Integrated office method and system based on AI assistant

Abstract

The application provides an integrated office method and system based on an AI assistant, which comprises the following steps: receiving a natural language instruction input by a user based on a workbench, and analyzing the natural language instruction based on an AI assistant to identify the user's intention; determining the operation to be performed by the user based on the user's intention, verifying the user's authority, and determining the accessible authorized application of the user; automatically logging in to the accessible authorized application based on SSO authentication, scheduling an RPA driving engine to automate a business process according to the operation to be performed by the user, and returning the execution result to the user terminal. The application realizes intelligent interaction and business automation under a unified portal, and significantly improves office efficiency and security.

Description

Technical Field

[0001] This invention relates to the fields of enterprise informatization and artificial intelligence technology, and in particular to an integrated office method and system based on an AI assistant. Background Technology

[0002] Currently, enterprises generally face many pain points in their daily office operations. First, the phenomenon of "application silos" is serious. Employees need to remember the access addresses, accounts and passwords of multiple intranet systems. Switching between different systems is cumbersome and inefficient. Second, there is a lack of a unified, intelligent and configurable office portal, resulting in a fragmented office experience. Third, traditional office clients lack proactive service and automation capabilities, and the operation process is not intelligent enough. In addition, the enterprise's knowledge assets, such as rules and regulations, project documents, and historical data, are often disconnected from daily office processes, forming "knowledge isolation". Finally, application access, identity authentication and permission management are usually scattered in different systems, making management complex and security policies difficult to implement uniformly. Existing solutions, such as traditional enterprise portals or simple application aggregation platforms, mostly only solve the problem of centralized display of applications. They cannot seamlessly integrate functions such as application access, knowledge management, AI collaboration and business process automation, and are unable to meet the deep-seated needs of modern enterprises for intelligent and automated office work. Therefore, in order to overcome the above-mentioned defects, the present invention provides an integrated office method and system based on an AI assistant. Summary of the Invention

[0003] This invention provides an integrated office method and system based on an AI assistant. It receives natural language commands and automatically parses user intent, intelligently matches accessible applications based on permission verification results, achieves seamless authentication with single sign-on technology, schedules an automated process engine to execute specific operations, and finally feeds back the results to the user. This effectively solves the problems of cumbersome switching between multiple systems, complex operation paths, and fragmented permission management, and realizes intelligent interaction and business automation under a unified entry point, significantly improving office efficiency and security.

[0004] This invention provides an integrated office method based on an AI assistant, comprising: Step 1: Receive natural language commands input by the user based on the workbench, and use the AI ​​assistant to parse the natural language commands and identify the user's intent; Step 2: Determine the user's operation to be performed based on the user's intent, and verify the user's permissions to determine the user's authorized applications; Step 3: Automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user's terminal.

[0005] Preferably, in an integrated office method based on an AI assistant, step 1 involves receiving natural language commands input by the user on the workbench and parsing these commands with the AI ​​assistant to identify the user's intent, including: The workbench-based interactive interface receives natural language commands input by the user, and the AI ​​assistant also receives natural language commands. The natural language instructions are preprocessed, and the resulting instruction vector sequence is input into a pre-defined intent understanding model. The preprocessing includes word segmentation and vectorization of the natural language instructions. Named entity recognition is performed on instruction vector sequences based on the intent understanding model, and key parameter entities related to operation behavior are extracted from natural language instructions based on the named entity recognition results; Meanwhile, the intent category of the instruction vector is identified based on the intent understanding model, and the preset operation category to which the natural language instruction belongs is determined based on the intent category identification result; Based on the key parameter entities extracted by named entity recognition and the preset operation categories of intent category recognition results, a structured description of user intent is generated.

[0006] Preferably, an integrated office method based on an AI assistant preprocesses natural language commands, including: Load a pre-built Chinese word segmentation dictionary, and based on the Chinese word segmentation dictionary, use the maximum positive matching algorithm to segment natural language instructions into word segments, dividing continuous text into independent lexical unit sequences; The segmented word unit sequence is standardized, and the standardized word unit sequence is input into the pre-trained word embedding model; Based on the word embedding model, each word unit is mapped to a fixed-dimensional real vector, and based on the mapping, the real vectors corresponding to all word units are combined in the original sequence order to generate an instruction vector sequence.

[0007] Preferably, an integrated office method based on an AI assistant, which performs named entity recognition on instruction vector sequences based on an intent understanding model, includes: Obtain a pre-defined intent understanding model, wherein the intent understanding model adopts a neural network architecture that integrates BERT, bidirectional long short-term memory network and conditional random field; The obtained instruction vector sequence is input into the BERT encoding layer in the intent understanding model, and the instruction vector sequence is context-encoded based on the BERT encoding layer to generate a feature vector sequence containing contextual semantic information. The feature vector sequence is input into the bidirectional long short-term memory network layer in the intent understanding model, and the forward and backward semantic dependencies of the feature vector sequence are modeled based on the bidirectional long short-term memory network layer, outputting the context-enhanced feature vector corresponding to each lexical unit. The context-enhanced feature vector sequence is input into the conditional random field layer in the intent understanding model, and the probability distribution of each word unit belonging to a preset entity label is calculated based on the context-enhanced feature vector according to the conditional random field layer. The preset entity labels include time, type, and quantity. The globally optimal entity label sequence is determined based on the probability distribution, and the corresponding key parameter entities and types are extracted from the original natural language instructions based on the entity label sequence.

[0008] Preferably, in an integrated office method based on an AI assistant, step 2 involves determining the user's intended operation based on the user's intent, verifying the user's permissions, and determining the user's authorized applications, including: The user intent is obtained and parsed to determine the target application identifier, operation type, and key parameter entities extracted from the natural language instructions. Based on the target application identifier, the pre-set application metadata database is queried to determine the application access information and associated permission group requirements corresponding to the target application identifier; At the same time, obtain the current user's identity identifier, and based on the user's identity identifier, query the role-based access control library to determine one or more roles to which the user's identity identifier belongs; Based on the user's role, query the role-permission relationship and determine the set of permission groups granted to the user based on the query results; Based on the operation type and key parameter entities, the corresponding verification rules are matched in the preset fine-grained permission policy library. The application's associated permission group requirements are matched and verified against the set of permission groups granted to the user. At the same time, based on the verification rules, the operation type and the attachment permissions of key parameter entities are verified and checked during the verification process. When the matching verification passes, the target application is identified as an accessible and authorized application.

[0009] Preferably, an integrated office method based on an AI assistant determines the target application identifier, operation type, and key parameter entities extracted from natural language instructions corresponding to the user's intent, including: The target application identifier, operation type, and key parameter entities extracted from natural language instructions are obtained, and a pre-set operation category mapping table is queried based on the operation type to determine the operation execution interface category corresponding to the operation type. Based on the target application identifier, query the application metadata database to determine the application access interface information corresponding to the target application identifier; Generate an initial operation command template based on the operation execution interface category and application access interface information; Extract the operation parameters corresponding to the operation instruction to be executed, extract the necessary parameter values ​​that match the initial operation command template from the operation parameters, and fill the necessary parameter values ​​into the parameter slots of the initial operation command template to generate a complete operation instruction to be executed. The operation instruction to be executed is associated with the target application identifier, and the output is a schedulable operation context to be executed.

[0010] Preferably, in an integrated office method based on an AI assistant, step 3 involves automatically logging into authorized applications based on SSO authentication, scheduling the RPA-driven engine to automate business processes according to the user's pending operations, and returning the execution results to the user's terminal, including: The system obtains the user's pending operations and authorized applications, parses the pending operations to determine the corresponding pending operation instructions, and extracts the target application identifier of the authorized applications. The SSO authentication module obtains a valid access token for the authorized application based on the target application identifier, and obtains the authentication session status of the network client based on the SSO authentication module. Extract the original session token from the authentication session state and convert it into a unified security token based on preset conversion rules; Extract the configuration parameters of the authorized applications and determine the access mode configured for the authorized applications based on the configuration parameters; When the access mode is H5 container mode, the unified security token is entered into the HTTP request header of the H5 container, and access to the authorized application is initiated based on the entry result. When the access mode is configured as browser mode, a deep link URL carrying a unified security token is constructed, and the deep link URL is opened based on the preset default browser. Once an authentication session with an authorized application is successfully established, the operation command to be executed is sent to the RPA driver engine, and the RPA driver engine is controlled to match the corresponding automated execution script from the preset script library according to the operation command to be executed. If no match is found, an executable script fragment is generated in real time based on the operation type, parameter structure and target application interface characteristics of the operation command to be executed. Extract the operation parameters corresponding to the operation instructions to be executed, and dynamically input the operation parameters into the matching automated execution script based on the RPA driving engine; The automated execution script is driven by the RPA driver engine after the operation parameters are entered. The execution status is monitored in real time and a fault-tolerant rollback mechanism is triggered to restore to the previous stable state when errors or exceptions occur. The output data during the execution process is obtained. After the automated script finishes execution, the output data during the execution process is summarized and the summarized output data is formatted. The formatted output data is returned to the user terminal.

[0011] Preferably, an integrated office method based on an AI assistant, which obtains a valid access token for authorized applications based on the target application identifier using an SSO authentication module, includes: The SSO authentication module queries the application metadata database based on the target application identifier and obtains the authentication protocol type and token generation parameters of the authorized application based on the query results. Based on the authentication protocol type and token generation parameters, the system queries the preset token cache library to determine whether there is an unexpired access token corresponding to the current user session and the authorized applications that can be accessed. If an unexpired access token exists, return a valid access token that allows access to the authorized application from the preset token cache. When no expired access token exists, a token issuance request is sent to the central authentication server based on the SSO authentication module. The token issuance request includes the user identity identifier and the target application identifier. The central authentication server parses the token issuance request and builds a service connection with the authentication source based on the parsing result. The service connection control authentication source verifies the user's identity and sends an authentication response back to the central authentication server after successful verification. When the authentication response is successful, the central authentication server generates token claim data containing user identity information, target application identifier and timestamp, and digitally signs the token claim data based on a preset encryption key to obtain a valid access token that can access the authorized application. The obtained valid access token is returned to the SSO authentication module.

[0012] Preferably, in an integrated office method based on an AI assistant, step 3, which involves scheduling the RPA-driven engine to automate business processes, further includes: Based on the AI ​​assistant's real-time collection of target semantic understanding confidence of currently undefined graphical user interface elements; The target semantic understanding confidence level is compared with a preset confidence threshold. At the same time, the interaction intent and contextual logic relationship of the currently undefined graphical user interface elements are obtained. When the confidence level of the target semantic understanding is greater than or equal to the preset confidence threshold, the target programming logic operation sequence is generated based on the confidence level of the target semantic understanding, the interaction intent, and the contextual logical relationship. Otherwise, a preset safety operation instruction based on a preset confidence threshold will be generated; Based on the target programming logic operation sequence or the preset security operation instructions, the native instruction set executable by the RPA drive engine is generated, and the native instruction set is integrated into the currently running automated business process as a dynamically generated sub-process. When the sub-process fails, a rollback operation is performed according to the recorded context, and recovery from the breakpoint or transfer to the manual intervention process is supported. The undefined graphical user interface elements are processed.

[0013] This invention provides an integrated office system based on an AI assistant, comprising: The intent recognition module is used to receive natural language commands input by the user on the workbench and to identify the user's intent by parsing the natural language commands with the AI ​​assistant. The permission verification module is used to determine the user's pending operation based on the user's intent, and to verify the user's permissions to determine the authorized applications that the user can access. The operation execution module is used to automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user terminal.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: By receiving natural language commands and automatically parsing user intent, intelligently matching accessible applications based on permission verification results, achieving seamless authentication with the help of single sign-on technology, scheduling an automated process engine to execute specific operations, and finally feeding back the results to the user, this effectively solves the problems of cumbersome switching between multiple systems, complex operation paths, and fragmented permission management. It realizes intelligent interaction and business automation under a unified entry point, significantly improving office efficiency and security.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] 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 flowchart illustrating an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the overall architecture and data flow of an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the permission verification principle in an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Figure 4 This is a schematic diagram of the SSO authentication principle in an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Figure 5 This is a diagram illustrating the automated execution framework of RPA in an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Figure 6 This is an example diagram illustrating the execution flow of the RPA automated execution framework in an integrated office method based on an AI assistant, as described in an embodiment of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] Example 1: This example provides an integrated office method based on an AI assistant, such as... Figure 1 As shown, it includes: Step 1: Receive natural language commands input by the user based on the workbench, and use the AI ​​assistant to parse the natural language commands and identify the user's intent; Step 2: Determine the user's operation to be performed based on the user's intent, and verify the user's permissions to determine the user's authorized applications; Step 3: Automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user's terminal.

[0020] In this embodiment, the workbench refers to a client platform that provides a unified application entry point and interactive interface.

[0021] In this embodiment, the AI ​​assistant refers to an intelligent interaction module with the ability to understand natural language, recognize intent, and perform application operations.

[0022] In this embodiment, natural language instructions refer to operation requests that are input by the user through text or voice and conform to everyday expression habits.

[0023] In this embodiment, user intent refers to the structured operation description obtained by parsing natural language instructions, including the target application, operation type, and parameters.

[0024] In this embodiment, the operation to be executed refers to an executable instruction generated according to the user's intent, which contains specific application interface call information and parameters.

[0025] In this embodiment, permission verification refers to the authorization check process that determines a user's eligibility to access the target application based on the relationship between user roles and permission groups.

[0026] In this embodiment, an accessible authorized application refers to a specific application instance that a user is allowed to access after passing permission verification.

[0027] In this embodiment, SSO authentication refers to a single sign-on authentication mechanism that allows users to access multiple mutually trusted application systems after logging in once.

[0028] In this embodiment, the RPA driving engine refers to the core of robotic process automation execution that can simulate user operations and automatically execute business processes.

[0029] In this embodiment, a schematic diagram of the overall architecture and data flow of an integrated office method based on an AI assistant is shown below. Figure 2 As shown.

[0030] In this embodiment, the RPA automated execution framework diagram in an integrated office method based on an AI assistant is shown below. Figure 5 As shown.

[0031] In this embodiment, an example diagram of the execution flow of the RPA automated execution framework in an integrated office method based on an AI assistant is shown below. Figure 6 As shown.

[0032] The beneficial effects of the above technical solution are as follows: by receiving natural language instructions and automatically parsing user intent, intelligently matching accessible applications based on permission verification results, achieving seamless authentication with the help of single sign-on technology, scheduling an automated process engine to execute specific operations, and finally feeding back the results to the user, it effectively solves the problems of cumbersome switching between multiple systems, complex operation paths, and scattered permission management, realizes intelligent interaction and business automation under a unified entry point, and significantly improves office efficiency and security.

[0033] Example 2: Based on Example 1, this example provides an integrated office method based on an AI assistant. In step 1, the system receives natural language commands input by the user on the workbench and parses the natural language commands using the AI ​​assistant to identify the user's intent, including: The workbench-based interactive interface receives natural language commands input by the user, and the AI ​​assistant also receives natural language commands. The natural language instructions are preprocessed, and the resulting instruction vector sequence is input into a pre-defined intent understanding model. The preprocessing includes word segmentation and vectorization of the natural language instructions. Named entity recognition is performed on instruction vector sequences based on the intent understanding model, and key parameter entities related to operation behavior are extracted from natural language instructions based on the named entity recognition results; Meanwhile, the intent category of the instruction vector is identified based on the intent understanding model, and the preset operation category to which the natural language instruction belongs is determined based on the intent category identification result; Based on the key parameter entities extracted by named entity recognition and the preset operation categories of intent category recognition results, a structured description of user intent is generated.

[0034] In this embodiment, the intent understanding model adopts a neural network architecture that integrates BERT, bidirectional long short-term memory network and conditional random field, which can simultaneously execute a composite algorithm model of named entity recognition and intent classification.

[0035] In this embodiment, the instruction vector sequence refers to a numerical instruction representation that retains the original semantics and sequence relationship after being generated through word segmentation and vectorization.

[0036] In this embodiment, the key parameter entity refers to specific parameter information such as time, quantity, and type extracted from natural language instructions that are directly related to the operation behavior.

[0037] In this embodiment, the preset operation category refers to the set of operation types predefined by the system for classifying user intentions, including categories such as open, query, operate, and create.

[0038] In this embodiment, the user intent description refers to a machine-readable intent representation with a clear structure and semantics, generated based on key parameter entities and preset operation categories.

[0039] The beneficial effects of the above technical solution are as follows: by automatically segmenting and vectorizing natural language instructions, and combining an intent understanding model that integrates multi-level semantic analysis, key parameter entity extraction and operation intent classification are completed simultaneously, and finally a structured user intent description is generated. This significantly improves the accuracy and completeness of natural language understanding, and can effectively parse instructions containing complex parameters and multi-step operations, providing an accurate and reliable input foundation for subsequent permission verification and automated execution.

[0040] Example 3: Based on Example 2, this example provides an integrated office method based on an AI assistant, which preprocesses natural language commands, including: Load a pre-built Chinese word segmentation dictionary, and based on the Chinese word segmentation dictionary, use the maximum positive matching algorithm to segment natural language instructions into word segments, dividing continuous text into independent lexical unit sequences; The segmented word unit sequence is standardized, and the standardized word unit sequence is input into the pre-trained word embedding model; Based on the word embedding model, each word unit is mapped to a fixed-dimensional real vector, and based on the mapping, the real vectors corresponding to all word units are combined in the original sequence order to generate an instruction vector sequence.

[0041] In this embodiment, the Chinese word segmentation dictionary refers to a vocabulary set that includes general vocabulary and enterprise-specific terms.

[0042] In this embodiment, the standardization process includes converting full-width characters to half-width characters and converting English letters to lowercase.

[0043] In this embodiment, the vocabulary unit sequence refers to the set of independent words obtained after word segmentation and arranged in the original text order.

[0044] In this embodiment, the word embedding model refers to a pre-trained neural network model that maps lexical units to fixed-dimensional real vectors.

[0045] In this embodiment, the instruction vector sequence refers to a numerical sequence composed of word vectors arranged in the original text order, representing the complete semantics of the instruction.

[0046] The beneficial effects of the above technical solution are: by adopting word segmentation processing that combines domain dictionary with standardized text processing, and using word embedding model to generate vector representations that preserve semantics and sequence relationships, efficient and accurate structured conversion of natural language instructions is achieved, providing a standardized and machine-readable input foundation for subsequent deep semantic understanding.

[0047] Example 4: Building upon Example 2, this example provides an integrated office method based on an AI assistant, which performs named entity recognition on instruction vector sequences based on an intent understanding model, including: Obtain a pre-defined intent understanding model, wherein the intent understanding model adopts a neural network architecture that integrates BERT, bidirectional long short-term memory network and conditional random field; The obtained instruction vector sequence is input into the BERT encoding layer in the intent understanding model, and the instruction vector sequence is context-encoded based on the BERT encoding layer to generate a feature vector sequence containing contextual semantic information. The feature vector sequence is input into the bidirectional long short-term memory network layer in the intent understanding model, and the forward and backward semantic dependencies of the feature vector sequence are modeled based on the bidirectional long short-term memory network layer, outputting the context-enhanced feature vector corresponding to each lexical unit. The context-enhanced feature vector sequence is input into the conditional random field layer in the intent understanding model, and the probability distribution of each word unit belonging to a preset entity label is calculated based on the context-enhanced feature vector according to the conditional random field layer. The preset entity labels include time, type, and quantity. The globally optimal entity label sequence is determined based on the probability distribution, and the corresponding key parameter entities and types are extracted from the original natural language instructions based on the entity label sequence.

[0048] In this embodiment, the conditional random field layer uses the Viterbi algorithm to decode the probability distribution and outputs a globally optimal entity label sequence.

[0049] In this embodiment, the instruction vector sequence refers to the numerical representation sequence that preserves the word order relationship after the original natural language instructions have been processed by word segmentation and vectorization.

[0050] In this embodiment, the feature vector sequence refers to the numerical feature sequence generated after processing by the context-aware encoder, in which each lexical unit contains full-text semantic information.

[0051] In this embodiment, the context-enhanced feature vector refers to a deep semantic feature representation that is obtained by further learning through a bidirectional sequence model and simultaneously integrates the forward and backward contextual dependencies of the current word.

[0052] In this embodiment, the probability distribution refers to the set of probabilities that each word unit is likely to belong to each predefined entity label, calculated by the model.

[0053] In this embodiment, the entity label sequence refers to the optimal entity type label sequence assigned to each lexical unit in the input instruction by a global decoding algorithm.

[0054] In this embodiment, the key parameter entity and type refer to the specific parameter values ​​with clear semantic types (such as time, type, quantity) extracted from the original instruction based on the entity tag sequence.

[0055] The beneficial effects of the above technical solution are as follows: by adopting a multi-layer neural network architecture that integrates BERT, bidirectional long short-term memory network and conditional random field, context encoding, bidirectional semantic dependency modeling and global sequence labeling are performed in sequence, which realizes high-precision identification and extraction of key parameters such as time, type and quantity in instructions. It can effectively capture complex semantic relationships and long-distance dependencies in natural language, significantly improve the accuracy and robustness of entity recognition, and provide reliable structured parameter input for subsequent operation execution.

[0056] Example 5: Based on Example 1, this example provides an integrated office method based on an AI assistant. In step 2, the user's operation to be performed is determined based on the user's intent, and the user's permissions are verified to determine the user's authorized applications, including: The user intent is obtained and parsed to determine the target application identifier, operation type, and key parameter entities extracted from the natural language instructions. Based on the target application identifier, the pre-set application metadata database is queried to determine the application access information and associated permission group requirements corresponding to the target application identifier; At the same time, obtain the current user's identity identifier, and based on the user's identity identifier, query the role-based access control library to determine one or more roles to which the user's identity identifier belongs; Based on the user's role, query the role-permission relationship and determine the set of permission groups granted to the user based on the query results; Based on the operation type and key parameter entities, the corresponding verification rules are matched in the preset fine-grained permission policy library. The application's associated permission group requirements are matched and verified against the set of permission groups granted to the user. At the same time, based on the verification rules, the operation type and the attachment permissions of key parameter entities are verified and checked during the verification process. When the matching verification passes, the target application is identified as an accessible and authorized application.

[0057] In this embodiment, the application metadata database refers to a dedicated database that stores the access methods of various application systems (such as login address, interface URL) and their required permission attributes (such as required roles or permission groups).

[0058] In this embodiment, permission group requirements refer to a set of specific permissions that must be met when accessing or operating a target application.

[0059] In this embodiment, the role-based access control library refers to the authorization policy database that stores the correspondence between users and roles, as well as basic information about the roles.

[0060] In this embodiment, the role-permission association refers to the data set defined in the permission model that describes the mapping relationship between each role and the granted permission group.

[0061] In this embodiment, the permission group set refers to the collection of all permission groups granted to the current user, which is calculated by querying the role-permission association relationship.

[0062] In this embodiment, the operation type and the key parameter entities extracted from the natural language command refer to the core action category (such as query, create, modify) and the specific object information that the action acts upon, identified by the system after parsing the natural language command input by the user. For example, in the command "View Zhang San's contract", the operation type is "View", and the key parameter entities are "contract" and "Zhang San".

[0063] In this embodiment, matching the corresponding verification rule in a pre-set fine-grained permission policy library based on the operation type and key parameter entities means that the system uses the identified operation type and specific parameters as query conditions to search in a pre-set policy library containing detailed permission rules. The purpose is to find permission check standards that match the current specific operation scenario, such as "only allow viewing of contracts within this department" or "prohibit modification of orders with an amount greater than 1 million".

[0064] In this embodiment, verifying and checking the operation type and attachment permissions of key parameter entities based on verification rules during the verification process means that when performing permission verification, it is not only necessary to verify whether the user has basic access rights to the application, but also to conduct a deep review of the actions and parameters involved in this operation according to the specific rules matched. For example, verifying whether the user has the right to perform the "export" operation, or checking whether the "customer list" parameter that the user wants to access is within the scope of their authorized data.

[0065] The beneficial effects of the above technical solution are: by dynamically parsing user intent and associating it with application permission requirements, and by combining user roles and permission group relationships for multi-layered matching and verification, the permission verification process is deeply coupled with business intent; it can accurately define the scope of user operations in different applications, effectively avoiding the problems of over-allocation or insufficient verification of permissions commonly found in traditional permission management, thereby improving the security of automated processes while ensuring the accurate execution of business operations.

[0066] Example 6: Building upon Example 5, this example provides an integrated office method based on an AI assistant, which determines the target application identifier, operation type, and key parameter entities extracted from natural language instructions corresponding to the user's intent, including: The target application identifier, operation type, and key parameter entities extracted from natural language instructions are obtained, and a pre-set operation category mapping table is queried based on the operation type to determine the operation execution interface category corresponding to the operation type. Based on the target application identifier, query the application metadata database to determine the application access interface information corresponding to the target application identifier; Generate an initial operation command template based on the operation execution interface category and application access interface information; Extract the operation parameters corresponding to the operation instruction to be executed, extract the necessary parameter values ​​that match the initial operation command template from the operation parameters, and fill the necessary parameter values ​​into the parameter slots of the initial operation command template to generate a complete operation instruction to be executed. The operation instruction to be executed is associated with the target application identifier, and the output is a schedulable operation context to be executed.

[0067] In this embodiment, the operation category mapping table refers to a configuration table that stores the correspondence between operation types described in natural language and the underlying execution interface types of the system.

[0068] In this embodiment, the application access interface information refers to the description of the technical access specifications provided by the target application to support external calls, including interface address, protocol type, and authentication method.

[0069] In this embodiment, the initial operation command template refers to a standardized operation command framework that contains a fixed structure and parameter placeholders but is not filled with specific values.

[0070] In this embodiment, the necessary parameter value refers to the specific data value that must be filled in in order to make the operation command template executable, as parsed from the user instruction.

[0071] In this embodiment, the parameter slot refers to a predefined placeholder mark in the operation command template used to receive and fill specific parameter values.

[0072] In this embodiment, the operation context to be executed refers to a standardized task data package that contains the target application identifier and complete operation instructions and can be directly invoked by subsequent processes.

[0073] In this embodiment, the operation parameters refer to the execution parameters corresponding to the specific behavior in the operation instruction to be executed, including the business type parameters to be processed and the specific execution purpose corresponding to the business.

[0074] In this embodiment, the principle diagram of permission verification is combined with that of embodiments 5 and 6, as follows: Figure 3 As shown.

[0075] The beneficial effects of the above technical solution are: by automatically mapping user intent to standardized operation interfaces and dynamically generating complete operation instructions, the accurate conversion of natural language instructions to system executable commands is achieved; the complexity of constructing operation instructions can be effectively reduced, execution failures caused by manual configuration errors can be avoided, and the standardization and reliability of operation instructions are ensured through the parameterized template mechanism, which significantly improves the construction efficiency and execution accuracy of automated processes.

[0076] Example 7: Based on Example 1, this example provides an integrated office method based on an AI assistant. In step 3, the system automatically logs in to authorized applications based on SSO authentication, schedules the RPA-driven engine to automate business processes according to the user's pending operations, and returns the execution results to the user terminal, including: The system obtains the user's pending operations and authorized applications, parses the pending operations to determine the corresponding pending operation instructions, and extracts the target application identifier of the authorized applications. The SSO authentication module obtains a valid access token for the authorized application based on the target application identifier, and obtains the authentication session status of the network client based on the SSO authentication module. Extract the original session token from the authentication session state and convert it into a unified security token based on preset conversion rules; Extract the configuration parameters of the authorized applications and determine the access mode configured for the authorized applications based on the configuration parameters; When the access mode is H5 container mode, the unified security token is entered into the HTTP request header of the H5 container, and access to the authorized application is initiated based on the entry result. When the access mode is configured as browser mode, a deep link URL carrying a unified security token is constructed, and the deep link URL is opened based on the preset default browser. Once an authentication session with an authorized application is successfully established, the operation command to be executed is sent to the RPA driver engine, and the RPA driver engine is controlled to match the corresponding automated execution script from the preset script library according to the operation command to be executed. If no match is found, an executable script fragment is generated in real time based on the operation type, parameter structure and target application interface characteristics of the operation command to be executed. Extract the operation parameters corresponding to the operation instructions to be executed, and dynamically input the operation parameters into the matching automated execution script based on the RPA driving engine; The automated execution script is driven by the RPA driver engine after the operation parameters are entered. The execution status is monitored in real time and a fault-tolerant rollback mechanism is triggered to restore to the previous stable state when errors or exceptions occur. The output data during the execution process is obtained. After the automated script finishes execution, the output data during the execution process is summarized and the summarized output data is formatted. The formatted output data is returned to the user terminal.

[0077] In this embodiment, obtaining the authentication session status of the network client based on the SSO authentication module includes: The SSO authentication module is registered as a status listener for network clients; When a user completes network authentication, the network client generates an authentication success event and broadcasts it. The SSO authentication module intercepts the authentication success event and extracts the original session token from the event parameters; Based on pre-configured token conversion rules, the original session token is converted into a standardized intermediate token; Based on the authentication protocol type of the target application, the intermediate token is formatted and encapsulated. The encapsulated access token is stored in the token cache library, and its issuance time and validity period are recorded.

[0078] In this embodiment, the operation instruction to be executed refers to an executable instruction generated according to the user's intent, which contains specific application operation commands and parameters.

[0079] In this embodiment, the target application identifier refers to an identifier used to uniquely identify an enterprise's internal application system.

[0080] In this embodiment, the unified security token refers to a standardized security authentication credential suitable for application layer access.

[0081] In this embodiment, a deep link URL refers to a hyperlink address that carries authentication token parameters and can directly open a specific page of the target application.

[0082] In this embodiment, the automated execution script refers to a predefined set of instructions used to simulate user operations to complete a specific business process in the target application.

[0083] In this embodiment, the token cache library refers to a temporary storage area used to store valid access tokens and their metadata.

[0084] In this embodiment, the SSO authentication principle diagram is as follows: Figure 4 As shown.

[0085] In this embodiment, the unified security token refers to the unified security token that converts the original session token into a unified security token with an additional digital signature, timestamp, and random salt value according to a preset conversion rule.

[0086] In this embodiment, if no match is found, the system will automatically assemble and generate an executable script fragment based on the operation type, parameter structure, and target application interface characteristics of the operation instruction to be executed. This means that when there is no existing script in the preset script library, the system will automatically assemble and generate a program code fragment that can complete the specific task according to the action type of this operation, the combination of required parameters, and the interaction interface specifications of the target application, so as to realize dynamic automated process creation.

[0087] In this embodiment, real-time monitoring of the execution status and triggering a fault-tolerant rollback mechanism to restore to the previous stable state in the event of errors or anomalies refers to continuously monitoring the execution progress and results of the automated script during its execution. Once an error or unexpected anomaly is detected, a preset fault-tolerant handling procedure is automatically initiated to roll back the relevant application data and system state to the correct point in time before the start of this operation, thereby ensuring the atomicity of the business process and data consistency.

[0088] In this embodiment, the configuration parameters of the accessible authorized application refer to data related to the performance of the accessible authorized application itself, such as the processing rate of the accessible authorized application.

[0089] The beneficial effects of the above technical solution are as follows: by automatically acquiring and converting network authentication tokens, intelligently injecting tokens according to application access patterns to achieve seamless login, and then scheduling the automation engine to match and execute parameterized scripts, monitoring the execution status in real time and returning formatted results, the full-process automation of cross-system authentication and business operations is realized, effectively eliminating manual login and operation links, significantly improving task execution efficiency and system security, while ensuring the reliability and traceability of the operation process.

[0090] Example 8: Based on Example 7, this example provides an integrated office method based on an AI assistant. It uses an SSO authentication module to obtain a valid access token for authorized applications based on the target application identifier, including: The SSO authentication module queries the application metadata database based on the target application identifier and obtains the authentication protocol type and token generation parameters of the authorized application based on the query results. Based on the authentication protocol type and token generation parameters, the system queries the preset token cache library to determine whether there is an unexpired access token corresponding to the current user session and the authorized applications that can be accessed. If an unexpired access token exists, return a valid access token that allows access to the authorized application from the preset token cache. When no expired access token exists, a token issuance request is sent to the central authentication server based on the SSO authentication module. The token issuance request includes the user identity identifier and the target application identifier. The central authentication server parses the token issuance request and builds a service connection with the authentication source based on the parsing result. The service connection control authentication source verifies the user's identity and sends an authentication response back to the central authentication server after successful verification. When the authentication response is successful, the central authentication server generates token claim data containing user identity information, target application identifier and timestamp, and digitally signs the token claim data based on a preset encryption key to obtain a valid access token that can access the authorized application. The obtained valid access token is returned to the SSO authentication module.

[0091] In this embodiment, the application metadata database refers to the configuration database that stores the application system's authentication protocol types and token generation parameters.

[0092] In this embodiment, the token cache library refers to a temporary storage area used to store valid access tokens and their metadata.

[0093] In this embodiment, the central authentication server refers to the core service component that is uniformly responsible for identity verification and token issuance.

[0094] In this embodiment, the authentication source refers to the underlying system that provides user authentication services, including Active Directory domains, LDAP directories, etc.

[0095] In this embodiment, token claim data refers to an unsigned token data body containing information such as user identity, target application, and timestamp.

[0096] The beneficial effects of the above technical solution are: by using intelligent query token caching and dynamic issuance mechanism, digital signature tokens are generated on demand after verifying the legitimacy of the user's identity, realizing unified management and secure distribution of access tokens, effectively avoiding repeated authentication operations, ensuring the security of cross-application access, and significantly improving the response efficiency and user experience of the authentication process.

[0097] Example 9: Based on Example 1, this example provides an integrated office method based on an AI assistant. Step 3, which schedules the RPA-driven engine to automate business processes, also includes: Based on the AI ​​assistant's real-time collection of target semantic understanding confidence of currently undefined graphical user interface elements; The target semantic understanding confidence level is compared with a preset confidence threshold. At the same time, the interaction intent and contextual logic relationship of the currently undefined graphical user interface elements are obtained. When the confidence level of the target semantic understanding is greater than or equal to the preset confidence threshold, the target programming logic operation sequence is generated based on the confidence level of the target semantic understanding, the interaction intent, and the contextual logical relationship. Otherwise, a preset safety operation instruction based on a preset confidence threshold will be generated; Based on the target programming logic operation sequence or the preset security operation instructions, the native instruction set executable by the RPA drive engine is generated, and the native instruction set is integrated into the currently running automated business process as a dynamically generated sub-process. When the sub-process fails, a rollback operation is performed according to the recorded context, and recovery from the breakpoint or transfer to the manual intervention process is supported. The undefined graphical user interface elements are processed.

[0098] In this embodiment, during the execution of the above content, for example, when a company employee is performing an expense reimbursement process, the user issues a command to the AI ​​assistant via natural language, "Reimburse my travel expenses from yesterday." The AI ​​assistant has interpreted the user's intent and logged into the financial system through permission verification and SSO. The RPA-driven engine is automatically executing a series of predefined reimbursement operations, such as navigating to the reimbursement page and filling out forms. However, if the system suddenly pops up an undefined warning pop-up during execution, and this pop-up is not within the RPA's preset process script, the pop-up prompts: "Duplicate reimbursement application for the same trip detected. Please confirm whether to continue submitting?" and provides two buttons: "Continue Submitting" and "Cancel." At this time, the RPA-driven engine faces the risk of interruption because it cannot recognize this pop-up. The present invention activates the following response mechanism: The AI ​​assistant (with its built-in computer vision and DOM parser) collects the target semantic understanding confidence level of currently undefined graphical user interface elements (capturing corresponding warning pop-ups) in real time (semantically understanding the text content of the pop-up to identify it as a "confirmation prompt" about "duplicate reimbursement"). Simultaneously, the AI ​​assistant outputs a quantified target semantic understanding confidence level, such as 92% (because the pop-up text is clear and the intent is explicit). The target semantic understanding confidence level (e.g., 92%) is compared with the preset confidence threshold (e.g., 85%). At the same time, the interaction intent and contextual logic relationship of the currently undefined graphical user interface element are obtained (i.e., the pop-up is a modal dialog box that requires user decision, its appearance is directly related to the "travel expense reimbursement" business being processed, and its logic is to prevent possible duplicate submissions and require confirmation). When the target semantic understanding confidence level (e.g., 92%) is greater than or equal to the preset confidence threshold (85%, where the preset confidence threshold is pre-set based on the risk level of different business operations. For example, for high-risk operations, such as fund transfers and data deletion, a high confidence threshold of 95% is set; for medium-risk operations, such as data modification and submitting approval processes, a medium confidence threshold of 75% is set; and for low-risk operations, such as information queries, a low confidence threshold of 65% is set to encourage the system to make more adaptive attempts), a target programming logic operation sequence is generated based on the target semantic understanding confidence level, interaction intent, and contextual logical relationship. (The target programming logic operation sequence refers to a micro-script written by AI in real time, containing process control logic such as conditional judgments, used to instruct the RPA engine to cope with complex and unknown scenarios; it is an intelligent bridge connecting the AI's "cognitive" ability and the RPA's operational ability.) Otherwise, a preset safety operation instruction based on a pre-set confidence threshold is generated. (For comparison, suppose the pop-up window is blurry and the AI ​​assistant can only recognize it as an "unknown dialog box," with a semantic understanding confidence of only 70%, which is below the 85% threshold. In this case, the system will not risk generating complex logic but will instead activate a safety precaution strategy, i.e., generate a preset safety operation instruction based on a pre-set confidence threshold. This instruction might be: "Send a notification to the user terminal requesting manual intervention," or "Minimize the current window and record the error status," thereby avoiding incorrect operations on the system.) Based on the target programming logic operation sequence or the preset security operation instructions, a native instruction set executable by the RPA driver engine is generated (the native instruction set is, for example, a sequence of underlying API calls simulating mouse movement or clicking). This native instruction set is then integrated as a dynamically generated sub-process into the currently running automated business process to process the undefined graphical user interface elements. (Subsequently, the native instruction set is integrated as a dynamically generated sub-process into the currently running "expense reimbursement" automated business process. Specifically, the RPA engine will pause the current form filling task and execute this newly inserted sub-process for handling the "duplicate reimbursement confirmation pop-up"). Once the sub-process is completed, click the "Continue Submit" button to close the pop-up window. Then, the RPA engine resumes the execution of the previously paused main task of the reimbursement process, thereby ensuring the semantic correctness and uninterrupted continuation of the entire automated business process.

[0099] The beneficial effects of the above technical solution are: by introducing a confidence-based intelligent decision-making mechanism, the RPA system is able to perceive, understand and autonomously handle unknown scenarios during runtime, which fundamentally improves the robustness, adaptability and reliability of business process automation.

[0100] Example 10: This example provides an integrated office system based on an AI assistant, including: The intent recognition module is used to receive natural language commands input by the user on the workbench and to identify the user's intent by parsing the natural language commands with the AI ​​assistant. The permission verification module is used to determine the user's pending operation based on the user's intent, and to verify the user's permissions to determine the authorized applications that the user can access. The operation execution module is used to automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user terminal.

[0101] The beneficial effects of the above technical solution are as follows: by receiving natural language commands and automatically parsing user intent, intelligently matching accessible applications based on permission verification results, achieving seamless authentication with the help of single sign-on technology, scheduling an automated process engine to execute specific operations, and finally feeding back the results to the user, it effectively solves the problems of cumbersome switching between multiple systems, complex operation paths, and scattered permission management, realizes intelligent interaction and business automation under a unified entry point, significantly improves office efficiency and security, and has good scalability and ease of use.

[0102] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An integrated office method based on an AI assistant, characterized in that, include: Step 1: Receive natural language commands input by the user based on the workbench, and use the AI ​​assistant to parse the natural language commands and identify the user's intent; Step 2: Determine the user's operation to be performed based on the user's intent, and verify the user's permissions to determine the user's authorized applications; Step 3: Automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user's terminal.

2. The integrated office method based on an AI assistant according to claim 1, characterized in that, In step 1, the system receives natural language commands input by the user from the workbench and uses an AI assistant to parse these commands and identify the user's intent, including: The workbench-based interactive interface receives natural language commands input by the user, and the AI ​​assistant also receives natural language commands. The natural language instructions are preprocessed, and the resulting instruction vector sequence is input into a pre-defined intent understanding model. The preprocessing includes word segmentation and vectorization of the natural language instructions. Named entity recognition is performed on instruction vector sequences based on the intent understanding model, and key parameter entities related to operation behavior are extracted from natural language instructions based on the named entity recognition results; Meanwhile, the intent category of the instruction vector is identified based on the intent understanding model, and the preset operation category to which the natural language instruction belongs is determined based on the intent category identification result; Based on the key parameter entities extracted by named entity recognition and the preset operation categories of intent category recognition results, a structured description of user intent is generated.

3. The integrated office method based on an AI assistant according to claim 2, characterized in that, Preprocessing of natural language instructions includes: Load a pre-built Chinese word segmentation dictionary, and based on the Chinese word segmentation dictionary, use the maximum positive matching algorithm to segment natural language instructions into word segments, dividing continuous text into independent lexical unit sequences; The segmented word unit sequence is standardized, and the standardized word unit sequence is input into the pre-trained word embedding model; Based on the word embedding model, each word unit is mapped to a fixed-dimensional real vector, and based on the mapping, the real vectors corresponding to all word units are combined in the original sequence order to generate an instruction vector sequence.

4. The integrated office method based on an AI assistant according to claim 2, characterized in that, Named entity recognition based on intent understanding model of instruction vector sequence, including: Obtain a pre-defined intent understanding model, wherein the intent understanding model adopts a neural network architecture that integrates BERT, bidirectional long short-term memory network and conditional random field; The obtained instruction vector sequence is input into the BERT encoding layer in the intent understanding model, and the instruction vector sequence is context-encoded based on the BERT encoding layer to generate a feature vector sequence containing contextual semantic information. The feature vector sequence is input into the bidirectional long short-term memory network layer in the intent understanding model, and the forward and backward semantic dependencies of the feature vector sequence are modeled based on the bidirectional long short-term memory network layer, outputting the context-enhanced feature vector corresponding to each lexical unit. The context-enhanced feature vector sequence is input into the conditional random field layer in the intent understanding model, and the probability distribution of each word unit belonging to a preset entity label is calculated based on the context-enhanced feature vector according to the conditional random field layer. The preset entity labels include time, type, and quantity. The globally optimal entity label sequence is determined based on the probability distribution, and the corresponding key parameter entities and types are extracted from the original natural language instructions based on the entity label sequence.

5. The integrated office method based on an AI assistant according to claim 1, characterized in that, In step 2, the user's intended action is determined based on the user's intent, and the user's permissions are verified to determine the user's authorized applications, including: The user intent is obtained and parsed to determine the target application identifier, operation type, and key parameter entities extracted from the natural language instructions. Based on the target application identifier, the pre-set application metadata database is queried to determine the application access information and associated permission group requirements corresponding to the target application identifier; At the same time, obtain the current user's identity identifier, and based on the user's identity identifier, query the role-based access control library to determine one or more roles to which the user's identity identifier belongs; Based on the user's role, query the role-permission relationship and determine the set of permission groups granted to the user based on the query results; Based on the operation type and key parameter entities, the corresponding verification rules are matched in the preset fine-grained permission policy library. The application's associated permission group requirements are matched and verified against the set of permission groups granted to the user. At the same time, based on the verification rules, the operation type and the attachment permissions of key parameter entities are verified and checked during the verification process. When the matching verification passes, the target application is identified as an accessible and authorized application.

6. The integrated office method based on an AI assistant according to claim 5, characterized in that, Determine the target application identifier, operation type, and key parameter entities extracted from the natural language instructions corresponding to the user intent, including: The target application identifier, operation type, and key parameter entities extracted from natural language instructions are obtained, and a pre-set operation category mapping table is queried based on the operation type to determine the operation execution interface category corresponding to the operation type. Based on the target application identifier, query the application metadata database to determine the application access interface information corresponding to the target application identifier; Generate an initial operation command template based on the operation execution interface category and application access interface information; Extract the operation parameters corresponding to the operation instruction to be executed, extract the necessary parameter values ​​that match the initial operation command template from the operation parameters, and fill the necessary parameter values ​​into the parameter slots of the initial operation command template to generate a complete operation instruction to be executed. The operation instruction to be executed is associated with the target application identifier, and the output is a schedulable operation context to be executed.

7. The integrated office method based on an AI assistant according to claim 1, characterized in that, In step 3, SSO authentication is used to automatically log in to authorized applications, and the RPA-driven engine is used to automate business processes based on the user's pending operations. The execution results are then returned to the user's terminal, including: The system obtains the user's pending operations and authorized applications, parses the pending operations to determine the corresponding pending operation instructions, and extracts the target application identifier of the authorized applications. The SSO authentication module obtains a valid access token for the authorized application based on the target application identifier, and obtains the authentication session status of the network client based on the SSO authentication module. Extract the original session token from the authentication session state and convert it into a unified security token based on preset conversion rules; Extract the configuration parameters of the authorized applications and determine the access mode configured for the authorized applications based on the configuration parameters; When the access mode is H5 container mode, the unified security token is entered into the HTTP request header of the H5 container, and access to the authorized application is initiated based on the entry result. When the access mode is configured as browser mode, a deep link URL carrying a unified security token is constructed, and the deep link URL is opened based on the preset default browser. Once an authentication session with an authorized application is successfully established, the operation command to be executed is sent to the RPA driver engine, and the RPA driver engine is controlled to match the corresponding automated execution script from the preset script library according to the operation command to be executed. If no match is found, an executable script fragment is generated in real time based on the operation type, parameter structure and target application interface characteristics of the operation command to be executed. Extract the operation parameters corresponding to the operation instructions to be executed, and dynamically input the operation parameters into the matching automated execution script based on the RPA driving engine; The automated execution script is driven by the RPA driver engine after the operation parameters are entered. The execution status is monitored in real time and a fault-tolerant rollback mechanism is triggered to restore to the previous stable state when errors or exceptions occur. The output data during the execution process is obtained. After the automated script finishes execution, the output data during the execution process is summarized and the summarized output data is formatted. The formatted output data is returned to the user terminal.

8. The integrated office method based on an AI assistant according to claim 7, characterized in that, The SSO authentication module obtains a valid access token for authorized applications based on the target application identifier, including: The SSO authentication module queries the application metadata database based on the target application identifier and obtains the authentication protocol type and token generation parameters of the authorized application based on the query results. Based on the authentication protocol type and token generation parameters, the system queries the preset token cache library to determine whether there is an unexpired access token corresponding to the current user session and the authorized applications that can be accessed. If an unexpired access token exists, return a valid access token that allows access to the authorized application from the preset token cache. When no expired access token exists, a token issuance request is sent to the central authentication server based on the SSO authentication module. The token issuance request includes the user identity identifier and the target application identifier. The central authentication server parses the token issuance request and builds a service connection with the authentication source based on the parsing result. The service connection control authentication source verifies the user's identity and sends an authentication response back to the central authentication server after successful verification. When the authentication response is successful, the central authentication server generates token claim data containing user identity information, target application identifier and timestamp, and digitally signs the token claim data based on a preset encryption key to obtain a valid access token that can access the authorized application. The obtained valid access token is returned to the SSO authentication module.

9. The integrated office method based on an AI assistant according to claim 1, characterized in that, Step 3, which schedules the RPA-driven engine to automate business processes, also includes: Based on the AI ​​assistant's real-time collection of target semantic understanding confidence of currently undefined graphical user interface elements; The target semantic understanding confidence level is compared with a preset confidence threshold. At the same time, the interaction intent and contextual logic relationship of the currently undefined graphical user interface elements are obtained. When the confidence level of the target semantic understanding is greater than or equal to the preset confidence threshold, the target programming logic operation sequence is generated based on the confidence level of the target semantic understanding, the interaction intent, and the contextual logical relationship. Otherwise, a preset safety operation instruction based on a preset confidence threshold will be generated; Based on the target programming logic operation sequence or the preset security operation instructions, the native instruction set executable by the RPA drive engine is generated, and the native instruction set is integrated into the currently running automated business process as a dynamically generated sub-process. When the sub-process fails, a rollback operation is performed according to the recorded context, and recovery from the breakpoint or transfer to the manual intervention process is supported. The undefined graphical user interface elements are processed.

10. An integrated office system based on an AI assistant, characterized in that, include: The intent recognition module is used to receive natural language commands input by the user on the workbench and to identify the user's intent by parsing the natural language commands with the AI ​​assistant. The permission verification module is used to determine the user's pending operation based on the user's intent, and to verify the user's permissions to determine the authorized applications that the user can access. The operation execution module is used to automatically log in to authorized applications based on SSO authentication, schedule the RPA-driven engine to automate business processes according to the user's pending operations, and return the execution results to the user terminal.

Images