An RPA agent and its application
Through entity recognition and knowledge graph technology, the fine-grained permissions of employees are dynamically matched, and the data leakage problem caused by employees' excessive access to knowledge data under the traditional permission allocation method is solved, achieving higher data security and targeting.
Patent Information
- Application Number
- CN202411285859.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-13
AI Technical Summary
The traditional employee permission allocation method has excessive access to knowledge data, resulting in an increase in the risk of data leakage and the inability to dynamically adjust permissions based on employee work content.
Through the entity recognition layer, knowledge graph recognition layer, text feature extraction layer, permission feature extraction layer and graph neural network, the similarity between employee characteristics and knowledge data permissions is calculated, and fine-grained permissions are dynamically matched to ensure that employees only access necessary data.
It effectively reduces the risk of data breaches, improves the pertinence and security of knowledge data permissions, and ensures that employees only access the data needed for their work.
Smart Images

Figure CN118798251B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and more specifically, to an RPA agent and its application. Background Art
[0002] The traditional way of allocating employee permissions has some shortcomings. Although the coarse-grained static allocation based on job roles can meet the work needs of employees, it often leads to employees obtaining knowledge and data permissions that exceed the requirements of their work, thereby increasing the risk of data leakage.
[0003] Therefore, a new permission allocation scheme is needed that can dynamically allocate corresponding fine-grained permissions to employees based on their work content at different stages. In this way, employees can only access necessary data, thereby effectively reducing the potential risk of data leakage. Summary of the invention
[0004] The present invention provides an RPA agent and its application, which obtains different entities existing in the employee's project background knowledge text, expands the text content of the entity through the employee knowledge graph, extracts features from the text, and calculates the similarity between the employee characteristics of each entity and the permission characteristics of each knowledge data permission for different entities, thereby matching a knowledge data permission for different entities of the current employee. The matched knowledge data permission refers to the permission of a sub-file or file under a certain type of knowledge database, which solves the problem of data leakage caused by employees' excessive acquisition of knowledge data.
[0005] The present invention provides an RPA agent, comprising: an entity recognition layer, which is used to input the project background knowledge of employees and then obtain named entities;
[0006] The knowledge graph recognition layer matches the extracted named entities with the entities in the employee knowledge graph and marks the matched entities, and performs m non-repeating random walks with the marked entities as the center; the entities in each random walk are sorted according to the order of the random walk to obtain the second text;
[0007] The text feature extraction layer extracts features from the second text obtained by random walking of the same marked entity through the BERT model to obtain a text feature, averages the obtained text features to obtain the employee feature of the marked entity, and sequentially obtains the employee features of entities with different marks;
[0008] The employee feature calculation formula for entities with different tags is as follows:
[0009] ;
[0010] in, represents the text features extracted by the i-th random walk of each marked entity, represents the employee features of the i-th marked entity, and m represents the total number of random walks of the marked entity;
[0011] The permission feature extraction layer calculates the permission features of each knowledge data permission;
[0012] The calculation formula of the permission characteristic is as follows:
[0013] ;
[0014] in represents the authority feature of the yth knowledge data authority relative to the i-th marked entity, represents the set of employees to whom the yth knowledge data authority has been assigned, and d represents The total number of employees in express The employee features of the ith labeled entity of the pth employee in , represents the employee features of the i-th labeled entity of the target employee, represents the similarity function;
[0015] The authority identification layer calculates the similarity between the employee characteristics of the target employee's i-th marked entity and the authority characteristics of the knowledge data authority relative to the i-th marked entity; selects the knowledge data authority with the greatest similarity as the knowledge data authority that matches the i-th marked entity of the target employee; and puts the duplicates into the matching authority set;
[0016] The graph structure data encoding layer is centered on the marked entity corresponding to the permission characteristics of the knowledge data permission in the matching permission set, and randomly walks on the employee knowledge graph to obtain the corresponding entity set; the graph structure data is established based on the representation characteristics of the matching permission set, entity set, and marked entity;
[0017] Graph neural network, which inputs graph structure data and outputs recognition results of knowledge data authority.
[0018] In a preferred embodiment, the identification result of the knowledge data authority includes the allocation order or authority level of the knowledge data authority.
[0019] In a preferred embodiment, the knowledge graph recognition layer performs the following steps:
[0020] Step 101, entity matching, the matching rule is: if the cosine similarity between the entity in the employee knowledge graph and the named entity extracted by the entity recognition layer in step 101 is greater than a set first threshold, then the entity in the employee knowledge graph is matched with the named entity extracted by the entity recognition layer, otherwise it is not matched;
[0021] Step 102, mark the matching entities in the employee knowledge graph, perform random walks, and perform m non-repeated random walks for each marked entity. The step length of the walk is 1, and the step length of each walk is n steps. The default values of m and n are 10;
[0022] Step 103, entity sequence generation, arrange the results of each random walk in the order of the walk to obtain the second text.
[0023] In a preferred embodiment, the entity recognition layer includes a Transformer encoder and a conditional random field layer, and the calculation process of the Transformer encoder is as follows:
[0024] Compute the query matrix for the i-th head of the Transformer encoder , key matrix Sum Matrix :
[0025] ;
[0026] in , , denote the query weight matrix, key weight matrix, and value weight matrix of the i-th head, respectively. is the project background knowledge, which is represented by a matrix, and a row of the matrix represents the embedding feature of a word in a vocabulary sequence of the employee's background project text information;
[0027] Calculate the attention weight matrix of the i-th head :
[0028] ;
[0029] represents the normalized exponential function, The number of dimensions of the row vector representing the project background knowledge, T represents transpose;
[0030] The weight matrix of the i-th head and the value matrix of the i-th head Multiply to get the output of the i-th head :
[0031] ;
[0032] Multi-head splicing:
[0033] ;
[0034] in represents the attention encoding matrix, , and Represent the outputs of the 1st, 2nd and Bth heads respectively, is the output layer weight matrix, Concat represents vector concatenation;
[0035] The attention encoding matrix inputs the conditional random field layer and outputs a label sequence. The rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence. The entity types include unnamed entities and named entities.
[0036] In a preferred embodiment, the method for acquiring project background knowledge is as follows:
[0037] Collect background project text information of employees and employees to whom knowledge data permissions are assigned. Background project text information includes: employee name, position, skills, responsibilities, and text information of current projects involved;
[0038] Convert the background project text information of the current employee into a vocabulary sequence;
[0039] The vocabulary sequence is input into the word embedding model to obtain the embedding features, and the embedding features are combined to obtain the project background knowledge.
[0040] In a preferred embodiment, the calculation formula of the hidden layer of the graph neural network is as follows:
[0041] ;
[0042] in represents the encoding feature of node i, represents the hidden layer weight parameter, and Represent the input features of the information represented by the i-th and j-th nodes respectively, represents the set of nodes directly connected to the i-th node, express The total number of nodes in represents the proximity coefficient, It is a sigmoid function;
[0043] ;
[0044] Among them, tanh represents the hyperbolic tangent function, Indicates adjustable parameters; The default value is 1 / N, where N is the dimension of the node feature and T represents transposition.
[0045] In a preferred embodiment, the calculation formula of the graph neural network output layer is as follows:
[0046] ;
[0047] is the encoding feature of node v, , A collection of nodes representing knowledge data permissions, represents the identification weight parameter, represents the identification bias parameter, Represents an output vector, where one component of the output vector represents the probability value of a sequence number of an allocation order, and the sequence number of the allocation order with the largest probability value is marked as the knowledge data authority corresponding to the node v.
[0048] In a preferred embodiment, the input features of a node representing a marked entity are the characterization features of the marked entity, the input features of a node representing an entity in the entity set H are obtained by encoding the entity information of the entity, and the input features of a node representing an entity of knowledge data authority are obtained by encoding the text information of the knowledge data authority.
[0049] The beneficial effects of the present invention are:
[0050] 1. The present invention can identify different entities existing in the project background knowledge text of the current employee through entity recognition and knowledge graph technology, thereby obtaining more fine-grained employee characteristics of the current employee;
[0051] 2. The present invention calculates the different employee characteristics of different entities in the current employee's project background knowledge text and the text characteristics of the employee's project background knowledge text assigned under different knowledge data authorities through an attention mechanism to obtain the authority characteristics of the knowledge data authority relative to different entities of the current employee, thereby improving the targeted nature of the knowledge data authority. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a module diagram of an RPA agent of the present invention.
[0053] In the figure: entity recognition layer 100, knowledge graph recognition layer 200, text feature extraction layer 300, authority feature extraction layer 400, authority recognition layer 500, graph structure data encoding layer 600, graph neural network 700. DETAILED DESCRIPTION
[0054] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.
[0055] At least one embodiment of the present invention discloses an RPA agent, such as Figure 1 As shown, including:
[0056] The entity recognition layer 100 is used to input the project background knowledge of the employees and then obtain the named entities;
[0057] The entity recognition layer includes:
[0058] Transformer encoder and conditional random field layer ConditionalRandomField; the calculation process of Transformer encoder is as follows:
[0059] Compute the query matrix for the i-th head of the Transformer encoder , key matrix Sum Matrix :
[0060] ; ;
[0061] in , , denote the query weight matrix, key weight matrix, and value weight matrix of the i-th head, respectively. is the project background knowledge, which is represented by a matrix, and a row of the matrix represents the embedding feature of a word in a vocabulary sequence of the employee's background project text information;
[0062] Calculate the attention weight matrix of the i-th head :
[0063] ;
[0064] represents the normalized exponential function, The number of dimensions of the row vector representing the project background knowledge, T represents transpose;
[0065] The weight matrix of the i-th head and the value matrix of the i-th head Multiply to get the output of the i-th head :
[0066] ;
[0067] Multi-head splicing: ;
[0068] in represents the attention encoding matrix, , and Represent the outputs of the 1st, 2nd and Bth heads respectively, is the output layer weight matrix, Concat represents vector concatenation;
[0069] The attention encoding matrix inputs the conditional random field layer and outputs a label sequence. The rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence. The entity types include unnamed entities and named entities.
[0070] The method of obtaining project background knowledge is as follows:
[0071] Collect background project text information of employees and employees to whom knowledge data permissions are assigned. Background project text information includes: employee name, position, skills, responsibilities, and text information of current projects involved;
[0072] Convert the background project text information of the current employee into a vocabulary sequence;
[0073] The vocabulary sequence is input into the word embedding model to obtain the embedding features, and the embedding features are combined to obtain the project background knowledge.
[0074] In one embodiment of the present invention, a simple example of a vocabulary sequence is as follows:
[0075] Data analyst, programming language, Python, financial data automation report project, responsible for data analysis and modeling.
[0076] The corresponding tag sequence is shown below:
[0077] ;
[0078] The meanings of the tags are as follows:
[0079] O: non-named entity, B-Parameters: named entity;
[0080] The obtained named entity representation is as follows:
[0081] Data analyst, programming language, Python, financial data automation report project, data analysis, modeling.
[0082] It should be noted that when an employee changes projects, knowledge data permissions need to be redone.
[0083] Therefore, one employee may correspond to multiple background project text information, and when he changes the project, a new background project text information will be generated.
[0084] The knowledge graph recognition layer 200 matches the extracted named entities with the entities in the employee knowledge graph and marks the matched entities, and performs m non-repetitive random walks with the marked entities as the center; the entities in each random walk are sorted according to the order of the random walk to obtain the second text;
[0085] The employee knowledge graph includes entities representing the following information: project name, project content, project process, project position, project responsibilities, employee position, employee name, employee skills, employee responsibilities, school name, major name, and knowledge data permissions.
[0086] The employee knowledge graph expresses the relationship between employees, projects, positions, knowledge data permissions, educational background and other information.
[0087] It can be obtained by collating information from the company's history.
[0088] In one embodiment of the present invention, the knowledge graph recognition layer performs the following steps:
[0089] Step 101, entity matching, the matching rule is: if the cosine similarity between the entity in the employee knowledge graph and the named entity extracted by the entity recognition layer in step 101 is greater than a set first threshold, then the entity in the employee knowledge graph is matched with the named entity extracted by the entity recognition layer, otherwise it is not matched;
[0090] The default value of the first threshold is 0.85.
[0091] Step 102, mark the matching entities in the employee knowledge graph, perform random walks, and perform m non-repeated random walks for each marked entity. The step length of the walk is 1, and the step length of each walk is n steps. The default values of m and n are 10;
[0092] Step 103, entity sequence generation, arrange the results of each random walk in the order of the walk to obtain the second text.
[0093] The text feature extraction layer 300 performs feature extraction on the second text obtained by random walking of the same marked entity through the BERT (Bidirectional Encoder Representations from Transformers) model to obtain a text feature, averages the obtained text features to obtain the employee feature of the marked entity, and sequentially obtains the employee features of entities with different marks;
[0094] In one embodiment of the present invention, the employee feature calculation formula for entities with different tags is as follows:
[0095] ;
[0096] in, represents the text features extracted by the i-th random walk of each marked entity, represents the employee features of the i-th labeled entity, and m represents the total number of random walks of the labeled entity.
[0097] The permission feature extraction layer 400 calculates the permission feature of each knowledge data permission;
[0098] In one embodiment of the present invention, the calculation formula of the permission feature is as follows:
[0099] ;
[0100] in represents the authority feature of the yth knowledge data authority relative to the i-th marked entity, represents the set of employees to whom the yth knowledge data authority has been assigned, and d represents The total number of employees in express The employee features of the ith labeled entity of the pth employee in , represents the employee features of the i-th labeled entity of the target employee, Represents the similarity function.
[0101] The authority identification layer 500 calculates the similarity between the employee characteristics of the target employee's i-th marked entity and the authority characteristics of the knowledge data authority relative to the i-th marked entity; selects the knowledge data authority with the greatest similarity as the knowledge data authority matched by the i-th marked entity of the target employee;
[0102] After deleting duplicates, put them into the matching permission set;
[0103] In one embodiment of the present invention, the similarity calculation formula of the permission identification layer is as follows:
[0104] ;
[0105] in, represents the similarity value between the employee feature of the ith marked entity and the yth knowledge data authority relative to the authority feature of the ith marked entity, Represents the similarity function.
[0106] The graph structure data encoding layer 600 is centered around the marked entities corresponding to the permission features of the knowledge data permissions in the permission set (e.g. The corresponding marked entity is the i-th marked entity), and a random walk is performed on the employee knowledge graph (the step length of the random walk is 1, the number of steps is n, the number of times is m, the default value of n is 10, and the default value of m is 10), and the corresponding entity set is obtained, and the obtained entity set is named H. The entities in an entity set H include employees, projects, and positions; based on the matching permission set, entity set H, and the characterization features of the marked entities, a graph structure data is established, and the graph structure data includes nodes and edges connecting nodes. A node represents a knowledge data permission of the matching permission set or an entity in the entity set H or a marked entity;
[0107] The edges between the nodes representing the entities or knowledge data permissions in the entity set H represent the entity relationships between the entities or knowledge data permissions in the entity set H on the employee knowledge graph;
[0108] The two nodes represent the characterization features of the marked entity and the knowledge data authority respectively. If the similarity between the characterization features of the marked entity and the knowledge data authority relative to the authority features of the marked entity is the largest, then there is an edge between the two nodes, otherwise there is no edge.
[0109] A graph neural network 700, which inputs graph structure data and outputs the order of assigning knowledge data permissions;
[0110] In one embodiment of the present invention, the graph neural network includes a hidden layer and an output layer;
[0111] The calculation formula of the hidden layer is as follows:
[0112] ;
[0113] in represents the encoding feature of node i, represents the hidden layer weight parameter, and Represent the input features of the information represented by the i-th and j-th nodes respectively, represents the set of nodes directly connected to the i-th node, express The total number of nodes in represents the proximity coefficient, It is a S-shaped function.
[0114] The input feature of the node representing the marked entity is the characterization feature of the marked entity. The input feature of the node representing the entity in the entity set H is obtained by encoding the entity information of the entity. The input feature of the node representing the entity of the knowledge data authority is obtained by encoding the text information of the knowledge data authority.
[0115] The entity information of an entity in set H represents the information associated with the entity. If an entity is a position, then its associated entity information is the introduction text corresponding to the position;
[0116] ;
[0117] Among them, tanh represents the hyperbolic tangent function, Indicates adjustable parameters; The default value is 1 / N, where N is the dimension of the node feature and T represents transposition.
[0118] The output layer inputs the encoding features of the nodes representing the knowledge data authority, and outputs the value of the serial number representing the allocation order of the knowledge data authority.
[0119] The calculation formula of the output layer is as follows:
[0120] ;
[0121] is the encoding feature of node v, , A collection of nodes representing knowledge data permissions, represents the identification weight parameter, represents the identification bias parameter, Represents an output vector, where one component of the output vector represents the probability value of a sequence number of an allocation order, and the sequence number of the allocation order with the largest probability value is marked as the knowledge data authority corresponding to the node v.
[0122] It should be noted that the order of knowledge data authority allocation is defined as the stage of the project at which the allocation is made. For example, if the order of knowledge data authority allocation is 3, it will be allocated to employees at the beginning of the third stage of the project.
[0123] The order of assigning more than one knowledge data permission may be the same, because an employee may need multiple knowledge data permissions at one stage.
[0124] In one embodiment of the present invention, the graph neural network outputs the permission level of the knowledge data permission;
[0125] The output layer of the graph neural network inputs the encoded features of the nodes representing the knowledge data permissions, and outputs the value representing the permission level of the knowledge data permissions.
[0126] For example, the permission level includes three levels. The first level permission includes editing, saving, and viewing permissions. The second level permission includes saving and viewing permissions. The third level permission includes viewing permissions.
[0127] The output layer of the graph neural network outputs the probability values corresponding to the three levels of permissions, and selects the permission level with the largest probability value as the output.
[0128] In one embodiment of the present invention, an application of an RPA agent is provided, which uses the aforementioned RPA agent to assign knowledge data permissions to employees and mark the stages of the projects to which the knowledge data permissions are assigned.
[0129] In one embodiment of the present invention, an RPA agent application is provided, which uses the aforementioned RPA agent to assign knowledge data permissions to employees and mark the permission levels of the knowledge data permissions.
[0130] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.
Claims
1. An RPA agent system, characterized in that: include: The entity recognition layer is used to input the employee's project background knowledge and then obtain the named entities; The knowledge graph recognition layer matches the extracted named entities with the entities in the employee knowledge graph and marks the matched entities, and performs m non-repeating random walks with the marked entities as the center; the entities in each random walk are sorted according to the order of the random walk to obtain the second text; The text feature extraction layer extracts features from the second text obtained by random walking of the same marked entity through the BERT model to obtain a text feature, averages the obtained text features to obtain the employee feature of the marked entity, and sequentially obtains the employee features of entities with different marks; The employee feature calculation formula for entities with different tags is as follows: ; in, represents the text features extracted by the i-th random walk of each marked entity, represents the employee features of the i-th marked entity, and m represents the total number of random walks of the marked entity; The permission feature extraction layer calculates the permission features of each knowledge data permission; The calculation formula of the permission characteristic is as follows: ; in represents the authority feature of the yth knowledge data authority relative to the i-th marked entity, represents the set of employees to whom the yth knowledge data authority has been assigned, and d represents The total number of employees in express The employee features of the ith labeled entity of the pth employee in , represents the employee features of the i-th labeled entity of the target employee, represents the similarity function; The authority identification layer calculates the similarity between the employee characteristics of the target employee's i-th marked entity and the authority characteristics of the knowledge data authority relative to the i-th marked entity; selects the knowledge data authority with the greatest similarity as the knowledge data authority that matches the i-th marked entity of the target employee; and puts the duplicates into the matching authority set; The graph structure data encoding layer is centered on the marked entity corresponding to the permission characteristics of the knowledge data permission in the matching permission set, and randomly walks on the employee knowledge graph to obtain the corresponding entity set; the graph structure data is established based on the representation characteristics of the matching permission set, entity set, and marked entity; Graph neural network, which inputs graph structure data and outputs recognition results of knowledge data authority.
2. The RPA agent system according to claim 1, characterized in that: The identification result of the knowledge data authority includes the allocation order or authority level of the knowledge data authority.
3. The RPA agent system according to claim 1, characterized in that: The knowledge graph recognition layer performs the following steps: Step 101, entity matching, the matching rule is: if the cosine similarity between the entity in the employee knowledge graph and the named entity extracted by the entity recognition layer in step 101 is greater than a set first threshold, then the entity in the employee knowledge graph is matched with the named entity extracted by the entity recognition layer, otherwise it is not matched; Step 102, mark the matching entities in the employee knowledge graph, perform random walks, and perform m non-repeated random walks for each marked entity. The step length of the walk is 1, and the step length of each walk is n steps. The default values of m and n are 10; Step 103, entity sequence generation, arrange the results of each random walk in the order of the walk to obtain the second text.
4. The RPA agent system according to claim 1, characterized in that: The entity recognition layer includes the Transformer encoder and the conditional random field layer. The calculation process of the Transformer encoder is as follows: Compute the query matrix for the i-th head of the Transformer encoder , key matrix Sum Matrix : ; in , , denote the query weight matrix, key weight matrix, and value weight matrix of the i-th head, respectively. is the project background knowledge, which is represented by a matrix, and a row of the matrix represents the embedding feature of a word in a vocabulary sequence of the employee's background project text information; Calculate the attention weight matrix of the i-th head : ; represents the normalized exponential function, The number of dimensions of the row vector representing the project background knowledge, T represents transpose; The weight matrix of the i-th head and the value matrix of the i-th head Multiply to get the output of the i-th head : ; Multi-head splicing: ; in represents the attention encoding matrix, , and Represent the outputs of the 1st, 2nd and Bth heads respectively, is the output layer weight matrix, Concat represents vector concatenation; The attention encoding matrix inputs the conditional random field layer and outputs a label sequence. The rth label of the label sequence represents the entity type of the rth word in the vocabulary sequence. The entity types include unnamed entities and named entities.
5. The RPA agent system according to claim 4, characterized in that: The method of obtaining project background knowledge is as follows: Collect background project text information of employees and employees to whom knowledge data permissions are assigned. Background project text information includes: employee name, position, skills, responsibilities, and text information of current projects involved; Convert the background project text information of the current employee into a vocabulary sequence; The vocabulary sequence is input into the word embedding model to obtain the embedding features, and the embedding features are combined to obtain the project background knowledge.
6. The RPA agent system according to claim 1, characterized in that: The calculation formula of the hidden layer of the graph neural network is as follows: ; in represents the encoding feature of node i, represents the hidden layer weight parameter, and Represent the input features of the information represented by the i-th and j-th nodes respectively, represents the set of nodes directly connected to the i-th node, express The total number of nodes in represents the proximity coefficient, It is a sigmoid function; ; Among them, tanh represents the hyperbolic tangent function, Indicates adjustable parameters; The default value is 1 / N, where N is the dimension of the node feature and T represents transposition.
7. The RPA agent system according to claim 6, characterized in that: The calculation formula of the output layer of the graph neural network is as follows: ; is the encoding feature of node v, , A collection of nodes representing knowledge data permissions, represents the identification weight parameter, represents the identification bias parameter, Represents an output vector, where one component of the output vector represents the probability value of a sequence number of an allocation order, and the sequence number of the allocation order with the largest probability value is marked as the knowledge data authority corresponding to the node v.
8. The RPA agent system according to claim 6, characterized in that: The input features of a node representing a marked entity are the characterization features of the marked entity. The input features of a node representing an entity in the entity set H are obtained by encoding the entity information of the entity. The input features of a node representing an entity of knowledge data authority are obtained by encoding the text information of the knowledge data authority.
9. A method for applying an RPA agent system, characterized in that: Use an RPA agent system as described in any one of claims 1-8 to assign knowledge data permissions to employees and mark the stages of the projects to which the knowledge data permissions are assigned.
10. A method for applying an RPA agent system, characterized in that: Use an RPA agent system as described in any one of claims 1-8 to assign knowledge data permissions to employees and mark the permission levels of the knowledge data permissions.
