.

[0008] Furthermore, the process of selecting nouns representing demand instructions based on the clustering algorithm specifically includes the following steps: Form a matrix X of all words in the demand instruction text, calculate the covariance matrix S of the sample of matrix X, and calculate the eigenvector of the covariance matrix S 、 ,..., 、 and eigenvalues t = 1, 2, ..., n; project the data into the space of eigenvectors, using the formula: ,in, The value is the value of the dimension corresponding to the word in the demand instruction text. Filter reference values for words and set reference thresholds ,Will Greater than The corresponding m words are selected and recorded as words to be screened; the words to be screened form a data set to be screened The K value is set based on the size and shape of the data set to be screened, and any word to be screened is randomly selected as the initial cluster centroid. When the number of initial centroids is less than K, is the initial center of mass; based on the objective function Calculate the distance between each word to be filtered in the dataset to be filtered and the existing initial cluster centroid ,Will The word to be filtered corresponding to the maximum value is used as the next initial centroid; K initial centroids are obtained in sequence, and the data set to be filtered is clustered based on the k initial centroids to obtain the noun representing the demand instruction.

[0009] Furthermore, parsing the dependency relationship between the continuous vector sequence and the preset authority topic words specifically includes the following process: Given a continuous vector sequence , Represents a continuous vector sequence Middle Morphemes, where morphemes are roots or affixes, ,in, Indicates the constituent morphemes No. syllables; A sequence of consecutive vectors Syllables represented by low-dimensional real vectors obtained through Word2Vec word embedding technology , Indicates the The first morpheme in A vector of syllables; calculate Corresponding to the preset permission topic words The sum of similarities dist: ; in, Indicates the first word in the preset authority topic The first morpheme in A vector of syllables, represents dot product, express The modulus value, express The modulus value of Determine whether the sum of similarities dist exceeds the preset similarity threshold. If so, the dependency relationship between the continuous vector sequence and the preset authority topic words is determined to be a content similarity relationship. If not, determine whether the nouns corresponding to the continuous vector sequence and the preset authority topic words have common upstream data or data sources. If the same upstream data or data sources exist, determine that the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship.

[0010] Furthermore, generating a word vector representation of the demand instruction text based on the dependency relationship specifically includes the following process: When the dependency relationship between the continuous vector sequence and the preset authority topic words is a content similarity relationship, Input into the bidirectional TextCNN network to obtain the last state of the forward TextCNN network and the last state of the backward TextCNN network, and concatenate the two states into a word vector representation; When the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship, the preset authority topic words are processed by Word2Vec word embedding technology to obtain a low-dimensional real number vector recorded as the word vector representation.

[0011] Furthermore, the word vector representation is input into the BERT-TextCNN hybrid model to output the demand category corresponding to the user's demand instruction text. The specific process includes the following: BERT encoding layer: Use the pre-trained BERT model to generate context-related vectors for each word vector representation; Output: A tensor of shape [batch_size, seq_len, hidden_dim], where hidden_dim is the hidden layer dimension of BERT. TextCNN layer: Convolutional layer: Use multiple convolution kernels of different sizes to slide on the word vector sequence to extract local features; Pooling layer: Perform maximum pooling on the output of each convolution kernel to obtain a feature vector of fixed length; Splicing features: Splicing the pooling results of different convolution kernels to form the final text feature representation; Fully connected layer: The feature vector output by TextCNN is mapped to the required category space through the fully connected layer, and the Softmax activation function is used to output the category probability.

[0012] Furthermore, activating the dynamic permission engine based on the requirement category to retrieve the permission matrix specifically includes the following process: Determine the number of relevant factors affecting permissions , build a set of relevant influencing factors of the authority matrix , ,in, is the user identity weight subset, where the user identity weight subset is a set of weights assigned to each user. is the user access time subset, where the access time subset is the set of the duration of each user's access. Access a subset of frequencies for the user, from the set Take out any two subsets from the set and compare them, and use It represents the ratio of importance, and assigns the corresponding importance according to the preset ratio, and combines the importance to form an authority matrix.

[0013] Furthermore, adaptive navigation for users based on the multimodal interaction mechanism and access path specifically includes the following processes: By integrating voiceprint verification, voice prompts, and AR guidance technologies, combined with user behavior analysis, it automatically skips redundant steps and embeds secondary verification at key nodes to achieve efficient and secure user-adaptive navigation.

[0014] On the other hand, a dynamic permission adaptation navigation system based on natural language processing includes: An instruction text parsing unit is used to obtain the user's demand instruction text, convert the word sequence corresponding to the demand instruction text into a continuous vector sequence, parse the dependency relationship between the continuous vector sequence and the preset permission topic words, and generate a word vector representation of the demand instruction text based on the dependency relationship; The instruction text classification unit is used to input the word vector representation into the BERT-TextCNN hybrid model and output the demand category corresponding to the user's demand instruction text; An access path determination unit is used to activate the dynamic permission engine based on the demand category to retrieve the permission matrix, generate a weighted optimal path using the DAG algorithm, and determine the access path corresponding to the user's demand instruction based on the weighted optimal path; The user adaptive navigation unit is used to perform adaptive navigation for users based on the multimodal interaction mechanism and access path.

[0015] Compared with the existing solutions, the present invention achieves the following beneficial effects: The present invention obtains the user's demand instruction text, converts the word sequence corresponding to the demand instruction text into a continuous vector sequence, analyzes the dependency relationship between the continuous vector sequence and preset permission topic words, and generates a word vector representation of the demand instruction text based on the dependency relationship; inputs the word vector representation into the BERT-TextCNN hybrid model, and outputs the demand category corresponding to the user's demand instruction text; activates the dynamic permission engine based on the demand category to call the permission matrix, combines the DAG algorithm to generate the weighted optimal path, and determines the access path corresponding to the user's demand instruction based on the weighted optimal path; and performs adaptive navigation on users based on the multimodal interaction mechanism and the access path, which can improve the efficiency of permission allocation, navigation interaction efficiency and interaction security.

[0016] By converting user demand instruction text into a continuous vector sequence and parsing its dependency on preset permission topic words, it can accurately capture the semantic associations in the instructions, eliminate ambiguity in natural language, and significantly reduce the misjudgment rate compared to traditional keyword matching methods.

[0017] The dynamic permission engine is activated based on the demand category, and the minimum necessary permissions are quickly matched from the permission matrix based on user roles, historical behaviors, and context, avoiding permission overload or insufficient, and shortening the permission allocation response time.

[0018] The weight of the access path is calculated through the directed acyclic graph (DAG) algorithm to generate the global optimal path and reduce redundant steps. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0020] Figure 1 This is a workflow diagram of the first dynamic permission adaptation navigation method based on natural language processing according to an embodiment of the present invention; Figure 2 This is a workflow diagram of a second dynamic permission adaptation navigation method based on natural language processing according to an embodiment of the present invention; Figure 3 This is a system block diagram of a dynamic permission adaptive navigation based on natural language processing in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure can be practiced while omitting one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0023] This embodiment provides a dynamic permission adaptation navigation method based on natural language processing. Figure 1 This is a workflow diagram of the first dynamic permission adaptation navigation method based on natural language processing according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps: Step S101: Obtain the user's demand instruction text, and convert the word sequence corresponding to the demand instruction text into a continuous vector sequence; Step S102: parsing the dependency relationship between the continuous vector sequence and the preset authority topic words, and generating a word vector representation of the demand instruction text based on the dependency relationship; Step S103: Input the word vector representation into the BERT-TextCNN hybrid model to output the demand category corresponding to the user's demand instruction text; Step S104: Activate the dynamic permission engine based on the demand category to retrieve the permission matrix, combine the DAG algorithm to generate the weighted optimal path, and determine the access path corresponding to the user's demand instruction based on the weighted optimal path; Step S105: performing adaptive navigation for the user based on the multimodal interaction mechanism and the access path.

[0024] In summary, the present invention converts user demand instruction text into a continuous vector sequence and analyzes its dependency relationship with preset permission topic words. This can accurately capture the semantic association in the instruction and eliminate ambiguity in natural language. Compared with traditional keyword matching methods, it significantly reduces the misjudgment rate. It activates the dynamic permission engine based on the demand category, and quickly matches the minimum necessary permissions from the permission matrix based on user roles, historical behaviors and contexts to avoid permission overload or insufficiency, shorten the permission allocation response time, calculate the weight of the access path through the directed acyclic graph (DAG) algorithm, generate the global optimal path, reduce redundant steps, and thus improve the permission allocation efficiency, navigation interaction efficiency and interaction security.

[0025] In some embodiments, converting the word sequence corresponding to the demand instruction text into a continuous vector sequence specifically includes the following process: The parts of speech of words in the demand instruction text are determined based on Stanford NLP Parser, and the text nouns representing the demand instructions are screened out based on the clustering algorithm. The text nouns are converted into continuous vector sequences through the embedding layer of the encoder.

[0026] Specifically, StanfordNLP Parser can segment Chinese text and mark the part of speech for each word, such as noun, verb, adjective, etc.

[0027] The specific process of filtering out nouns representing demand instructions based on the clustering algorithm includes the following: Form a matrix X of all words in the demand instruction text, calculate the covariance matrix S of the sample of matrix X, and calculate the eigenvector of the covariance matrix S 、 ,..., 、 and eigenvalues t = 1, 2, ..., n; project the data into the space of eigenvectors, using the formula: ,in, The value is the value of the dimension corresponding to the word in the demand instruction text. Filter reference values for words and set reference thresholds ,Will Greater than The corresponding m words are selected and recorded as words to be screened; the words to be screened form a data set to be screened The K value is set based on the size and shape of the data set to be screened, and any word to be screened is randomly selected as the initial cluster centroid. When the number of initial centroids is less than K, is the initial center of mass; based on the objective function Calculate the distance between each word to be filtered in the dataset to be filtered and the existing initial cluster centroid ,Will The word to be filtered corresponding to the maximum value is used as the next initial centroid; K initial centroids are obtained in sequence, and the data set to be filtered is clustered based on the k initial centroids to obtain the noun representing the demand instruction.

[0028] Converting text nouns into continuous vector sequences through the encoder's embedding layer includes: The encoder's embedding layer is a core component in natural language processing (NLP) models that converts discrete sequences of text words into continuous sequences of vectors. This process maps each word into a dense vector of fixed dimension (i.e., a word embedding), capturing the word's semantic and grammatical information. The following is a detailed description of the embedding layer: 1. The role of the embedding layer Discrete-to-continuous conversion: convert each word (or subword) in the text from a discrete index (such as a position in a dictionary) to a continuous vector representation.

[0029] Semantic representation: Through training, the embedding layer can learn the semantic relationship between words (for example, "cat" and "dog" are close in the vector space).

[0030] Dimensionality compression: Compress high-dimensional discrete representations (such as vocabulary size) into low-dimensional continuous vectors (such as 100-300 dimensions) to facilitate subsequent model processing.

[0031] 2. Implementation of the Embedding Layer The embedding layer is usually implemented through a trainable lookup table. The specific steps are as follows: 2.1 Input: Discrete Word Sequence The input is a sequence of integers, where each integer corresponds to a word (or subword) in the vocabulary.

[0032] Example: Sentence “I love NLP” → vocabulary index [1, 2, 3] (assuming “I” = 1, “love” = 2, “NLP” = 3).

[0033] 2.2 Lookup Table (Embedding Matrix) The embedding layer maintains a matrix Q∈ ,in: B: vocabulary size (e.g. 50,000).

[0034] d: Embedding dimension (e.g., 300).

[0035] Each word corresponds to a row in the matrix (i.e., the vector representation of the word).

[0036] 2.3 Output: Continuous Vector Sequence The input integer sequence is converted into a vector sequence through table lookup operation.

[0037] Example: Input [1, 2, 3] → Output [Q[1], Q[2], Q[3]] (for each Q[i]∈ ).

[0038] 3. Training the Embedding Layer The parameters of the embedding layer (i.e., the embedding matrix Q) are usually trained as follows: Random initialization: At the beginning of model training, the weights of the embedding matrix are randomly initialized.

[0039] End-to-end training: The embedding layer is updated along with other model parameters (such as LSTM and Transformer weights) via backpropagation.

[0040] Pre-trained embeddings: You can use pre-trained word vectors (such as Word2Vec, GloVe) to initialize the embedding layer and then fine-tune it on a specific task.

[0041] In some embodiments, parsing the dependency relationship between the continuous vector sequence and the preset authority topic words specifically includes the following process: Given a continuous vector sequence , Represents a continuous vector sequence Middle Morphemes, where morphemes are roots or affixes, ,in, Indicates the constituent morphemes No. syllables; A sequence of consecutive vectors Syllables represented by low-dimensional real vectors obtained through Word2Vec word embedding technology , Indicates the The first morpheme in A vector of syllables; calculate Corresponding to the preset permission topic words The sum of similarities dist: ; in, Indicates the first word in the preset authority topic The first morpheme in A vector of syllables, represents dot product, express The modulus value, express The modulus value of Determine whether the sum of similarities dist exceeds the preset similarity threshold. If so, the dependency relationship between the continuous vector sequence and the preset authority topic words is determined to be a content similarity relationship. If not, determine whether the nouns corresponding to the continuous vector sequence and the preset authority topic words have common upstream data or data sources. If the same upstream data or data sources exist, determine that the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship.

[0042] In some embodiments, generating a word vector representation of a requirement instruction text based on dependency relationships specifically includes the following process: When the dependency relationship between the continuous vector sequence and the preset authority topic words is a content similarity relationship, Input into the bidirectional TextCNN network to obtain the last state of the forward TextCNN network and the last state of the backward TextCNN network, and concatenate the two states into a word vector representation; When the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship, the preset authority topic words are processed by Word2Vec word embedding technology to obtain a low-dimensional real number vector recorded as the word vector representation.

[0043] In some embodiments, inputting the word vector representation into the BERT-TextCNN hybrid model and outputting the demand category corresponding to the user's demand instruction text specifically includes the following process: BERT encoding layer: Use the pre-trained BERT model to generate context-related vectors for each word vector representation; Output: A tensor of shape [batch_size, seq_len, hidden_dim], where hidden_dim is the hidden layer dimension of BERT. TextCNN layer: Convolutional layer: Use multiple convolution kernels of different sizes to slide on the word vector sequence to extract local features; Pooling layer: Perform maximum pooling on the output of each convolution kernel to obtain a feature vector of fixed length; Splicing features: Splicing the pooling results of different convolution kernels to form the final text feature representation; Fully connected layer: The feature vector output by TextCNN is mapped to the required category space through the fully connected layer, and the Softmax activation function is used to output the category probability.

[0044] In some embodiments, activating the dynamic permission engine based on the requirement category to retrieve the permission matrix specifically includes the following process: Determine the number of relevant factors affecting permissions , build a set of relevant influencing factors of the authority matrix , ,in, is the user identity weight subset, where the user identity weight subset is a set of weights assigned to each user. is the user access time subset, where the access time subset is the set of the duration of each user's access. Access a subset of frequencies for the user, from the set Take out any two subsets from the set and compare them, and use It represents the ratio of importance, and assigns the corresponding importance according to the preset ratio, and combines the importance to form an authority matrix.

[0045] In some embodiments, Figure 2 This is a workflow diagram of a second dynamic permission adaptation navigation method based on natural language processing according to an embodiment of the present invention. Figure 2 As shown, the DAG algorithm is combined to generate the weighted optimal path. The access path corresponding to the user's demand instruction is determined based on the weighted optimal path. The specific process includes the following: Step S201: Activate the dynamic permission engine and retrieve the permission matrix; The core mechanism of the dynamic permission engine is that the engine dynamically adjusts permissions based on real-time context (such as user identity, device status, access time, network environment) and predefined policies (RBAC / ABAC model).

[0046] The permission matrix serves as the data foundation, defines the mapping relationship between roles and resources, and supports fine-grained access control.

[0047] In the permission matrix retrieval process, when a user initiates a request, the engine intercepts the request through the API gateway and queries the user permissions from the permission matrix based on the identity authentication information (such as JWT token).

[0048] The engine assesses risks in real time (such as remote logins and high-frequency operations) and dynamically adjusts permissions (such as upgrade verification and restricted access).

[0049] Step S202: Optimize access paths based on the DAG algorithm; DAG model construction: Tasks or resource access requests are abstracted into DAG nodes, edges represent dependencies or access order, and weights represent costs (such as time and resource consumption).

[0050] Example: In the DAG of task A→B→C, the edge weight is the task execution time, and the goal is to find the path with the smallest total weight.

[0051] Shortest path algorithm selection, topological sorting + dynamic programming: applicable to DAGs without negative weight edges, with a time complexity of O(Z + P).

[0052] Bellman-Ford algorithm: If there are negative weight edges, it can detect negative weight cycles and calculate the shortest path, with a time complexity of O(ZP).

[0053] Steps for path optimization implementation: Input: DAG graph G=(Z,P), source point s, edge weight T(u,v).

[0054] Initialization: dis[s]=0, and dis for other nodes is ∞.

[0055] Perform topological sorting to obtain the node sequence.

[0056] Process nodes according to the sequence, with relaxation operation: If dist[u]+w(u,v)<dist[v], then update dist[v].

[0057] Output: The shortest paths and weights from the source point to each node.

[0058] Step S203: Combine the permission engine with the optimized path generation of the DAG; Association mapping of permissions and paths: Map the role-resource relationships in the permission matrix to DAG nodes, and the edge weights represent the costs of accessing resources (such as security verification time).

[0059] Example: When role R1 accesses resource R2, it needs to pass MFA authentication, and the edge weight increases the authentication time cost.

[0060] Dynamic path adjustment, where the engine adjusts the DAG edge weights according to the real-time context (such as increasing the weight for high-risk operations). Recalculate the shortest path to ensure the dynamic consistency of permissions and paths.

[0061] Balance between security and efficiency, achieving multi-objective optimization through weight design (such as security cost vs. execution efficiency). Example: For emergency operations, reduce the security check weight to prioritize response, and for regular operations, increase the weight to ensure security.

[0062] In some embodiments, integrate the three technologies of voiceprint verification, voice prompt, and AR guidance, automatically skip redundant steps in combination with user behavior analysis, and embed secondary verification at key nodes to achieve efficient and secure user adaptation navigation: 1. Core module design 1.1 Voiceprint verification module Function: Authenticate the user's identity through the user's voice features.

[0063] Process: The user issues a voice command (such as "start operation"), and the system extracts voiceprint features (MFCC, spectral features, etc.) from the voice, compares them with pre-stored voiceprint templates, and confirms the identity.

[0064] Advantages: Contactless verification is suitable for screenless devices or when hands are busy. It can be combined with voice content understanding (such as command keywords) to enhance security.

[0065] 1.2 Voice Prompt Module Function: Guide user operations through voice feedback.

[0066] Implementation: Dynamically generate prompts based on context (e.g., "Please aim the device at the target area"), supporting multiple languages and speeds to adapt to different user needs.

[0067] Optimization points: Avoid redundant prompts (e.g., simplify prompts when the user is already familiar with the process) and combine voice emotion recognition (e.g., provide soothing prompts when the user is anxious).

[0068] 1.3 AR Guidance Module Function: Provide visual operation guidance through augmented reality technology.

[0069] Implementation method: Use the camera to capture the environment and superimpose virtual arrows, highlighted areas and other guidance.

[0070] Example: In a maintenance scenario, AR marks parts that need to be disassembled.

[0071] Advantages: Reduced operational complexity, reduced learning costs, and real-time feedback on operational status (e.g., displaying “√” when alignment is correct).

[0072] 1.4 Redundant Step Skipping Mechanism Logic: Dynamic judgment based on user historical behavior, operational proficiency, and environmental context.

[0073] Example: Expert users skip basic steps and go directly to core operations.

[0074] In repetitive tasks, omit verified steps (such as equipment that has been calibrated).

[0075] Implementation technology: Machine learning models predict user intent (such as LSTM analysis of action sequences).

[0076] The rules engine defines skip conditions (e.g. "skip step X if it completed successfully 3 times in the past").

[0077] 1.5 Secondary Verification Module Trigger conditions: high-risk operations (such as payment, privilege escalation), abnormal environment (such as insufficient light, network fluctuations), and abnormal user behavior (such as excessively fast operation speed).

[0078] Verification method: Dynamic combination of voiceprint, fingerprint, face and other multi-modal verification.

[0079] Example: "Voiceprint + Blink Verification" is required for payment.

[0080] In some embodiments, Figure 3 This is a system block diagram of a dynamic permission adaptive navigation based on natural language processing according to an embodiment of the present invention, which is applicable to the above-mentioned dynamic permission adaptive navigation method based on natural language processing, such as Figure 3 As shown, the system includes: An instruction text parsing unit is used to obtain the user's demand instruction text, convert the word sequence corresponding to the demand instruction text into a continuous vector sequence, parse the dependency relationship between the continuous vector sequence and the preset permission topic words, and generate a word vector representation of the demand instruction text based on the dependency relationship; The instruction text classification unit is used to input the word vector representation into the BERT-TextCNN hybrid model and output the demand category corresponding to the user's demand instruction text; An access path determination unit is used to activate the dynamic permission engine based on the demand category to retrieve the permission matrix, generate a weighted optimal path using the DAG algorithm, and determine the access path corresponding to the user's demand instruction based on the weighted optimal path; The user adaptive navigation unit is used to perform adaptive navigation for users based on the multimodal interaction mechanism and access path.

[0081] The above embodiments can be implemented in whole or in part via software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in the embodiments of this application are fully or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0082] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0083] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0085] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0086] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A dynamic permission adaptation navigation method based on natural language processing, characterized in that: include: Obtain the user's demand instruction text, convert the word sequence corresponding to the demand instruction text into a continuous vector sequence, analyze the dependency relationship between the continuous vector sequence and the preset permission topic words, and generate a word vector representation of the demand instruction text based on the dependency relationship; Input the word vector representation into the BERT-TextCNN hybrid model to output the demand category corresponding to the user's demand instruction text; Based on the demand category, the dynamic permission engine is activated to retrieve the permission matrix, and the weighted optimal path is generated by combining the DAG algorithm. Based on the weighted optimal path, the access path corresponding to the user's demand instruction is determined; Adapt navigation for users based on multimodal interaction mechanisms and access paths.

2. The dynamic permission adaptation navigation method based on natural language processing according to claim 1 is characterized in that: Converting the word sequence corresponding to the demand instruction text into a continuous vector sequence specifically includes the following process: The parts of speech of words in the demand instruction text are determined based on Stanford NLP Parser, and the text nouns representing the demand instructions are screened out based on the clustering algorithm. The text nouns are converted into continuous vector sequences through the embedding layer of the encoder.

3. The dynamic permission adaptation navigation method based on natural language processing according to claim 2 is characterized in that: Based on the clustering algorithm, the specific nouns representing the demand instructions are selected. The following processes are included: Form a matrix X of all words in the demand instruction text, calculate the covariance matrix S of the sample of matrix X, and calculate the eigenvector of the covariance matrix S 、 ,..., 、 and eigenvalues t = 1, 2, ..., n; project the data into the space of eigenvectors, using the formula: ,in, The value is the value of the dimension corresponding to the word in the demand instruction text. Filter reference values for words and set reference thresholds ,Will Greater than The corresponding m words are selected and recorded as words to be screened; the words to be screened form a data set to be screened The K value is set based on the size and shape of the data set to be screened, and any word to be screened is randomly selected as the initial cluster centroid. When the number of initial centroids is less than K, is the initial center of mass; based on the objective function Calculate the distance between each word to be filtered in the dataset to be filtered and the existing initial cluster centroid ,Will The word to be filtered corresponding to the maximum value is used as the next initial centroid; K initial centroids are obtained in sequence, and the data set to be filtered is clustered based on the k initial centroids to obtain the noun representing the demand instruction.

4. The dynamic permission adaptation navigation method based on natural language processing according to claim 1 is characterized in that: Parsing the dependency relationship between the continuous vector sequence and the preset authority topic words specifically includes the following process: Given a continuous vector sequence , Represents a continuous vector sequence Middle Morphemes, where morphemes are roots or affixes, ,in, Indicates the constituent morphemes No. syllables; A sequence of consecutive vectors Syllables represented by low-dimensional real vectors obtained through Word2Vec word embedding technology , Indicates the The first morpheme in A vector of syllables; calculate Corresponding to the preset permission topic words The sum of similarities dist: ； in, Indicates the first word in the preset authority topic The first morpheme in A vector of syllables, represents dot product, express The modulus value, express The modulus value of Determine whether the sum of similarities dist exceeds the preset similarity threshold. If so, the dependency relationship between the continuous vector sequence and the preset authority topic words is determined to be a content similarity relationship. If not, determine whether the nouns corresponding to the continuous vector sequence and the preset authority topic words have common upstream data or data sources. If the same upstream data or data sources exist, determine that the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship.

5. The dynamic permission adaptation navigation method based on natural language processing according to claim 4 is characterized in that: Generating word vector representations of requirement instruction text based on dependency relationships specifically includes the following steps: When the dependency relationship between the continuous vector sequence and the preset authority topic words is a content similarity relationship, Input into the bidirectional TextCNN network to obtain the last state of the forward TextCNN network and the last state of the backward TextCNN network, and concatenate the two states into a word vector representation; When the dependency relationship between the continuous vector sequence and the preset authority topic words is a data flow relationship, the preset authority topic words are processed by Word2Vec word embedding technology to obtain a low-dimensional real number vector recorded as the word vector representation.

6. The dynamic permission adaptation navigation method based on natural language processing according to claim 1 is characterized in that: Inputting the word vector representation into the BERT-TextCNN hybrid model and outputting the demand category corresponding to the user's demand instruction text specifically involves the following process: BERT encoding layer: Use the pre-trained BERT model to generate context-related vectors for each word vector representation; Output: A tensor of shape [batch_size, seq_len, hidden_dim], where hidden_dim is the hidden layer dimension of BERT. TextCNN layer: Convolutional layer: Use multiple convolution kernels of different sizes to slide on the word vector sequence to extract local features; Pooling layer: Perform maximum pooling on the output of each convolution kernel to obtain a feature vector of fixed length; Splicing features: Splicing the pooling results of different convolution kernels to form the final text feature representation; Fully connected layer: The feature vector output by TextCNN is mapped to the required category space through the fully connected layer, and the Softmax activation function is used to output the category probability.

7. The dynamic permission adaptation navigation method based on natural language processing according to claim 1 is characterized in that: Activating the dynamic permission engine based on the requirement category to retrieve the permission matrix specifically involves the following process: Determine the number of relevant factors affecting permissions , build a set of relevant influencing factors of the authority matrix , ,in, is the user identity weight subset, where the user identity weight subset is a set of weights assigned to each user. is the user access time subset, where the access time subset is the set of the duration of each user's access. Access a subset of frequencies for the user, from the set Take out any two subsets from the set and compare them, and use It represents the ratio of importance, and assigns the corresponding importance according to the preset ratio, and combines the importance to form an authority matrix.

8. The dynamic permission adaptation navigation method based on natural language processing according to claim 1 is characterized in that: Adaptive navigation for users based on multimodal interaction mechanisms and access paths specifically includes the following processes: It integrates voiceprint verification, voice prompts, and AR guidance technologies, combines user behavior analysis to automatically skip redundant steps, and embeds secondary verification at key nodes to achieve efficient and secure user-adapted navigation.

9. Dynamic permission adaptation navigation system based on natural language processing, characterized by: The dynamic permission adaptation navigation method based on natural language processing applicable to any one of claims 1 to 8 comprises: An instruction text parsing unit is used to obtain the user's demand instruction text, convert the word sequence corresponding to the demand instruction text into a continuous vector sequence, parse the dependency relationship between the continuous vector sequence and the preset permission topic words, and generate a word vector representation of the demand instruction text based on the dependency relationship; The instruction text classification unit is used to input the word vector representation into the BERT-TextCNN hybrid model and output the demand category corresponding to the user's demand instruction text; An access path determination unit is used to activate the dynamic permission engine based on the demand category to retrieve the permission matrix, generate a weighted optimal path using the DAG algorithm, and determine the access path corresponding to the user's demand instruction based on the weighted optimal path; The user adaptive navigation unit is used to perform adaptive navigation for users based on the multimodal interaction mechanism and access path.

Images