A workflow automatic processing system and method based on a large language model

Abstract

The application discloses a kind of workflow automatic processing system and method based on large language model, it is related to computer technology field.The system includes: instruction recognition module, communication module and plug-in module, and large language model service is directed to natural language content and judges: when natural language content includes directional content and contains running instruction, natural language content is sent to instruction recognition module and is handled;When natural language content includes directional content and semantics is consistent with plug-in description, natural language content is sent to plug-in module and is handled;When natural language content does not include directional content, large language model service generates answer information according to natural language content and sends to dialog box and is displayed.The application only needs to carry out simple communication with large language model service, can realize the operation of workflow, greatly improves work efficiency while reducing learning cost.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to an automated workflow processing system and method based on a large language model. Background Technology

[0002] Against the backdrop of the rapid development of LLM (Large Language Model) technology, various large models have emerged both domestically and internationally. The number of parameters in these large models has increased dramatically compared to the past, resulting in more accurate natural language generation that closely resembles human thought. Furthermore, their knowledge bases cover various industries, demonstrating outstanding performance in human-computer dialogue. At the same time, most enterprises are using workflow software to handle their daily workflows. Currently, workflow creation and approval still rely on manual processes. However, these workflow software programs vary widely, and the types of work orders and submission steps offered by each software are quite different, significantly increasing the difficulty of learning and using them, thereby affecting work efficiency. Summary of the Invention

[0003] The present invention aims to address the problem that existing workflow management systems are difficult to learn and use, thus affecting work efficiency. In order to solve the above-mentioned technical problems, the present invention provides an automatic workflow processing system and method based on a large language model.

[0004] Firstly, a workflow automation system based on a large language model is provided, comprising: an instruction recognition module, a communication module, and a plug-in module. The communication module includes a large language model service and a dialog box. The dialog box is used to acquire natural language content input by the user and send it to the large language model service. The large language model service is used to perform directional judgment on the natural language content.

[0005] When the natural language content includes directional content and contains execution instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the execution information to obtain a first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the module calls the address of the corresponding service. The execution information is an instruction configuration file maintained by the instruction recognition module, and the instruction configuration file includes execution instructions and addresses with binding relationships.

[0006] When the natural language content includes directional content and semantically matches the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software and a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain a second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through the dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is invoked.

[0007] When the natural language content does not include directional content, the large language model service is used to generate answer information based on the natural language content and send it to the dialog box for display.

[0008] In one possible implementation of the first aspect, the address includes a service address and a backend interface address, the source of the running instruction, the service address, and the backend interface address is workflow software, and the instruction recognition module is specifically used to call the service corresponding to the running instruction through the service address or the backend interface address.

[0009] In one possible implementation of the first aspect, the instruction recognition module predefines a prompt template, and the instruction recognition module is specifically used to add the natural language content and the running information to the corresponding position of the prompt template to obtain a first prompt text.

[0010] In one possible implementation of the first aspect, the plugin module predefines a prompt template, and the plugin module is specifically used to add the natural language content and the plugin description to the corresponding position of the prompt template to obtain a second prompt text.

[0011] In one possible implementation of the first aspect, the plugin module is specifically used to manage the functions in the workflow software through a yml configuration file, which maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

[0012] Secondly, an automated workflow processing method based on a large language model is provided, including:

[0013] The dialog box retrieves the natural language content input by the user and sends it to the large language model service;

[0014] The large language model service performs a directional determination on the natural language content:

[0015] When the natural language content includes directional content and contains execution instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the execution information to obtain a first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the address of the corresponding service is called. The execution information is an instruction configuration file maintained by the instruction recognition module, and the instruction configuration file includes execution instructions and addresses with binding relationships.

[0016] When the natural language content includes directional content and semantically matches the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software and a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain a second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through the dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is invoked.

[0017] When the natural language content does not include directional information, the large language model service generates a response based on the natural language content and sends it to the dialog box for display.

[0018] In one possible implementation of the second aspect, the address includes a service address and a backend interface address. The source of the running instruction, the service address, and the backend interface address is workflow software. The instruction recognition module specifically calls the service corresponding to the running instruction through the service address or the backend interface address.

[0019] In one possible implementation of the second aspect, the instruction recognition module predefines a prompt template, and the instruction recognition module specifically adds the natural language content and the running information to the corresponding positions of the prompt template to obtain the first prompt text.

[0020] In one possible implementation of the second aspect, the plugin module predefines a prompt template, and the plugin module specifically adds the natural language content and the plugin description to the corresponding position of the prompt template to obtain the second prompt text.

[0021] One possible implementation of the second aspect also includes:

[0022] The plugin module manages the functions in the workflow software through a yml configuration file;

[0023] The yml configuration file maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

[0024] By leveraging the powerful understanding capabilities of large language models and combining them with plug-in technology, large models can be integrated with workflow products. This allows large models to intervene in workflow processing through plug-ins, seamlessly connecting with existing products. In addition to intelligent order submission, it also enables intelligent recommendation, intelligent data entry, and intelligent customer service. Users no longer need to spend time and effort understanding the function and operation steps of various process orders; they simply ask questions to the large model. Through question-and-answer dialogue, the large model can automatically complete the operation of the work order based on the requirements. This achieves intelligent operation of the application software without affecting existing functions, eliminating cumbersome work order processes. Workflow operation can be achieved through simple communication with the large model, greatly improving work efficiency while reducing learning costs.

[0025] The advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0026] Figure 1 A schematic diagram of the structural framework of an automated workflow processing system is provided for embodiments of the present invention;

[0027] Figure 2 A schematic diagram of an automated workflow processing method provided for other embodiments of the present invention. Detailed Implementation

[0028] The principles and features of the present invention are described below with reference to the accompanying drawings. The embodiments described are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0029] like Figure 1The diagram illustrates the structural framework of an automated workflow processing system according to an embodiment of the present invention. This automated workflow processing system, based on a large language model, includes: an instruction recognition module, a communication module, and a plug-in module. The instruction recognition module provides intelligent input functionality, meaning it intelligently recognizes semantics and keywords based on user input and provides corresponding services accordingly. The plug-in module sends feedback information to the dialog box of the communication module to guide the user to provide more explicit and effective information, repeating the above process until a clear user requirement is obtained. The communication module includes a large language model service and a dialog box. The dialog box is used to obtain the natural language content input by the user and send it to the large language model service. The large language model service is used to perform directional judgment on the natural language content.

[0030] It should be noted that the directionality judgment refers to judging whether the natural language content contains directionality content. Directional content refers to the content in the natural language content that can be understood by the large language model and is targeted at a specific function or need. For example, querying xx work order can be understood by the large language model and is targeted at the work order query function, so it belongs to directionality content.

[0031] When the natural language content includes directional content and contains execution instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the execution information to obtain the first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the address of the corresponding service is called. The execution information is the instruction configuration file maintained by the instruction recognition module. The instruction configuration file includes execution instructions and addresses with binding relationships.

[0032] It should be noted that the execution instructions are a set of natural language provided by the instruction recognition module. For example, they can be: [work order query, work order creation, work order statistics, financial reimbursement process...] etc.

[0033] It should be noted that the instruction recognition module binds one or a group of instructions to a specific service address or backend interface address. The source of the instruction and the service address or backend interface address is the workflow software, and users enter this information into the instruction recognition module through configuration files.

[0034] The following is an example command configuration file:

[0035] The instruction configuration file is a properties file, with content in key=value format, where the key is the instruction name and the value is the process interface address. For example, the process.properties file:

[0036] Create a reimbursement process: http: / / xxx:xxx / xx / xx

[0037] View maintenance work orders: http: / / xxx:xxx / xx / xx

[0038] The instruction recognition module loads the above instruction configuration file as dictionary entries and maintains it, which is called runtime information.

[0039] When the natural language content includes directional information and its semantics match the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software, as well as a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain a second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through a dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is invoked.

[0040] It should be understood that a plugin description refers to descriptive information such as the features, functions, and processes of a plugin.

[0041] The feedback information from the large language model service includes answers to questions, as well as application services provided to solve problems. Users can call specific applications as needed.

[0042] It should be understood that the plugin module is similar to the ChatGPT plugin function. The workflow software is abstracted into a plugin. When the communication module decides to forward the request to the plugin module, the plugin module decides which function of the workflow software to use and what content to return.

[0043] It should be understood that the plugin module is used to manage the various functions of the workflow software, and the function information needs to be manually maintained in the yml configuration file;

[0044] Below is an example of a yml configuration file:

[0045]

[0046] desc: "This interface is called when a user needs to create a new event ticket."

[0047] #After the call ends, if a page needs to be opened, the link to open when returning should be dynamic parameters concatenated with ${}, in the form of JSON values ​​returned after the request, such as data.id, etc.

[0048]

[0049]

[0050]

[0051] It should be understood that after the plugin module is called, it will return interface parameter information. The plugin module will generate questions based on these parameters and present these questions to the user one by one, step by step, until all information is obtained, the corresponding plugin function is called, and the result is returned to the user.

[0052] It should be understood that once all the questions listed in the question configuration have been interacted with by the user, it can be considered that all the necessary information has been obtained.

[0053] When the natural language content does not include directional information, the large language model service is used to generate response information based on the natural language content and send it to the dialog box for display.

[0054] It should be understood that large language model services refer to large language models, also known as large-scale language models, which are artificial intelligence models designed to understand and generate human language. They are trained on large amounts of text data and can perform a wide range of tasks, including text summarization, translation, sentiment analysis, and more.

[0055] It should be understood that the Ratcliff-Obershelp algorithm can be used to match user input with execution instructions, thereby making a directional judgment on the natural language content.

[0056] It should be understood that, such as Figure 1 As shown, both the instruction recognition module and the plug-in module can interface with existing workflow software to obtain runtime and application information, respectively. This means users can manually input necessary information from the workflow software into the instruction recognition and plug-in modules via configuration files. The plug-in module, in addition to interfaceing with workflow software, can theoretically also be integrated with other software.

[0057] It should be noted that the large language model service can be located on the backend, which can be a server, while the dialog box can be located on the frontend, which can be a mobile phone, computer, etc.

[0058] The large language model in this embodiment can be GPT-3, BERT, T5, Tongyi Qianwen, Wenxin Yiyan, or Xunfei Xinghuo, etc.

[0059] This embodiment leverages the powerful understanding capabilities of a large language model, combined with plug-in technology, to integrate the large model with workflow products. This allows the large model to participate in workflow processing through plug-ins, seamlessly connecting with existing products. In addition to intelligent order submission, it also enables intelligent recommendation, intelligent data entry, and intelligent customer service. Users don't need to spend time and effort understanding the function and operation steps of various process orders; they simply ask the large model a question, and through question-and-answer dialogue, the large model automatically completes the operation of the work order based on the requirements. Without affecting existing functionality, it achieves intelligent operation of the application software, shielding users from cumbersome work order processes. Workflow operation can be achieved simply by communicating with the large model, greatly improving work efficiency while reducing learning costs.

[0060] Optionally, in some possible implementations, the address includes a service address and a backend interface address. The source of the running instruction, the service address, and the backend interface address is the workflow software. The instruction recognition module is specifically used to call the service corresponding to the running instruction through the service address or the backend interface address.

[0061] Optionally, in some possible implementations, the instruction recognition module predefines a prompt template, and the instruction recognition module is specifically used to add natural language content and running information to the corresponding positions of the prompt template to obtain the first prompt text.

[0062] For example, the prompt template content is:

[0063] "The following options are available: commandStr. Please select the option that is most semantically similar to the following sentence: input. Please answer directly with the option content, without generating unnecessary words. If you cannot make a selection, answer: Option content does not exist;"

[0064] In the prompt template, commandStr is the set of instructions, and input is the information entered by the user.

[0065] It should be understood that when a prompt text is sent to the large language model service of the chat module, the command recognition module can invoke a specific service address based on the feedback from the large language model service. For example, "Open the maintenance work order software" corresponds to the service address of the maintenance work order software.

[0066] Optionally, in some possible implementations, the plugin module predefines a prompt template, and the plugin module is specifically used to add natural language content and plugin description to the corresponding position of the prompt template to obtain the second prompt text.

[0067] It should be understood that the process of generating the second prompt text is the same as that of generating the first prompt text, only the content added is different, so it will not be described again.

[0068] Optionally, in some possible implementations, the plugin module is specifically used to manage the functions in the workflow software through a yml configuration file. The yml configuration file maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

[0069] like Figure 2 The diagram shown illustrates an automated workflow processing method according to another embodiment of the present invention. This automated workflow processing method based on a large language model includes:

[0070] S1, The dialog box obtains the natural language content input by the user and sends it to the large language model service;

[0071] S2, the large language model service performs directional judgment on natural language content:

[0072] S21, when the natural language content includes directional content and contains running instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the running information to obtain the first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the address of the corresponding service is called. The running information is the instruction configuration file maintained by the instruction recognition module. The instruction configuration file includes running instructions and addresses with binding relationships.

[0073] S22, when the natural language content includes directional content and the semantics match the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software, as well as a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain the second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through a dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is called.

[0074] S23, When the natural language content does not include directional content, the large language model service generates answer information based on the natural language content and sends it to the dialog box for display.

[0075] Optionally, in some possible implementations, the address includes a service address and a backend interface address. The source of the running instruction, the service address, and the backend interface address is the workflow software. The instruction recognition module specifically calls the service corresponding to the running instruction through the service address or the backend interface address.

[0076] Optionally, in some possible implementations, the instruction recognition module predefines a prompt template, and the instruction recognition module specifically adds natural language content and running information to the corresponding positions of the prompt template to obtain the first prompt text.

[0077] Optionally, in some possible implementations, the plugin module predefines a prompt template, and the plugin module specifically adds natural language content and plugin description to the corresponding positions in the prompt template to obtain the second prompt text.

[0078] Optionally, in some possible implementations, it also includes:

[0079] Plugin modules manage the functions in the workflow software through yml configuration files;

[0080] The yml configuration file maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

[0081] It should be understood that the above implementation method is a method implementation method corresponding to the prior product implementation method. For a description of the method implementation method, please refer to the description of the prior product implementation method, which will not be repeated here.

[0082] It should be understood that, without departing from the concept of the present invention, those skilled in the art can combine the above embodiments in any way, and all such combinations are within the protection scope of the present invention.

[0083] Readers should understand that in the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the method embodiments described above are merely illustrative. For instance, the division of steps is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple steps may be combined or integrated into another step, or some features may be ignored or not executed.

[0085] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An automated workflow processing system based on a large language model, characterized in that, include: The system includes an instruction recognition module, a communication module, and a plugin module. The communication module comprises a large language model service and a dialog box. The dialog box is used to acquire natural language content input by the user and send it to the large language model service. The large language model service is used to perform directional determination on the natural language content. When the natural language content includes directional content and contains execution instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the execution information to obtain a first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the module calls the address of the corresponding service. The execution information is an instruction configuration file maintained by the instruction recognition module, and the instruction configuration file includes execution instructions and addresses with binding relationships. When the natural language content includes directional content and semantically matches the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software and a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain a second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through the dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is invoked. When the natural language content does not include directional content, the large language model service is used to generate answer information based on the natural language content and send it to the dialog box for display.

2. The automated workflow processing system based on a large language model according to claim 1, characterized in that, The address includes a service address and a backend interface address. The source of the running instruction, the service address, and the backend interface address is the workflow software. The instruction recognition module is specifically used to call the service corresponding to the running instruction through the service address or the backend interface address.

3. The automated workflow processing system based on a large language model according to claim 1, characterized in that, The instruction recognition module predefines a prompt template. Specifically, the instruction recognition module is used to add the natural language content and the running information to the corresponding positions of the prompt template to obtain the first prompt text.

4. The automated workflow processing system based on a large language model according to claim 1, characterized in that, The plugin module predefines a prompt template. Specifically, the plugin module is used to add the natural language content and the plugin description to the corresponding position of the prompt template to obtain the second prompt text.

5. The automated workflow processing system based on a large language model according to any one of claims 1 to 4, characterized in that, The plugin module is specifically used to manage the functions in the workflow software through a yml configuration file. The yml configuration file maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

6. An automated workflow processing method based on a large language model, characterized in that, include: The dialog box retrieves the natural language content input by the user and sends it to the large language model service; The large language model service performs a directional determination on the natural language content: When the natural language content includes directional content and contains execution instructions, the large language model service sends the natural language content to the instruction recognition module. The instruction recognition module combines the natural language content with the execution information to obtain a first prompt text, and sends the first prompt text to the large language model service for recognition. Based on the feedback result of the large language model service, the address of the corresponding service is called. The execution information is an instruction configuration file maintained by the instruction recognition module, and the instruction configuration file includes execution instructions and addresses with binding relationships. When the natural language content includes directional content and semantically matches the plugin description, the large language model service sends the natural language content to the plugin module. The plugin module includes multiple plugins abstracted from workflow software and a plugin description for each plugin. The plugin module combines the natural language content with the plugin description to obtain a second prompt text, and sends the second prompt text to the large language model service for recognition. Based on the feedback information from the large language model service, the plugin and its corresponding function are selected. A question is generated based on the selected plugin function and displayed through the dialog box. The user's answer is obtained through the dialog box. The question and answer process is repeated until all the information required for the selected plugin function is obtained, and the corresponding plugin function is invoked. When the natural language content does not include directional information, the large language model service generates a response based on the natural language content and sends it to the dialog box for display.

7. The automated workflow processing method based on a large language model according to claim 6, characterized in that, The address includes a service address and a backend interface address. The source of the running instruction, the service address, and the backend interface address is the workflow software. The instruction recognition module specifically calls the service corresponding to the running instruction through the service address or the backend interface address.

8. The automated workflow processing method based on a large language model according to claim 6, characterized in that, The instruction recognition module predefines a prompt template. Specifically, the instruction recognition module adds the natural language content and the running information to the corresponding positions of the prompt template to obtain the first prompt text.

9. The automated workflow processing method based on a large language model according to claim 6, characterized in that, The plugin module predefines a prompt template. Specifically, the plugin module adds the natural language content and the plugin description to the corresponding positions in the prompt template to obtain the second prompt text.

10. The automated workflow processing method based on a large language model according to any one of claims 6 to 9, characterized in that, Also includes: The plugin module manages the functions in the workflow software through a yml configuration file; The yml configuration file maintains the function name, function description, application service call instructions and keywords corresponding to the function, interface address, interface parameter information, and information returned to the user.

Images