Digital intelligent employee agent implementation method and system based on large model

Abstract

The embodiment of the invention provides a digital intelligent employee agent implementation method and system based on a large model. According to the method, when a current task is received, current task information is firstly extracted, and a foundation is laid for subsequent processing; then, current task information is understood by utilizing a multi-technology fusion basic platform, and non-preset tasks are accurately identified in combination with a preset rule; calling a large and small model collaboration module to decompose the non-preset task and generate a non-preset collaboration execution result; then, non-preset target knowledge associated with the non-preset subtasks is retrieved from a plug-in knowledge base which is updated in real time and is safe and controllable through a semantic retrieval algorithm of a retrieval enhancement generation technology, and accurate knowledge support is provided for subtask execution; and finally, on the basis of the non-preset collaborative execution result and the non-preset target knowledge, each non-preset subtask is executed, and a result is generated and fed back to the current user, the process does not need manual intervention of complex task decomposition and knowledge retrieval, the manual processing error rate is reduced, and meanwhile, the office efficiency of an enterprise can be improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for implementing a digital intelligent employee agent based on a large model. Background Technology

[0002] With the rapid development of artificial intelligence technology, enterprises are facing increasingly complex business environments and growing business demands, and their business scale is also expanding. Therefore, higher requirements are being placed on the efficiency, stability and intelligence of office processes.

[0003] Currently, traditional office work relies heavily on manual operation, meaning that the entire process from task reception and information processing to result output requires human intervention.

[0004] However, the traditional office model that relies on manual intervention is prone to process backlog when faced with massive business scenarios, which cannot meet the needs of efficient business flow and thus leads to low office efficiency. Summary of the Invention

[0005] This invention provides a method and system for implementing a digital intelligent employee agent based on a large model, which can improve the office efficiency of enterprises.

[0006] In a first aspect, embodiments of the present invention provide a method for implementing a digital intelligent employee agent based on a large model, the method comprising:

[0007] Receive the current task and extract the current task information from the current task, wherein the current task information includes: task type, task objective and key information;

[0008] The system utilizes a multi-technology fusion platform to perform task understanding operations on the extracted current task information and identifies whether the current task is a non-preset task based on preset rules.

[0009] When the current task is identified as a non-preset task, the size model collaboration module is invoked to decompose the current task into at least one non-preset sub-task, and preliminary processing is performed to generate a non-preset collaborative execution result.

[0010] A semantic retrieval algorithm based on retrieval enhancement generation technology retrieves non-preset target knowledge associated with the at least one non-preset subtask from an external knowledge base;

[0011] Based on the non-preset collaborative execution results and the non-preset target knowledge, each non-preset sub-task after decomposition is executed, non-preset task execution results are generated and fed back to the current user.

[0012] Preferably,

[0013] After receiving the current task and before extracting the current task information from the current task, the method further includes:

[0014] Determine whether the current task is in voice format;

[0015] When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task.

[0016] The step of utilizing a multi-technology fusion platform to perform task understanding operations on the extracted current task information and identifying whether the current task is a non-preset task based on preset rules includes:

[0017] Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information;

[0018] The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph.

[0019] Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

[0020] Preferably,

[0021] The semantic retrieval algorithm based on retrieval enhancement generation technology retrieves non-preset target knowledge associated with the at least one non-preset subtask from the external knowledge base, including:

[0022] The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base;

[0023] The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base;

[0024] The first formula is:

[0025] Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base;

[0026] Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

[0027] Preferably,

[0028] After executing the decomposed non-preset sub-tasks based on the non-preset collaborative execution results and the non-preset target knowledge, generating non-preset task execution results and feeding them back to the current user, the process further includes:

[0029] When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task.

[0030] Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library;

[0031] The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result;

[0032] Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base;

[0033] Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

[0034] Secondly, embodiments of the present invention provide a digital intelligent employee agent implementation system based on a large model, the system comprising:

[0035] A receiving module is used to receive the current task and extract the current task information from the current task, wherein the current task information includes: task type, task objective and key information;

[0036] Identification module: used to perform task understanding operation on the current task information extracted by the receiving module using a multi-technology fusion platform, and to identify whether the current task is a non-preset task based on preset rules;

[0037] Collaboration module: When the identification module identifies the current task as the non-preset task, it calls the size model collaboration module to decompose the current task into at least one non-preset sub-task and performs preliminary processing to generate a non-preset collaborative execution result.

[0038] Retrieval module: Used to retrieve non-preset target knowledge associated with the at least one non-preset subtask from the external knowledge base using a semantic retrieval algorithm based on retrieval enhancement generation technology;

[0039] Non-preset task processing module: Based on the non-preset collaborative execution results generated by the collaborative module and the non-preset target knowledge retrieved by the retrieval module, it executes each decomposed non-preset sub-task, generates non-preset task execution results, and feeds them back to the current user.

[0040] Preferably,

[0041] The receiving module is further configured to perform the following after receiving the current task and before extracting the current task information from the current task:

[0042] Determine whether the current task is in voice format;

[0043] When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task.

[0044] The aforementioned identification module is also used to perform:

[0045] Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information;

[0046] The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph.

[0047] Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

[0048] Preferably,

[0049] The retrieval module is also used to perform:

[0050] The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base;

[0051] The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base;

[0052] The first formula is:

[0053] Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base;

[0054] Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

[0055] Preferably,

[0056] Following the non-preset task processing module, a preset task processing module is further included;

[0057] The preset task processing module is used to execute:

[0058] When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task.

[0059] Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library;

[0060] The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result;

[0061] Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base;

[0062] Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

[0063] Thirdly, embodiments of the present invention provide a digital intelligent employee agent implementation system based on a large model, comprising: at least one memory and at least one processor;

[0064] The at least one memory is used to store a machine-readable program;

[0065] The at least one processor is configured to invoke the machine-readable program to execute any of the methods described in the first aspect.

[0066] Fourthly, embodiments of the present invention provide a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform any of the methods described in the first aspect.

[0067] This invention provides a method and system for implementing a digital intelligent employee agent based on a large model. Upon receiving a current task, the method first extracts current task information, including task type, task objectives, and key information, to lay a data foundation for subsequent processing. Then, it utilizes a multi-technology fusion platform to understand the current task information and accurately identify non-preset tasks using preset rules to ensure accurate task classification. Next, it calls a large-scale model collaboration module to decompose the non-preset tasks and generate non-preset collaborative execution results. Subsequently, it uses a semantic retrieval algorithm with enhanced generation technology to accurately retrieve non-preset target knowledge associated with non-preset sub-tasks from a real-time updated and secure external knowledge base, providing precise knowledge support for sub-task execution. Finally, based on the non-preset collaborative execution results and non-preset target knowledge, it executes each non-preset sub-task and generates results to feed back to the current user. This process eliminates the need for manual intervention in complex task decomposition and knowledge retrieval steps, significantly improving enterprise office efficiency while reducing human error rates. Attached Figure Description

[0068] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a flowchart of a method for implementing a digital intelligent employee agent based on a large model, provided by an embodiment of the present invention;

[0070] Figure 2 This is a flowchart of another method for implementing a digital intelligent employee agent based on a large model, provided by an embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram of a digital intelligent employee agent implementation system based on a large model, provided by an embodiment of the present invention;

[0072] Figure 4 This is a schematic diagram of another digital intelligent employee agent implementation system based on a large model provided in an embodiment of the present invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0074] like Figure 1 As shown, this embodiment of the invention provides a method for implementing a digital intelligent employee agent based on a large model. This method may include the following steps:

[0075] Step 101: Receive the current task and extract the current task information from the current task, which includes: task type, task objective and key information;

[0076] Step 102: Utilize a multi-technology fusion platform to perform task understanding operations on the extracted current task information, and identify whether the current task is a non-preset task based on preset rules;

[0077] Step 103: When the current task is identified as a non-preset task, the size model collaboration module is invoked to decompose the current task into at least one non-preset sub-task, and preliminary processing is performed to generate non-preset collaborative execution results;

[0078] Step 104: The semantic retrieval algorithm based on retrieval enhancement generation technology retrieves non-preset target knowledge associated with at least one non-preset subtask from the external knowledge base;

[0079] Step 105: Based on the non-preset collaborative execution results and non-preset target knowledge, execute each decomposed non-preset sub-task, generate non-preset task execution results, and feed them back to the current user.

[0080] This invention provides a method for implementing a digital intelligent employee agent based on a large model. Upon receiving a current task, the method first extracts current task information, including task type, task objectives, and key information, to lay a data foundation for subsequent processing. Then, it utilizes a multi-technology fusion platform to understand the current task information and accurately identifies non-preset tasks using preset rules to ensure accurate task classification. Next, it calls a large-scale model collaboration module to decompose the non-preset tasks and generate non-preset collaborative execution results. Subsequently, it uses a semantic retrieval algorithm with enhanced generation technology to accurately retrieve non-preset target knowledge associated with non-preset sub-tasks from a real-time updated and secure external knowledge base, providing precise knowledge support for sub-task execution. Finally, based on the non-preset collaborative execution results and non-preset target knowledge, it executes each non-preset sub-task and generates results to feed back to the current user. This process eliminates the need for manual intervention in complex task decomposition and knowledge retrieval steps, significantly improving enterprise office efficiency while reducing human error rates.

[0081] To identify non-preset tasks, in one embodiment of the present invention, after receiving the current task in the above embodiment and before extracting the current task information from the current task, the method further includes:

[0082] Determine whether the current task is in voice format;

[0083] When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task.

[0084] The step of utilizing a multi-technology fusion platform to perform task understanding operations on the extracted current task information and identifying whether the current task is a non-preset task based on preset rules includes:

[0085] Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information;

[0086] The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph.

[0087] Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

[0088] In this embodiment of the invention, upon receiving the current task, since the current task may be in the form of voice or text, it is necessary to process voice-based tasks and convert them into text to solve the problem that non-text tasks are difficult to parse directly, thus providing a unified data format for subsequent task information extraction and understanding. To save computing power, the execution flow of non-preset tasks and preset tasks are different, so it is necessary to identify the task type. First, the natural language processing technology of the multi-technology fusion platform is used to parse the current task information and extract the business entities, time nodes, and constraints from the current task information. Then, the knowledge graph module of the platform is called to realize the association between the extracted information and the pre-stored business knowledge and generate a target knowledge graph. The binding between task information and enterprise business scenarios is strengthened through knowledge graph association. Finally, non-preset tasks are identified based on the target knowledge graph and preset rules to avoid task classification deviations that may lead to deviations in execution results and improve the reliability of the results.

[0089] In order to retrieve target knowledge associated with each subtask, in one embodiment of the present invention, step 104 in the above embodiment may specifically include the following steps:

[0090] The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base;

[0091] The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base;

[0092] The first formula is:

[0093] Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base;

[0094] Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

[0095] In this embodiment of the invention, to retrieve target knowledge, at least one non-preset subtask is first transformed into a non-preset subtask vector, and the external knowledge base is transformed into a vector knowledge base. This achieves the structuring of the non-preset subtasks and the knowledge base content, providing a computable data foundation for subsequent semantic retrieval. Secondly, a semantic retrieval algorithm based on retrieval enhancement generation technology calculates the cosine similarity between each non-preset subtask vector and the vector knowledge base using the cosine similarity calculation formula. This accurately measures the semantic association between the non-preset subtasks and the knowledge in the knowledge base, ensuring the relevance of the retrieval results and avoiding the semantic bias problem that easily occurs in traditional keyword retrieval. Finally, based on a preset similarity threshold, at least one piece of target knowledge with a cosine similarity higher than the threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask. Threshold filtering ensures the accuracy of the target knowledge, eliminating the need for manual screening of information matching non-preset subtasks from a massive knowledge base. This reduces the manual involvement in knowledge retrieval and matching, lowers the time consumption and error of manual operations, and thus ensures the overall processing efficiency of non-preset tasks.

[0096] In order to process the preset task, in one embodiment of the present invention, after step 104, the above embodiment further includes:

[0097] When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task.

[0098] Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library;

[0099] The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result;

[0100] Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base;

[0101] Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

[0102] In this embodiment of the invention, the preset task is a static workflow orchestration, which uses a process orchestration framework to define the various steps and execution order of the workflow, and executes strictly according to the plan during workflow execution. Therefore, when the current task is identified as a preset task, the natural language processing technology of the multi-technology fusion platform is first used to parse the preset task to determine the task type; then, based on the task type, the corresponding standardized execution process template is retrieved from the pre-stored static workflow template library to provide a standardized basis for the execution of the preset task; next, the large and small model collaboration module is called to decompose the preset task into at least one preset sub-task and generate a preset collaborative execution result. The large model is responsible for interacting with the user, understanding the user's intention and guiding the user to complete parameter filling, thereby making the selection of execution branches. Although this mode has a relatively fixed interaction, it can ensure that the task is executed according to the established standard operating procedure, ensuring the rigor and accuracy of task execution (for example, in the enterprise leave approval process, the digital employee follows the preset...). The workflow guides employees to fill out leave application information, and then sends the approval request to the corresponding person in charge according to the leave type and approval authority. Next, based on semantic retrieval algorithms, it retrieves preset target knowledge associated with preset sub-tasks from an external knowledge base. Leveraging the "real-time updating and secure controllable" characteristics of the external knowledge base and the semantic matching capabilities of the retrieval enhancement generation technology, it maintains information timeliness. Combined with fine-grained access control and security measures, it ensures data security. Finally, it executes each preset sub-task based on preset collaborative execution results, preset target knowledge, and standardized execution process templates, and provides feedback on the results (for example, when handling enterprise contract review tasks, the digital intelligence employee can quickly retrieve relevant laws and regulations, contract templates, and past cases from the external knowledge base to assist the large model in generating accurate review opinions). Furthermore, the digital intelligence employee can adjust and optimize its behavior and decisions based on task execution results and user feedback. Through continuous learning and improvement, it continuously enhances service quality and task processing capabilities (for example, when users are dissatisfied with the generated document content, the digital intelligence employee can regenerate the document based on user feedback).

[0103] In one embodiment of the present invention, the large model and the small model have the following three collaboration modes:

[0104] Serial mode: Large and small models collaborate, with the large model performing preprocessing and post-processing, and the small model performing precise evaluation calculations. (For example, in a project risk assessment task, the large model first analyzes the overall project situation to determine the scope and focus of the risk assessment, while the small model quantifies the specific risk factors.)

[0105] Traffic splitting mode: Large models split tasks to small models (for example, when a user's business request is received, the large model distributes the task to the corresponding small model according to the request intent).

[0106] Coordination Mode: The large model coordinates various models and summarizes the results to generate a comprehensive solution (for example, in corporate strategic planning tasks, the large model coordinates market analysis models, financial models, human resource models, etc., and integrates the results of various models to generate a strategic planning solution).

[0107] like Figure 2 As shown, to more clearly illustrate the technical solution and advantages of the present invention, the following provides a detailed description of the implementation method of the digital intelligent employee agent based on a large model, which may include the following steps:

[0108] Step 201: Receive the current task and determine whether the current task is in voice format;

[0109] Through natural language processing technology, intelligent employees can understand user input commands, including text and voice. Simultaneously, they can perceive data generated by internal business systems, such as emails, documents, and reports, and extract key information from them.

[0110] Step 202: When it is determined that the current task is in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task;

[0111] Step 203: Extract the current task information from the current task, which includes: task type, task objective, and key information;

[0112] Step 204: Utilize the natural language processing technology of the multi-technology fusion platform to parse the current task information and extract the business entities, time nodes, and constraints from the current task information;

[0113] Step 205: Call the knowledge graph module of the multi-technology fusion platform to associate the extracted business entities, time nodes and constraints with the pre-stored business knowledge in the knowledge graph, and generate the target knowledge graph;

[0114] Step 206: Identify whether the current task is a non-preset task based on the target knowledge graph and preset rules;

[0115] Step 207: When the current task is identified as a non-preset task, the size model collaboration module is invoked to decompose the current task into at least one non-preset sub-task, and preliminary processing is performed to generate non-preset collaborative execution results;

[0116] For example, in financial statement analysis tasks, the large model is responsible for trend analysis and anomaly detection of the overall financial data, while the small model focuses on the accurate calculation of specific financial indicators and compliance checks.

[0117] Step 208: Convert at least one non-preset subtask and the plug-in knowledge base into at least one non-preset subtask vector and vector knowledge base, respectively;

[0118] Step 209: Based on the semantic retrieval algorithm of retrieval enhancement generation technology, calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base using the first formula;

[0119] Specifically, the first formula is:

[0120] Where cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base;

[0121] Step 210: Based on a preset similarity threshold, identify at least one target knowledge from the vector knowledge base whose cosine similarity is higher than the preset similarity threshold, and associate it with the corresponding non-preset subtask;

[0122] Step 211: Based on the non-preset collaborative execution results and non-preset target knowledge, execute each decomposed non-preset sub-task, generate the non-preset task execution results and feed them back to the current user;

[0123] Step 212: When the current task is identified as a preset task, natural language processing technology is used to parse the preset task and determine the task type of the preset task.

[0124] Step 213: Retrieve the corresponding standardized execution process template from the pre-stored static workflow template library based on the task type;

[0125] Step 214: Call the size model collaboration module to decompose the preset task into at least one preset sub-task, and perform preliminary processing to generate the preset collaborative execution result;

[0126] Step 215: Retrieve preset target knowledge associated with at least one preset subtask from the external knowledge base based on a semantic retrieval algorithm;

[0127] Step 216: Based on the preset collaborative execution results and preset target knowledge, execute each preset sub-task after decomposition according to the standardized execution process template, generate the preset task execution results and feed them back to the current user.

[0128] like Figure 3 As shown, this embodiment of the invention provides a digital intelligent employee agent implementation system based on a large model, the system comprising:

[0129] The receiving module 301 is used to receive the current task and extract the current task information from the current task, wherein the current task information includes: task type, task objective and key information;

[0130] Identification module 302: Used to perform task understanding operation on the current task information extracted by the receiving module 301 using a multi-technology fusion platform, and to identify whether the current task is a non-preset task based on preset rules;

[0131] Collaboration module 303: When the identification module 302 identifies the current task as the non-preset task, it calls the size model collaboration module to decompose the current task into at least one non-preset sub-task and performs preliminary processing to generate a non-preset collaborative execution result.

[0132] Retrieval module 304: Used for retrieving non-preset target knowledge associated with the at least one non-preset subtask from an external knowledge base using a semantic retrieval algorithm based on retrieval enhancement generation technology;

[0133] Non-preset task processing module 305: Based on the non-preset collaborative execution result generated by the collaborative module 303 and the non-preset target knowledge retrieved by the retrieval module 304, it executes each decomposed non-preset sub-task, generates non-preset task execution results, and feeds them back to the current user.

[0134] like Figure 4 As shown, the receiving module 301 is further configured to perform the following after receiving the current task and before extracting the current task information from the current task:

[0135] Determine whether the current task is in voice format;

[0136] When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task.

[0137] The identification module 302 is also used to perform:

[0138] Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information;

[0139] The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph.

[0140] Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

[0141] like Figure 4 As shown, the retrieval module 304 is also used to perform:

[0142] The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base;

[0143] The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base;

[0144] The first formula is:

[0145] Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base;

[0146] Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

[0147] based on Figure 3 The system shown is a digital intelligent employee agent implementation system based on a large model, such as... Figure 4 As shown, after the non-preset task processing module 305, a preset task processing module 306 is further included.

[0148] The preset task processing module 306 is used to execute:

[0149] When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task.

[0150] Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library;

[0151] The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result;

[0152] Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base;

[0153] Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

[0154] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on the implementation system of the large-model-based intelligent employee agent. In other embodiments of the present invention, the large-model-based intelligent employee agent implementation system may include more or fewer components than illustrated, or combine some components, split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0155] The information interaction and execution process between the various units in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0156] This invention also provides a digital intelligent employee agent implementation system based on a large model, comprising: at least one memory and at least one processor;

[0157] At least one memory for storing machine-readable programs;

[0158] At least one processor is configured to invoke a machine-readable program to execute the large-model-based intelligent employee agent implementation method of any embodiment of the present invention.

[0159] This invention also provides a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform the large-model-based intelligent employee agent implementation method according to any embodiment of this invention.

[0160] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0161] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0162] Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0163] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0164] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0165] The various embodiments of the present invention have at least the following beneficial effects:

[0166] 1. In this embodiment of the invention, when the current task is received, the method first extracts current task information covering task type, task objective, and key information to lay a data foundation for subsequent processing; then, it uses a multi-technology fusion platform to understand the current task information and accurately identifies non-preset tasks in combination with preset rules to ensure the accuracy of task classification; next, it calls the size model collaboration module to decompose the non-preset tasks and generate non-preset collaborative execution results; then, it uses a semantic retrieval algorithm of retrieval enhancement generation technology to accurately retrieve non-preset target knowledge associated with non-preset sub-tasks from a real-time updated and securely controllable plug-in knowledge base, providing accurate knowledge support for sub-task execution; finally, based on the non-preset collaborative execution results and non-preset target knowledge, it executes each non-preset sub-task and generates results to be fed back to the current user. The above process does not require manual intervention in the complex task decomposition and knowledge retrieval process, which can significantly improve the office efficiency of enterprises while reducing the error rate of manual processing.

[0167] 2. In this embodiment of the invention, after receiving the current task, since the current task may be in the form of voice or text, it is necessary to process the voice task and convert it into text to solve the problem that non-text tasks are difficult to parse directly, and to provide a unified data format for subsequent task information extraction and understanding. In order to save computing power, the execution flow of non-preset tasks and preset tasks are not the same, so it is necessary to identify the task type. First, the natural language processing technology of the multi-technology fusion basic platform is used to parse the current task information and extract the business entities, time nodes and constraints in the current task information. Then, the knowledge graph module of the platform is called to realize the association between the extracted information and the pre-stored business knowledge and generate the target knowledge graph. The binding between the task information and the enterprise business scenario is strengthened through the association of the knowledge graph. Finally, non-preset tasks are identified based on the target knowledge graph and preset rules to avoid task classification deviation and thus deviation of execution results, thereby improving the reliability of the results.

[0168] 3. In this embodiment of the invention, in order to retrieve target knowledge, at least one non-preset subtask is first transformed into a non-preset subtask vector, and the external knowledge base is transformed into a vector knowledge base. This realizes the structuring of the non-preset subtasks and the knowledge base content, providing a computable data foundation for subsequent semantic retrieval. Secondly, based on the semantic retrieval enhancement generation technology, the cosine similarity between each non-preset subtask vector and the vector knowledge base is calculated using the cosine similarity calculation formula. This can accurately measure the semantic association between the non-preset subtasks and the knowledge in the knowledge base, ensuring the relevance of the retrieval results and avoiding the semantic bias problem that is prone to occur in traditional keyword retrieval. Finally, based on a preset similarity threshold, at least one piece of target knowledge with a cosine similarity higher than the threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask. The threshold screening ensures the accuracy of the target knowledge, eliminating the need for manual screening of information matching non-preset subtasks from a massive knowledge base. This reduces the manual involvement in knowledge retrieval and matching, reduces the time and error of manual operations, and thus ensures the overall processing efficiency of non-preset tasks.

[0169] It should be noted that not all steps and modules in the above processes and system structure diagrams are mandatory; some steps or modules can be omitted as needed. The execution order of each step is not fixed and can be adjusted as required. The system structure described in the above embodiments can be a physical structure or a logical structure. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0170] In the above embodiments, the hardware units can be implemented mechanically or electrically. For example, a hardware unit may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0171] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for implementing a digital intelligent employee agent based on a large model, characterized in that, The method includes: Receive the current task and extract the current task information from the current task, wherein the current task information includes: task type, task objective and key information; The system utilizes a multi-technology fusion platform to perform task understanding operations on the extracted current task information and identifies whether the current task is a non-preset task based on preset rules. When the current task is identified as a non-preset task, the size model collaboration module is invoked to decompose the current task into at least one non-preset sub-task, and preliminary processing is performed to generate a non-preset collaborative execution result. A semantic retrieval algorithm based on retrieval enhancement generation technology retrieves non-preset target knowledge associated with the at least one non-preset subtask from an external knowledge base; Based on the non-preset collaborative execution results and the non-preset target knowledge, each non-preset sub-task after decomposition is executed, non-preset task execution results are generated and fed back to the current user.

2. The method according to claim 1, characterized in that, After receiving the current task and before extracting the current task information from the current task, the method further includes: Determine whether the current task is in voice format; When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task. The step of utilizing a multi-technology fusion platform to perform task understanding operations on the extracted current task information and identifying whether the current task is a non-preset task based on preset rules includes: Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information; The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph. Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

3. The method according to claim 1, characterized in that, The semantic retrieval algorithm based on retrieval enhancement generation technology retrieves non-preset target knowledge associated with the at least one non-preset subtask from the external knowledge base, including: The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base; The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base; The first formula is: Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base; Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

4. The method according to any one of claims 1-3, characterized in that, After executing the decomposed non-preset sub-tasks based on the non-preset collaborative execution results and the non-preset target knowledge, generating non-preset task execution results and feeding them back to the current user, the process further includes: When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task. Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library; The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result; Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base; Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

5. A digital intelligent employee agent implementation system based on a large model, characterized in that, The system includes: A receiving module is used to receive the current task and extract the current task information from the current task, wherein the current task information includes: task type, task objective and key information; Identification module: used to perform task understanding operation on the current task information extracted by the receiving module using a multi-technology fusion platform, and to identify whether the current task is a non-preset task based on preset rules; Collaboration module: When the identification module identifies the current task as the non-preset task, it calls the size model collaboration module to decompose the current task into at least one non-preset sub-task and performs preliminary processing to generate a non-preset collaborative execution result. Retrieval module: Used to retrieve non-preset target knowledge associated with the at least one non-preset subtask from the external knowledge base using a semantic retrieval algorithm based on retrieval enhancement generation technology; Non-preset task processing module: Based on the non-preset collaborative execution results generated by the collaborative module and the non-preset target knowledge retrieved by the retrieval module, it executes each decomposed non-preset sub-task, generates non-preset task execution results, and feeds them back to the current user.

6. The system according to claim 5, characterized in that, The receiving module is further configured to perform the following after receiving the current task and before extracting the current task information from the current task: Determine whether the current task is in voice format; When the current task is determined to be in the form of speech, the speech-to-text technology of the multi-technology fusion platform is used to convert the speech-to-text current task into the text-to-text current task. The aforementioned identification module is also used to perform: Using the natural language processing technology of the multi-technology fusion platform, the current task information is parsed to extract business entities, time nodes and constraints from the current task information; The knowledge graph module of the multi-technology fusion platform is invoked to associate the extracted business entities, time nodes, and constraints with the pre-stored business knowledge in the knowledge graph and generate the target knowledge graph. Based on the target knowledge graph and the preset rules, it is determined whether the current task is a non-preset task.

7. The system according to claim 6, characterized in that, The retrieval module is also used to perform: The at least one non-preset subtask and the plug-in knowledge base are respectively transformed into at least one non-preset subtask vector and a vector knowledge base; The semantic retrieval algorithm based on the retrieval enhancement generation technology uses the first formula to calculate the cosine similarity between each non-preset subtask vector and the vector knowledge base; The first formula is: Wherein, cosθ is the cosine similarity between each non-preset subtask vector and the vector knowledge base, A is each non-preset subtask vector, and B is the vector knowledge base; Based on a preset similarity threshold, at least one target knowledge with a cosine similarity higher than the preset similarity threshold is determined from the vector knowledge base and associated with the corresponding non-preset subtask.

8. The system according to any one of claims 5-7, characterized in that, Following the non-preset task processing module, a preset task processing module is further included; The preset task processing module is used to execute: When the current task is identified as a preset task, the natural language processing technology is used to parse the preset task and determine the task type of the preset task. Based on the task type, retrieve the corresponding standardized execution process template from the pre-stored static workflow template library; The size model collaboration module is invoked to decompose the preset task into at least one preset sub-task, and preliminary processing is performed to generate a preset collaborative execution result; Based on the semantic retrieval algorithm, preset target knowledge associated with the at least one preset subtask is retrieved from the plug-in knowledge base; Based on the preset collaborative execution results and the preset target knowledge, each preset sub-task is executed according to the standardized execution process template, and the preset task execution results are generated and fed back to the current user.

9. A digital intelligent employee agent implementation system based on a large model, characterized in that, include: At least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to invoke the machine-readable program to perform the method according to any one of claims 1 to 4.

10. A computer-readable medium, characterized in that, The computer-readable medium stores computer instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1 to 4.

Images