A natural language-based cross-platform task intelligent collaboration method, device and storage medium

Abstract

The present invention proposes a natural language-based cross-platform task intelligent collaboration method, device, and storage medium. The method includes: receiving natural language operation instructions input by a user, parsing the natural language operation instructions to obtain multiple tool operation instructions, and composing the multiple tool operation instructions into a tool call chain, wherein the multiple tool operation instructions are cross-platform operation instructions; calling the Web application API, local application, and / or cloud service corresponding to the multiple cross-platform tool operation instructions in the cloud based on the sequence of the tool call chain to perform corresponding operations; monitoring the execution status of the multiple cross-platform tool operation instructions in real time, issuing a warning message when the execution status is abnormal, and intelligently selecting an alternative solution for execution or switching to human-computer collaborative execution based on the type of abnormality, and displaying the execution status in real time on a WEBUI interface. The present invention improves the execution efficiency of natural language operation instructions.

Description

Technical Field

[0001] The present invention relates to the technical field of the integration of artificial intelligence and automated data processing, and specifically to a natural language-based cross-platform task intelligent collaboration method, device and storage medium. Background Art

[0002] Common methods for automatic processing of human-computer interaction in the prior art include:

[0003] The shortcomings of traditional RPA (Robotic Process Automation) systems are:

[0004] ◦ Reliance on preset process rules (such as UiPath and Automation Anywhere), requiring professional developers to write scripts and unable to respond to dynamic user needs in real time;

[0005] ◦ Cross-system data interaction requires manual definition of data mapping rules (such as field matching from Excel to ERP system), and cannot automatically adapt to data structure changes;

[0006] Typical case: A company needs to manually export data from CRM and import it into BI tools to generate reports every month, which takes up to 8 hours per month.

[0007] Conversational AI Tools:

[0008] Existing products (such as the ChatGPT plug-in and Microsoft Copilot) only support single in-app operations (such as document generation or email composition) and lack cross-platform task orchestration capabilities.

[0009] ◦ Users need to manually switch between multiple tools. For example, they need to generate an Excel chart through a dialog and then manually copy it into PowerPoint, which is inefficient and prone to errors.

[0010] The primary cause of these flaws is the lack of a standardized mapping mechanism between the semantic output of the natural language understanding (NLU) module and tool API calls. For example, a user command like "Compare last year's sales data with this year's" needs to be translated into a chain of multiple actions, including CRM data query, Excel calculation, and visualization tool rendering. Existing technology cannot automatically generate a coherent command. Summary of the Invention

[0011] The present invention aims to solve one or more technical deficiencies in the above-mentioned prior art and proposes the following technical solutions.

[0012] A cross-platform task intelligent collaboration method based on natural language, the method comprising:

[0013] a conversion step of receiving a natural language operation instruction input by a user, parsing the natural language operation instruction to obtain a plurality of tool operation instructions, and combining the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions;

[0014] A scheduling step, in the cloud, calling the Web application APIs, local applications and / or cloud services corresponding to the multiple cross-platform tool operation instructions based on the order of the tool call chain to perform corresponding operations;

[0015] The monitoring step monitors the execution status of multiple cross-platform tool operation instructions in real time, issues a warning message when the execution status is abnormal, and intelligently selects alternative execution options or switches to human-computer collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface.

[0016] Furthermore, the operation of the conversion step is: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instructions to obtain the action subject, object and constraint conditions of the operation instructions, and performing context dependency analysis on the natural language operation instructions through the knowledge graph to associate historical tasks, and outputting the user operation intention in JSON format; based on the user operation intention, select available tools that can realize user operations from the tool library, and dynamically evaluate the execution efficiency of the tool combination based on the Q-learning algorithm to select the tool combination ranked first in execution efficiency or use the trained graph neural network to determine the tool combination; based on the atomic capability description of each tool stored in the tool metadata database, convert the user operation intention into the operation instructions of each tool in the tool combination, perform dependency analysis on the operation instructions of each tool based on the action subject, object and constraint conditions, and organize the operation instructions of each tool into a tool call chain based on the result of the dependency analysis.

[0017] Furthermore, the graph of the graph neural network is constructed as follows: the function of implementing a tool operation instruction is used as a node of the graph neural network, and all historical execution parameters of the software that can implement the tool operation instruction are used as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions. For example, a tool operation instruction is to generate a chart, and the software corresponding to the tool operation instruction are excel, pdf, and ppt. The historical execution success rates and execution times corresponding to the three software are as follows. Therefore, the historical execution success rates and execution times of the three software constitute a vector as the feature vector of the node.

[0018] Furthermore, the characteristic value of each node in the graph of the graph neural network is calculated as follows:

[0019] + ;

[0020] in, represents the eigenvalue of the i-th node, Indicates the number of software available to execute the tool operation instructions corresponding to the i-th node, express The maximum value of the historical execution success rate of the software, express The minimum value of the historical execution time of the software, i≥2, ≥1.

[0021] Furthermore, the weight of the edge in the graph of the graph neural network is defined as: if there is a dependency relationship between the tool operation instructions corresponding to two nodes, the weight of the edge between the two nodes is 1, otherwise it is 0.

[0022] Furthermore, the JSON-LD (Linked Data) standard is used to define data models for cross-platform data operations, and when the target tool data structure changes, the mapping rule update is automatically triggered to achieve automatic conversion from the data source to the target tool data.

[0023] Furthermore, in the monitoring step, a graph database is used to record the task execution trajectory corresponding to each tool operation instruction, build a data lineage relationship graph, and store the execution status persistently.

[0024] Furthermore, the user can operate tools or adjust tool call chains in the WEBUI interface.

[0025] The present invention also proposes a cross-platform task intelligent collaboration device based on natural language, which includes:

[0026] a conversion unit, receiving a natural language operation instruction input by a user, parsing the natural language operation instruction to obtain a plurality of tool operation instructions, and combining the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions;

[0027] A scheduling unit, in the cloud, calls the Web application APIs, local applications, and / or cloud services corresponding to the plurality of cross-platform tool operation instructions based on the order of the tool call chain to perform corresponding operations;

[0028] The monitoring unit monitors the execution status of multiple cross-platform tool operation instructions in real time, and issues a warning message when the execution status is abnormal. It also intelligently selects alternative execution options or switches to human-computer collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface.

[0029] Furthermore, the operation of the conversion unit is as follows: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instructions to obtain the action subject, object and constraint conditions of the operation instructions, and performing context dependency analysis on the natural language operation instructions through the knowledge graph to associate historical tasks, and outputting the user operation intention in JSON format; based on the user operation intention, select available tools that can realize the user operation from the tool library, and dynamically evaluate the execution efficiency of the tool combination based on the Q-learning algorithm to select the tool combination ranked first in execution efficiency or use the trained graph neural network to determine the tool combination; based on the atomic capability description of each tool stored in the tool metadata database, convert the user operation intention into the operation instruction of each tool in the tool combination, perform dependency analysis on the operation instruction of each tool based on the action subject, object and constraint conditions, and organize the operation instructions of each tool into a tool call chain based on the result of the dependency analysis.

[0030] Furthermore, the graph of the graph neural network is constructed as follows: the function of implementing a tool operation instruction is used as a node of the graph neural network, and all historical execution parameters of the software that can implement the tool operation instruction are used as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions. For example, a tool operation instruction is to generate a chart, and the software corresponding to the tool operation instruction are excel, pdf, and ppt. The historical execution success rates and execution times corresponding to the three software are as follows. Therefore, the historical execution success rates and execution times of the three software constitute a vector as the feature vector of the node.

[0031] Furthermore, the characteristic value of each node in the graph of the graph neural network is calculated as follows:

[0032] + ;

[0033] in, represents the eigenvalue of the i-th node, Indicates the number of software available to execute the tool operation instructions corresponding to the i-th node, express The maximum value of the historical execution success rate of the software, express The minimum value of the historical execution time of the software, i≥2, ≥1.

[0034] Furthermore, the weight of the edge in the graph of the graph neural network is defined as: if there is a dependency relationship between the tool operation instructions corresponding to two nodes, the weight of the edge between the two nodes is 1, otherwise it is 0.

[0035] Furthermore, the JSON-LD (Linked Data) standard is used to define data models for cross-platform data operations, and when the target tool data structure changes, the mapping rule update is automatically triggered to achieve automatic conversion from the data source to the target tool data.

[0036] Furthermore, a graph database is used in the monitoring unit to record the task execution trajectory corresponding to each tool operation instruction, build a data lineage relationship graph, and store the execution status persistently.

[0037] Furthermore, the user can operate tools or adjust tool call chains in the WEBUI interface.

[0038] The present invention further provides a computer-readable storage medium, wherein the storage medium stores computer program code, and when the computer program code is executed by a computer, any one of the above methods is executed.

[0039] The technical effect of the present invention is: a natural language-based cross-platform task intelligent collaboration method, device and storage medium of the present invention, the method includes: a conversion step S101, receiving the natural language operation instructions input by the user, parsing the natural language operation instructions to obtain multiple tool operation instructions, and forming the multiple tool operation instructions into a tool call chain, wherein the multiple tool operation instructions are cross-platform operation instructions; a scheduling step S102, calling the Web application API, local application and / or cloud service corresponding to the multiple cross-platform tool operation instructions in the cloud based on the order of the tool call chain to perform corresponding operations; a monitoring step S103, real-time monitoring of the execution status of multiple cross-platform tool operation instructions, and issuing a warning message when the execution status is abnormal, and intelligently selecting an alternative solution to execute or switching to human-computer collaborative execution based on the type of abnormality, and displaying the execution status in real time on the WEBUI interface. To address the problem that operation instructions in existing AI products can only be executed in one application, and executing multiple instruction operations, especially cross-platform operations, requires manual operation, the present invention creatively proposes a cross-platform task intelligent collaboration method. First, the natural language operation instruction input by the user is parsed to obtain multiple tool operation instructions, and the multiple tool operation instructions are combined into a tool call chain, wherein the multiple tool operation instructions are cross-platform operation instructions. Then, based on the sequence of the tool call chain, the Web application API, local application and / or cloud service corresponding to the multiple cross-platform tool operation instructions are called in the cloud to perform the corresponding operations. The execution status of the multiple cross-platform tool operation instructions is monitored in real time, and a warning message is issued when the execution status is abnormal. Based on the type of abnormality, an alternative solution is intelligently selected for execution or switched to human-machine collaborative execution, and the execution status is displayed in real time on the WEBUI interface. That is, by constructing a cross-platform tool chain, the present invention realizes cross-platform tool calling, monitors the execution status of the tool operation instructions in real time, and intelligently selects an alternative solution for execution or switches to human-machine collaborative execution based on the type of abnormality, thereby improving the execution efficiency of natural language operation instructions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0041] Figure 1 This is a flowchart of a natural language-based cross-platform task intelligent collaboration method according to an embodiment of the present invention.

[0042] Figure 2 4 is a structural diagram of a natural language-based cross-platform task intelligent collaboration device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0044] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] Figure 1 A natural language-based cross-platform task intelligent collaboration method of the present invention is shown, and the method includes:

[0046] Conversion step S101: receiving a natural language operation instruction input by a user, parsing the natural language operation instruction to obtain a plurality of tool operation instructions, and combining the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions;

[0047] Scheduling step S102, calling the Web application APIs, local applications and / or cloud services corresponding to the plurality of cross-platform tool operation instructions based on the order of the tool call chain in the cloud to perform corresponding operations;

[0048] Monitoring step S103 monitors the execution status of multiple cross-platform tool operation instructions in real time, and issues a warning message when the execution status is abnormal, and intelligently selects alternative execution options or switches to human-computer collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface.

[0049] To address the problem that operation instructions in existing AI products can only be executed in one application, and executing multiple instruction operations, especially cross-platform operations, requires manual operation, the present invention creatively proposes a cross-platform task intelligent collaboration method. First, the natural language operation instruction input by the user is parsed to obtain multiple tool operation instructions, and the multiple tool operation instructions are combined into a tool call chain, wherein the multiple tool operation instructions are cross-platform operation instructions. Then, based on the order of the tool call chain, the Web application API, local application and / or cloud service corresponding to the multiple cross-platform tool operation instructions are called in the cloud to perform the corresponding operations. The execution status of the multiple cross-platform tool operation instructions is monitored in real time, and a warning message is issued when the execution status is abnormal. Based on the type of abnormality, an alternative solution is intelligently selected for execution or switched to human-machine collaborative execution, and the execution status is displayed in real time on the WEBUI interface. That is, by constructing a cross-platform tool chain, the present invention realizes cross-platform tool calling, and can monitor the execution status of the tool operation instruction in real time. It can also intelligently select an alternative solution for execution or switch to human-machine collaborative execution based on the type of abnormality, thereby improving the execution efficiency of natural language operation instructions. This is the important inventive concept of the present invention.

[0050] In the present invention, the intelligent selection of alternative solutions based on the type of anomaly can be, for example, if the originally selected execution software makes an error and reports an alarm, the optimal alternative solution can be selected based on the historical success rate. For example, if the original execution software of a tool instruction is Power BI, and an error occurs during the execution of the software, it can automatically switch to Excel with a historical success rate of 98% to generate charts. This is an innovation of the present invention, which can ensure that the user's natural language operation instructions are executed without interruption and ensure the correctness of the results. Of course, it is also possible to switch to the manual operation model after the alarm is issued and execute in a human-computer collaborative manner.

[0051] The execution status is displayed in real time on the WEBUI interface, which can be:

[0052] ▪ Data exported from CRM (30%)

[0053] ▪ Using Excel to generate charts (60%)

[0054] Prepare to send emails (80%)

[0055] ▪ If a step gets stuck (e.g. Excel crashes), the system will mark it in red as a warning message.

[0056] The key to the present invention is to determine a tool chain with high feasibility and high reliability. In one embodiment, the operation of the conversion step S101 is as follows: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instruction to obtain the action subject, object and constraint conditions of the operation instruction, and performing context dependency analysis on the natural language operation instruction through the knowledge graph to associate historical tasks, and outputting the user operation intention in JSON format, such as the user's original instruction ( For example, for an example scenario where a user attempts to perform an action, such as "Generate a Q3 sales report and email it to management," a Bi-LSTM+CRF model is used to identify the action subject ("Generate"), the object ("Sales Report"), and the constraints ("Q3," "Management"). Historical tasks are then associated through a knowledge graph (for example, "Sales Report" must be associated with a CRM data source and last month's report template). Based on the user's intended action, available tools that can implement the user's action are selected from a tool library. The execution efficiency of the tool combination is dynamically evaluated using a Q-learning algorithm to select the tool combination with the highest execution efficiency or to determine the tool combination using a trained graph neural network. The Q-learning algorithm is a commonly used combination evaluation algorithm and a relatively traditional method, resulting in fast execution but slightly lower effectiveness. Based on the atomic capability descriptions of each tool stored in the tool metadata database, the user's intended action is converted into operational instructions for each tool in the tool combination. Dependency analysis is performed on the operational instructions for each tool based on the action subject, object, and constraints. Based on the results of the dependency analysis, the operational instructions for each tool are organized into a tool call chain.

[0057] In order to improve the problem of slightly poor effect of the Q-learning algorithm, the present invention also proposes to use a trained graph neural network to determine the tool combination, so that the determined tool combination is more accurate, and then the user operation intention is converted into the operation instructions of each tool in the tool combination based on the atomic capability description of each tool stored in the tool metadata database. The operation instructions of each tool are subjected to dependency analysis based on the action subject, object and constraint conditions, and the operation instructions of each tool are combined into a tool call chain based on the result of the dependency analysis, that is, the tool call chain finally generated is the optimal tool combination, which improves the efficiency and accuracy of the execution of the natural language operation instructions input by the user. This is the important inventive concept of the present invention.

[0058] In one embodiment, the graph of the graph neural network is constructed as follows: the function of implementing a tool operation instruction is used as a node of the graph neural network, and all historical execution parameters of the software that can implement the tool operation instruction are used as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions. For example, a tool operation instruction is to generate a chart, and the software corresponding to the tool operation instruction are Excel, PDF, and PPT. The historical execution success rates and execution times corresponding to the three software are as follows. Therefore, the historical execution success rates and execution times of the three software constitute a vector as the feature vector of the node.

[0059] In one embodiment, the characteristic value of each node in the graph of the graph neural network is calculated as follows:

[0060] + ;

[0061] in, represents the eigenvalue of the i-th node, Indicates the number of software available to execute the tool operation instructions corresponding to the i-th node, express The maximum value of the historical execution success rate of the software, express The minimum value of the historical execution time of the software, i≥2, ≥1.

[0062] In one embodiment, the weight of an edge in a graph of a graph neural network is defined as follows: if there is a dependency relationship between the tool operation instructions corresponding to two nodes, the weight of the edge between the two nodes is 1, otherwise it is 0.

[0063] The present invention proposes a specific composition method of graph neural network, using a tool operation instruction as a node, and all the historical execution parameters of the software that can implement the tool operation instruction as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions, and the node feature value is calculated based on the historical execution success rate and execution time of the software, and a specific calculation formula is proposed. After simulation calculation, this method has an 8% higher execution success rate and 4% lower execution time than the tool combination determined by traditional Q-learning. This is another important inventive concept of the present invention.

[0064] In one embodiment, the JSON-LD (Linked Data) standard is used to define data schemas for cross-platform data operations. When the target tool's data structure changes, mapping rule updates are automatically triggered to automatically convert data from the data source to the target tool. This addresses the technical drawback of prior art, where differences in data formats across systems (e.g., CRM JSON data versus Excel spreadsheets) require manual processing, leading to an interruption rate of over 40% for automated processes.

[0065] In one embodiment, a graph database is used in the monitoring step to record the task execution trajectory corresponding to the operation instructions of each tool, build a data lineage relationship graph, and store the execution status persistently. That is, all operation records of the task (corresponding to a natural language operation instruction input by a user) are saved to facilitate future viewing or collaboration. For example, the system will record the source of each data. For example, a certain number in the bar chart comes from which module of the CRM, just like giving the data a "birth certificate." After each task is completed, the system will save a "snapshot." If someone in the team makes a mistake, they can restore to the previous version with one click. After the task is completed, the system automatically generates a report, which mainly includes: which tools were used, how long it took, whether there were any errors, and where the final results are saved.

[0066] In one embodiment, the user can operate the tool or adjust the tool call chain in the WEBUI interface. For example, if the user feels that the color of the bar chart is not good, the user can directly open Excel in the WEBUI to modify the color. The system will automatically save the modification and continue with the next step.

[0067] Figure 2 The present invention shows a natural language-based cross-platform task intelligent collaboration device, which includes:

[0068] The conversion unit 201 receives a natural language operation instruction input by a user, parses the natural language operation instruction to obtain a plurality of tool operation instructions, and combines the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions;

[0069] The scheduling unit 202 calls the Web application APIs, local applications and / or cloud services corresponding to the multiple cross-platform tool operation instructions based on the order of the tool call chain in the cloud to perform corresponding operations;

[0070] The monitoring unit 203 monitors the execution status of multiple cross-platform tool operation instructions in real time, and issues a warning message when the execution status is abnormal. It intelligently selects alternative execution plans or switches to human-computer collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface.

[0071] To address the problem that operation instructions in existing AI products can only be executed in one application, and executing multiple instruction operations, especially cross-platform operations, requires manual operation, the present invention creatively proposes a cross-platform task intelligent collaboration method. First, the natural language operation instruction input by the user is parsed to obtain multiple tool operation instructions, and the multiple tool operation instructions are combined into a tool call chain, wherein the multiple tool operation instructions are cross-platform operation instructions. Then, based on the order of the tool call chain, the Web application API, local application and / or cloud service corresponding to the multiple cross-platform tool operation instructions are called in the cloud to perform the corresponding operations. The execution status of the multiple cross-platform tool operation instructions is monitored in real time, and a warning message is issued when the execution status is abnormal. Based on the type of abnormality, an alternative solution is intelligently selected for execution or switched to human-machine collaborative execution, and the execution status is displayed in real time on the WEBUI interface. That is, by constructing a cross-platform tool chain, the present invention realizes cross-platform tool calling, and can monitor the execution status of the tool operation instruction in real time. It can also intelligently select an alternative solution for execution or switch to human-machine collaborative execution based on the type of abnormality, thereby improving the execution efficiency of natural language operation instructions. This is the important inventive concept of the present invention.

[0072] In the present invention, the intelligent selection of alternative solutions based on the type of anomaly can be, for example, if the originally selected execution software makes an error and reports an alarm, the optimal alternative solution can be selected based on the historical success rate. For example, if the original execution software of a tool instruction is Power BI, and an error occurs during the execution of the software, it can automatically switch to Excel with a historical success rate of 98% to generate charts. This is an innovation of the present invention, which can ensure that the user's natural language operation instructions are executed without interruption and ensure the correctness of the results. Of course, it is also possible to switch to the manual operation model after the alarm is issued and execute in a human-computer collaborative manner.

[0073] The execution status is displayed in real time on the WEBUI interface, which can be:

[0074] Data exported from CRM (30%)

[0075] ▪ Using Excel to generate charts (60%)

[0076] Prepare to send emails (80%)

[0077] ▪ If a step gets stuck (e.g. Excel crashes), the system will mark it in red as a warning message.

[0078] The key to the present invention is to determine a tool chain with high feasibility and high reliability. In one embodiment, the operation of the conversion unit 201 is as follows: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instruction to obtain the action subject, object and constraint conditions of the operation instruction, and performing context dependency analysis on the natural language operation instruction through the knowledge graph to associate historical tasks, and outputting the user operation intention in JSON format, such as the user's original instruction ( For example, for an example scenario where a user attempts to perform an action, such as "Generate a Q3 sales report and email it to management," a Bi-LSTM+CRF model is used to identify the action subject ("Generate"), the object ("Sales Report"), and the constraints ("Q3," "Management"). Historical tasks are then associated through a knowledge graph (for example, "Sales Report" must be associated with a CRM data source and last month's report template). Based on the user's intended action, available tools that can implement the user's action are selected from a tool library. The execution efficiency of the tool combination is dynamically evaluated using a Q-learning algorithm to select the tool combination with the highest execution efficiency or to determine the tool combination using a trained graph neural network. The Q-learning algorithm is a commonly used combination evaluation algorithm and a relatively traditional method, resulting in fast execution but slightly lower effectiveness. Based on the atomic capability descriptions of each tool stored in the tool metadata database, the user's intended action is converted into operational instructions for each tool in the tool combination. Dependency analysis is performed on the operational instructions for each tool based on the action subject, object, and constraints. Based on the results of the dependency analysis, the operational instructions for each tool are organized into a tool call chain.

[0079] In order to improve the problem of slightly poor effect of the Q-learning algorithm, the present invention also proposes to use a trained graph neural network to determine the tool combination, so that the determined tool combination is more accurate, and then the user operation intention is converted into the operation instructions of each tool in the tool combination based on the atomic capability description of each tool stored in the tool metadata database. The operation instructions of each tool are subjected to dependency analysis based on the action subject, object and constraint conditions, and the operation instructions of each tool are combined into a tool call chain based on the result of the dependency analysis, that is, the tool call chain finally generated is the optimal tool combination, which improves the efficiency and accuracy of the execution of the natural language operation instructions input by the user. This is the important inventive concept of the present invention.

[0080] In one embodiment, the graph of the graph neural network is constructed as follows: the function of implementing a tool operation instruction is used as a node of the graph neural network, and all historical execution parameters of the software that can implement the tool operation instruction are used as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions. For example, a tool operation instruction is to generate a chart, and the software corresponding to the tool operation instruction are Excel, PDF, and PPT. The historical execution success rates and execution times corresponding to the three software are as follows. Therefore, the historical execution success rates and execution times of the three software constitute a vector as the feature vector of the node.

[0081] In one embodiment, the characteristic value of each node in the graph of the graph neural network is calculated as follows:

[0082] + ;

[0083] in, represents the eigenvalue of the i-th node, Indicates the number of software available to execute the tool operation instructions corresponding to the i-th node, express The maximum value of the historical execution success rate of the software, express The minimum value of the historical execution time of the software, i≥2, ≥1.

[0084] In one embodiment, the weight of an edge in a graph of a graph neural network is defined as follows: if there is a dependency relationship between the tool operation instructions corresponding to two nodes, the weight of the edge between the two nodes is 1, otherwise it is 0.

[0085] The present invention proposes a specific composition method of graph neural network, using a tool operation instruction as a node, and all the historical execution parameters of the software that can implement the tool operation instruction as the feature vector of the node. If there is a dependency relationship between two tool operation instructions, there is an edge between the two tool operation instructions, and the node feature value is calculated based on the historical execution success rate and execution time of the software, and a specific calculation formula is proposed. After simulation calculation, this method has an 8% higher execution success rate and 4% lower execution time than the tool combination determined by traditional Q-learning. This is another important inventive concept of the present invention.

[0086] In one embodiment, the JSON-LD (Linked Data) standard is used to define data schemas for cross-platform data operations. When the target tool's data structure changes, mapping rule updates are automatically triggered to automatically convert data from the data source to the target tool. This addresses the technical drawback of prior art, where differences in data formats across systems (e.g., CRM JSON data versus Excel spreadsheets) require manual processing, leading to an interruption rate of over 40% for automated processes.

[0087] In one embodiment, a graph database is used in the monitoring step to record the task execution trajectory corresponding to the operation instructions of each tool, build a data lineage relationship graph, and store the execution status persistently. That is, all operation records of the task (corresponding to a natural language operation instruction input by a user) are saved to facilitate future viewing or collaboration. For example, the system will record the source of each data. For example, a certain number in the bar chart comes from which module of the CRM, just like giving the data a "birth certificate." After each task is completed, the system will save a "snapshot." If someone in the team makes a mistake, they can restore to the previous version with one click. After the task is completed, the system automatically generates a report, which mainly includes: which tools were used, how long it took, whether there were any errors, and where the final results are saved.

[0088] In one embodiment, the user can operate the tool or adjust the tool call chain in the WEBUI interface. For example, if the user feels that the color of the bar chart is not good, the user can directly open Excel in the WEBUI to modify the color. The system will automatically save the modification and continue with the next step.

[0089] Finally, it should be noted that although the embodiments of the present invention are described with respect to cross-platform and cross-software, the method can be fully applied to the execution of operating instructions within the same system.

[0090] In one embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored. When the computer program on the computer storage medium is executed by a processor, the above-mentioned method is implemented. The computer storage medium can be a hard disk, DVD, CD, flash memory or other memory.

[0091] For the convenience of description, the above device is described as being divided into various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0092] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the devices described in various embodiments of the present application or certain parts of the embodiments.

[0093] Finally, it should be noted that the above embodiments are only intended to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A cross-platform task intelligent collaboration method based on natural language, characterized by: The method includes: a conversion step of receiving a natural language operation instruction input by a user, parsing the natural language operation instruction to obtain a plurality of tool operation instructions, and combining the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions; A scheduling step, in the cloud, calling the Web application APIs, local applications and / or cloud services corresponding to the multiple cross-platform tool operation instructions based on the sequence of the tool call chain to perform corresponding operations; The monitoring step monitors the execution status of multiple cross-platform tool operation instructions in real time, issues a warning message when the execution status is abnormal, and intelligently selects alternative execution options or switches to human-machine collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface. Among them, the operation of the conversion step is: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instructions to obtain the action subject, object and constraint conditions of the operation instructions, and performing context dependency analysis on the natural language operation instructions through the knowledge graph to associate historical tasks, and output the user operation intention in JSON format; based on the user operation intention, select available tools that can realize user operations from the tool library, and dynamically evaluate the execution efficiency of the tool combination based on the Q-learning algorithm to select the tool combination with the highest execution efficiency or use the trained graph neural network to determine the tool combination; based on the atomic capability description of each tool stored in the tool metadata database, convert the user operation intention into the operation instructions of each tool in the tool combination, perform dependency analysis on the operation instructions of each tool based on the action subject, object and constraint conditions, and form the operation instructions of each tool into a tool call chain based on the result of the dependency analysis.

2. The method according to claim 1, characterized in that Use the JSON-LD (Linked Data) standard to define data models for cross-platform data operations. When the target tool data structure changes, the mapping rule update is automatically triggered to achieve automatic conversion from the data source to the target tool data.

3. The method according to claim 2, characterized in that In the monitoring step, a graph database is used to record the task execution trajectory corresponding to each tool operation instruction, build a data lineage relationship graph, and store the execution status persistently.

4. The method according to claim 3, characterized in that The user can operate the tool or adjust the tool call chain in the WEBUI interface.

5. A cross-platform task intelligent collaboration device based on natural language, characterized in that: The device includes: a conversion unit, receiving a natural language operation instruction input by a user, parsing the natural language operation instruction to obtain a plurality of tool operation instructions, and combining the plurality of tool operation instructions into a tool call chain, wherein the plurality of tool operation instructions are cross-platform operation instructions; A scheduling unit, in the cloud, calls the Web application APIs, local applications, and / or cloud services corresponding to the multiple cross-platform tool operation instructions based on the sequence of the tool call chain to perform corresponding operations; The monitoring unit monitors the execution status of multiple cross-platform tool operation instructions in real time, issues a warning message when the execution status is abnormal, and intelligently selects an alternative execution plan or switches to human-machine collaborative execution based on the type of abnormality, and displays the execution status in real time on the WEBUI interface. Among them, the operation of the conversion unit is: using the Bi-LSTM+CRF model to perform intent recognition on the natural language operation instructions to obtain the action subject, object and constraint conditions of the operation instructions, and performing context dependency analysis on the natural language operation instructions through the knowledge graph to associate historical tasks, and output the user operation intention in JSON format; based on the user operation intention, select available tools that can realize user operations from the tool library, and dynamically evaluate the execution efficiency of the tool combination based on the Q-learning algorithm to select the tool combination with the highest execution efficiency or use the trained graph neural network to determine the tool combination; based on the atomic capability description of each tool stored in the tool metadata database, convert the user operation intention into the operation instructions of each tool in the tool combination, perform dependency analysis on the operation instructions of each tool based on the action subject, object and constraint conditions, and organize the operation instructions of each tool into a tool call chain based on the result of the dependency analysis.

6. The device according to claim 5, characterized in that Use the JSON-LD (Linked Data) standard to define data models for cross-platform data operations. When the target tool data structure changes, the mapping rule update is automatically triggered to achieve automatic conversion from the data source to the target tool data.

7. The device according to claim 6, characterized in that A graph database is used in the monitoring unit to record the task execution trajectory corresponding to each tool operation instruction, build a data lineage relationship graph, and store the execution status persistently.

8. A computer storage medium, characterized in that The computer storage medium stores a computer program, and when the computer program on the computer storage medium is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images