Task Matching Method and System

Abstract

The present application relates to the field of artificial intelligence technology, and provides a task matching method and system, the method comprising: obtaining a target task, constructing a matching model, the matching model being used to process the target task; then inputting the target task, data information and a first prompt word into the matching model, so as to output the task decomposition steps through the matching model, the data information being the information of the target person, and the first prompt word being used to guide the matching model to output the task decomposition steps; then obtaining a second prompt word, and inputting the task decomposition steps, data information and the second prompt word into the matching model, so as to output the matching result through the matching model, and execute the matching result through the target person. The method can split and match tasks through different prompt words and the language processing capability of the matching model to solve the problem of low accuracy of task assignment or poor efficiency of task execution.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a task matching method and system. Background Art

[0002] Large models can be used in natural language processing, intelligent dialogue, image recognition, task planning, etc. Pre-trained models based on deep learning, such as ChatGPT, Llama, ChatGLM and other large models, can process large amounts of data in complex scenarios, understand and generate natural language text, and have efficient task decomposition and execution capabilities. As an important part of AI development, agents can independently complete tasks and optimize the interactive experience by integrating large models.

[0003] Since the large model can perform tasks such as semantic understanding, it can be used in areas that require communication and management. In the communication process, the communication efficiency between managers and those being managed will directly affect the execution effect of the task. However, in the communication process, there may be problems such as unclear communication, information asymmetry, inaccurate transmission of task instructions, and uneven understanding ability of personnel, resulting in low accuracy of task allocation or poor efficiency of task execution. Summary of the invention

[0004] The present application provides a task matching method and system to solve the problem of low task allocation accuracy or poor task execution efficiency.

[0005] In a first aspect, the present application provides a task matching method, comprising:

[0006] Get the target task;

[0007] Constructing a matching model, wherein the matching model is used to process the target task;

[0008] Inputting the target task, data information and a first prompt word into a matching model, so as to output task decomposition steps through the matching model, wherein the data information is information of the target person, and the first prompt word is used to guide the matching model to output the task decomposition steps;

[0009] Acquire a second prompt word, where the second prompt word is used to guide the matching model to output a matching result;

[0010] The task decomposition steps, data information and the second prompt word are input into a matching model, so as to output a matching result through the matching model, and to execute the matching result through a target person.

[0011] The method can split and match tasks through different prompt words and the language processing capability of the matching model, and can solve the problems of low task allocation accuracy or poor task execution efficiency.

[0012] In some feasible embodiments, constructing a matching model includes:

[0013] Obtain a pre-trained model and a data set, wherein the data set is text data in the same field as the target task;

[0014] The pre-trained model is fine-tuned using the data set to generate a matching model.

[0015] By fine-tuning the pre-trained model, you can improve model performance and make the results generated by the model more accurate.

[0016] In some feasible embodiments, the step of inputting the target task, the data information and the first prompt word into a matching model to output a task decomposition step through the matching model includes:

[0017] If the target task cannot be obtained through the data set, embedding a language processing model in the matching model to obtain an embedding model;

[0018] Converting the target task into a first vector by the embedding model, and converting the data set into a second vector;

[0019] Calculating the cosine similarity of the first vector and the second vector to obtain a reference task, where the reference task is text data in the data set that has the highest similarity to the target task;

[0020] The reference task is input into the matching model to output task decomposition steps through the matching model.

[0021] Through cosine similarity calculation, reference tasks can be matched and the response speed of the task decomposition process can be improved.

[0022] In some feasible embodiments, the step of inputting the target task, the data information and the first prompt word into a matching model to output a task decomposition step through the matching model includes:

[0023] Generate a first prompt word according to the target task and the reference task;

[0024] The target task, data information and the first prompt word are input into the matching model so as to output the task decomposition steps through the matching model.

[0025] The first prompt word can limit the task decomposition steps to a specific range, so that the task decomposition steps output by it are more consistent with the domain of the target task.

[0026] In some feasible embodiments, calculating the cosine similarity of the first vector and the second vector to obtain a reference task includes:

[0027] Performing an average pooling operation on the first vector and the second vector to calculate an average value, where the average value is an average value of the first vector and the second vector;

[0028] Calculating the cosine similarity of the first vector and the second vector according to the average value;

[0029] If the cosine similarity of the second vector is greater than or equal to the similarity threshold, the text data corresponding to the second vector is used as a reference task.

[0030] In some feasible embodiments, before acquiring the target task, the process includes:

[0031] Get data information;

[0032] An intelligent agent is constructed, where the intelligent agent is used to define data information of a target person, where the data information includes first data information and second data information.

[0033] Through the intelligent agent, the results of task matching and decomposition can be viewed more intuitively.

[0034] In some feasible embodiments, inputting the task decomposition steps, data information and the second prompt word into a matching model to output a matching result through the matching model includes:

[0035] Decomposing the task into steps to generate a task list;

[0036] Analyze the task types of the task list;

[0037] Generate a second prompt word according to the task type and the first data information;

[0038] Inputting the task list, the first data information and the second prompt word into the matching model to output a matching result through the matching model;

[0039] The second prompt word can guide the matching model to output a more accurate matching result after analyzing the task list and the first data information.

[0040] In some feasible embodiments, the method further includes:

[0041] generating a third prompt word according to the second data information, wherein the third prompt word is used to guide the matching model to output an evaluation result;

[0042] Inputting the second data information and the third prompt word into the matching model to output an evaluation result through the matching model, wherein the evaluation result includes a first type of result, a second type of result, and a third type of result;

[0043] The evaluation result and the matching result are input into the matching model to output a task description through the matching model, wherein the task description includes a first-category description, a second-category description, and a third-category description.

[0044] The second data information allows the model to evaluate the capabilities of each agent related to the second data information.

[0045] In some feasible embodiments, inputting the evaluation result and the matching result into the matching model to output the task description through the matching model includes:

[0046] If the evaluation result is a first-category result, a fourth prompt word is generated, the first-category result, the matching result and the fourth prompt word are input into the matching model, so that the matching model outputs a first-category description, and the fourth prompt word is used to guide the matching model to output the first-category description;

[0047] If the evaluation result is a second type of result, outputting a second type of description through the matching model, wherein the second type of description is a task decomposition step;

[0048] If the evaluation result is a third-category result, a fifth prompt word is generated, and the third-category result, the matching result and the fifth prompt word are input into the matching model to output a third-category description through the matching model, the complexity of the first-category description is greater than that of the second-category description, and the complexity of the second-category description is greater than that of the third-category description, and the fifth prompt word is used to guide the matching model to output a third-category description.

[0049] Based on the evaluation results, different task step instructions can be output to ensure that the task is accurately understood and efficiently executed.

[0050] In a second aspect, the present application provides a task matching system, comprising:

[0051] An acquisition unit, used for acquiring a target task;

[0052] A construction unit, used for constructing a matching model, wherein the matching model is used for processing the target task;

[0053] A processing unit, configured to input the target task, data information, and a first prompt word into a matching model, so as to output a task decomposition step through the matching model, wherein the data information is information of a target person, and the first prompt word is used to guide the matching model to output the task decomposition step;

[0054] The acquisition unit is further used to acquire a second prompt word, where the second prompt word is used to guide the matching model to output a matching result;

[0055] The matching unit is used to input the task decomposition steps, data information and the second prompt word into the matching model, so as to output the matching result through the matching model, so as to execute the matching result through the target person.

[0056] It can be seen from the above technical solutions that the present application provides a task matching method and system, the method comprising: obtaining a target task, constructing a matching model, the matching model being used to process the target task; then inputting the target task, data information and a first prompt word into the matching model, so as to output the task decomposition steps through the matching model, the data information being the information of the target person, and the first prompt word being used to guide the matching model to output the task decomposition steps; then obtaining a second prompt word, and inputting the task decomposition steps, data information and the second prompt word into the matching model, so as to output the matching result through the matching model, so as to execute the matching result through the target person. The method can split and match tasks through different prompt words and the language processing capability of the matching model to solve the problem of low task allocation accuracy or poor task execution efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0058] Figure 1 A flowchart of a task matching method provided in an embodiment of the present application;

[0059] Figure 2 A schematic diagram of first data information provided in an embodiment of the present application;

[0060] Figure 3 A schematic diagram of second data information provided in an embodiment of the present application;

[0061] Figure 4 A schematic diagram of a data set provided in an embodiment of the present application;

[0062] Figure 5 A schematic diagram of the task decomposition process provided in an embodiment of the present application;

[0063] Figure 6 A schematic diagram of generating a second prompt word provided in an embodiment of the present application;

[0064] Figure 7 A schematic diagram of generating a task description provided in an embodiment of the present application;

[0065] Figure 8 A schematic diagram comparing the first type of description and the third type of description provided for the embodiments of the present application;

[0066] Fig. 9 A schematic diagram of the structure of the task matching system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0068] During the communication process, there may be problems such as unclear communication, information asymmetry, inaccurate transmission of task instructions, and uneven understanding ability of personnel. In some embodiments, the big model can simplify complex information and restate it in plain language to help people with different understanding abilities better understand it.

[0069] In other embodiments, the large model can also divide the task into a series of subtasks and then assign them to agents. However, when breaking down the task into subtasks, if the background, goals and requirements of the task are not fully understood, the information may not be fully transmitted, thus affecting the accurate allocation of subtasks. After the task is broken down, the rights and responsibilities may become blurred, and the agent may not be able to clearly define the specific responsibilities of each member. When problems arise during the execution process, it is difficult to hold them accountable and solve them, all of which lead to low accuracy in task allocation or poor efficiency in task execution.

[0070] In order to solve the problem of low task assignment accuracy or poor task execution efficiency, some embodiments of the present application provide a task matching method, which utilizes the natural language processing capabilities of a large model and implements task decomposition, task assignment and task description adjustment in the form of prompt words to achieve efficient work interaction.

[0071] like Figure 1 As shown, the method comprises the following steps:

[0072] S100: Obtain target tasks.

[0073] The target task is a specific task or work that needs to be assigned and executed, and can be various types of activities, such as writing reports, designing software, completing goals, etc. It can be understood that the method provided in this embodiment can be applied to environments where communication is required in various fields. In this embodiment, the power system is taken as an example to illustrate the method, wherein the target task can be a task that needs to be executed assigned by the manager, for example, the target task is to perform routine inspections on power lines.

[0074] In order to improve data accuracy, enhance matching intelligence and optimize task allocation, in some embodiments, the target task pre-data information is obtained, and then an intelligent agent is constructed, wherein the intelligent agent is used to define the data information of the target person. The intelligent agent can be a software entity or program, which is used to define and manage the data information of the target person, wherein the target person is the person who performs the target task, and can be one or more, for example, agentA, agentB. The intelligent agent can actively collect, integrate and update the relevant information of the target person. The intelligent agent can intelligently analyze and process the data information of the target person based on rules or algorithms, and provide more accurate and comprehensive data support for task matching.

[0075] The data information includes first data information and second data information. The first data information is basic information directly related to the target person, such as name, age, gender, position, etc. The first data information provides basic statistical information for the agent, which is helpful for the subsequent preliminary classification and screening of the target person. The second data information is in-depth information related to the professional skills, work experience, etc. of the target person. This information is further collected and integrated by the agent after the initial definition of the target person. The second data information can provide a more critical and specific basis for task matching, which helps to ensure that the task is assigned to the most suitable person.

[0076] For example, Figure 2 , Figure 3 As shown, in the electric power field, the first data information may be a description of the professional field of the target personnel, and the second data information may be the historical task completion status of the target personnel.

[0077] Through the construction of intelligent agents and the comprehensive collection of data information, it can be ensured that the data used in the task matching process is accurate, complete and up-to-date, and the intelligent agent can intelligently analyze and process the data information of the target personnel based on rules or algorithms. Based on accurate data and intelligent analysis, the task matching method can more accurately find the most suitable personnel to perform specific tasks, thereby optimizing task allocation and resource allocation.

[0078] S200: Building a matching model.

[0079] The matching model is used to process the target task, assign and decompose the target task, etc. In this embodiment, the matching model is a fine-tuned model. In some embodiments, a pre-trained model and a data set are obtained, wherein the data set is text data in the same field as the target task, and then the pre-trained model is fine-tuned through the data set to generate a matching model.

[0080] The pre-trained model is the language model ChatGLM3-6B. GLM (General Language Model) is a pre-trained model for natural language processing tasks. This type of model is based on the Transformer architecture and is pre-trained on a large amount of text data to complete various tasks such as language understanding, generation, and translation.

[0081] In the field of electric power, the data set is a data set in the field of electric power, for example, a data set of electric power operation, which may include text data of 1000-2000 electric power operation steps, such as Figure 4 As shown in the figure, this is a text data about the power generation operation process in the data set.

[0082] After obtaining the pre-trained model and data set, the pre-trained model is iteratively trained using the data set, and the model parameters are gradually optimized. On the basis of iterative training, LoRA (Low-Rank Adaptation) fine-tuning is performed to make the model more adaptable to the task requirements of a specific field, and then a matching model is generated for subsequent task matching tasks. Through iterative training and fine-tuning, the matching model can learn the characteristics of a specific field and improve the accuracy and efficiency of task matching. LoRA fine-tuning can achieve fast and effective adaptation to specific tasks while keeping most of the model structure unchanged, reducing the complexity and time cost of model training.

[0083] S300: Inputting the target task, data information and the first prompt word into the matching model to output the task decomposition steps through the matching model.

[0084] The data information is the information of the target person, namely, the first data information and the second data information defined by the intelligent agent. It can be understood that the data information can also be defined as more types of information, for example, including the first data information, the second data information, the third data information and the fourth data information, wherein the multiple data information can provide different information for task decomposition or matching.

[0085] Before outputting the task decomposition steps, the target task can be searched through the data set. If the steps of the target task are found in the data set, the steps in the data set can be used. For example, if the target task is a power generation operation process, Figure 4 As shown, Figure 4 The detailed steps of the power generation operation process are included in it, and this step can be used as the task decomposition step.

[0086] If the target task cannot be obtained through the data set, the matching model is required to decompose the model output task steps through the first prompt word. In some embodiments, the language processing model is embedded in the matching model to obtain an embedding model. The target task is converted into a first vector through the embedding model, and the data set is converted into a second vector. The cosine similarity of the first vector and the second vector is calculated to obtain a reference task; the reference task is input into the matching model to output the task decomposition steps through the matching model.

[0087] In some embodiments, an average pooling operation is performed on the first vector and the second vector to calculate an average value, where the average value is the average value of the first vector and the second vector; based on the average value, the cosine similarity of the first vector and the second vector is calculated; if the cosine similarity of the second vector is greater than or equal to a similarity threshold, the text data corresponding to the second vector is used as a reference task.

[0088] In order to find the text data most similar to the target task, this embodiment embeds a sentence embedding representation model sentence-BERT in the large model. This model is a natural language processing model based on BERT (Bidirectional Encoder Representations from Transformers), which is an improvement on the pre-trained BERT network. It uses Siamese Network and Triplet Network structures to output semantically valuable sentence embeddings. Cosine similarity can be used for comparison to extract the text data with the highest similarity.

[0089] Sentence embedding first embeds the target task to generate a vector of fixed length, namely the first vector, and then performs an average pooling operation to average all the vectors obtained by the sentence model. The average value is used as the sentence vector of each sentence, and the cosine similarity is calculated to obtain the sentence with the highest similarity. Each task description in the vector data set is encoded into a vector by Sentence-BERT through the above operations. The cosine similarity is calculated between the first vector and the vector of each task description in the data set, namely the second vector, to filter out the text data with the highest similarity. The formula for calculating cosine similarity is as follows:

[0090] ;

[0091] Among them, A is the first vector, B is the second vector, and Embedding is the vector generated by Sentence-BERT.

[0092] The higher the similarity score, the stronger the semantic similarity between the task descriptions. Through the embedding representation and cosine similarity calculation of Sentence-BERT, new tasks can be efficiently matched with existing data, improving the response speed of the task decomposition process. By calculating the similarity scores of the target task and all task descriptions in the dataset, the task description with the highest score is selected as the reference task, which will be used to guide the matching model to generate new task steps.

[0093] To ensure that the generated task steps are highly professional and accurate, it is necessary to guide the model generation process through prompt words, rather than relying directly on the model to generate task steps. Prompt words limit the task steps generated by the matching model to a specific range, making them more in line with the standards and requirements of the professional field. Specifically, the input of prompt words needs to match the similarity between the tasks issued by the manager and the existing tasks in the data set to select the record that is most similar to the new task.

[0094] In some embodiments, a first prompt word is generated according to the target task and the reference task, and then the target task, data information and the first prompt word are input into the matching model to output the task decomposition steps through the matching model.

[0095] The first prompt word is used to guide the matching model to output the task decomposition steps. The first prompt word is generated based on the target task and the reference task, and can be one or more phrases, sentences, or instructions, which are used to guide the matching model on how to output the task decomposition steps based on the target task and the reference task. In this embodiment, the first prompt word can be "Please generate the decomposition steps of the target task based on the reference task as an example."

[0096] like Figure 5 As shown, through the guidance of the first prompt word, the matching model can use the existing data set and combine it with the requirements of the target task to generate more professional and accurate task steps, thereby completing the task decomposition.

[0097] S400: Obtain a second prompt word.

[0098] In order to enable the matching model to accurately understand the agent's expertise and make reasonable task allocation based on the characteristics of the target task, the second prompt word can guide the matching model to output allocation suggestions after analyzing the description of the target task and the agent's expertise. The second prompt word guides the matching model to combine the requirements of the target task with the agent's professional capabilities, so that the model can deeply understand the advantages and areas of expertise of each agent, guide the model to analyze the matching relationship between tasks and agents through the second prompt word, and give recommended allocation results.

[0099] In some embodiments, the task decomposition steps are used to generate a task list, the task type in the task list is analyzed, and a second prompt word is generated according to the task type and the first data information. The second prompt word may be "analyze the task type in the task list, and match the task to the agent according to the first data information".

[0100] S500: Inputting the task decomposition steps, data information and the second prompt word into the matching model, so as to output the matching result through the matching model, so as to execute the matching result through the target person.

[0101] After obtaining the second prompt word, the task list, the first data information and the second prompt word are used as inputs. The task characteristics and the agent's expertise are analyzed and matched through the matching model's natural language understanding ability, thereby achieving appropriate docking between the task and the agent. The matching model generates matching results with suitable agents based on these inputs, ensuring the rationality of task allocation and the agent's professional field.

[0102] like Figure 6 As shown in the figure, if agentA is good at detecting and troubleshooting high-voltage equipment, agentB is good at optimizing and regulating the power dispatching system, and agentC is good at laying and maintaining power lines, when the task type is a high-voltage equipment task, then the generated matching result is that the target task matches agentA, and the target personnel corresponding to agentA can perform the target task through the task decomposition steps.

[0103] The matching model can also be called by recording the second data information of the agent, that is, the historical task completion status, so that the matching model can evaluate the agent's understanding ability and evaluate it with different standards. After outputting the evaluation conclusion, the description detail of the task can be adjusted according to different understanding abilities to ensure that the task is accurately understood and efficiently executed. By adjusting the task description in a targeted manner, the agent can fully understand the work requirements when executing the task, thereby improving work efficiency.

[0104] like Figure 7 As shown, in some embodiments, a third prompt word is generated based on the second data information, and then the second data information and the third prompt word are input into the matching model to output an evaluation result through the matching model, and the evaluation result and the matching result are input into the matching model to output a task description through the matching model.

[0105] Since the comprehension ability can be evaluated through the second data information, the third prompt word is used to guide the matching model to output the evaluation result. For example, the third prompt word can be "Please evaluate the comprehension ability of the intelligent agent based on the second data information". The evaluation results include first-category results, second-category results and third-category results. For example, the first-category result is low comprehension ability, that is, more specific task guidance and step-by-step operating instructions are required to ensure that the task can be correctly understood and completed. The second-category result is normal comprehension ability, that is, it can understand the task description more accurately and can receive the task description normally. The third-category result is high comprehension ability, that is, it can quickly understand the core content and execution requirements of the task, and has a strong thinking comprehension ability of the task.

[0106] Correspondingly, the task description includes a first category description, a second category description and a third category description, wherein the complexity of the first category description is greater than that of the second category description, and the complexity of the second category description is greater than that of the third category description.

[0107] If the evaluation result is the first type, that is, the agent's comprehension ability is low, then a more detailed task description result needs to be generated, including each operation step, so that the agent that needs to communicate multiple times to fully understand the task can receive detailed tasks, thereby improving the interaction efficiency.

[0108] By constructing the fourth prompt word, the first category of results, matching results and the fourth prompt word are input into the matching model to output the first category of instructions through the matching model, wherein the fourth prompt word is used to guide the matching model to output the first category of instructions, and the fourth prompt word can be "Please generate a detailed task description for the agent of the first category of results, and explain each operation step and precautions step by step."

[0109] If the evaluation result is the second type of result, the second type of description is output through the matching model, wherein the second type of description is the task decomposition steps. That is to say, when the agent has normal comprehension ability, it directly outputs the task decomposition steps after inputting the target task, data information and the first prompt word into the matching model.

[0110] If the evaluation result is the third category, that is, the agent has a high comprehension ability, a fifth prompt word is constructed, wherein the fifth prompt word is used to guide the matching model to output the third category description. The fifth prompt word can be "Please generate a concise task description for the agent with the third category result". The third category result and the matching result are then input into the matching model to output the third category description through the matching model.

[0111] like Figure 8As shown, different tasks can be assigned to different agents. If the evaluation result of agentA is the first type of result and the evaluation result of agentB is the third type of result, the generated descriptions are different. Through this process, this embodiment can provide different task descriptions for agents with different understanding abilities to ensure the accurate communication of task information, with flexibility and adaptability, and can be adjusted according to the actual situation of the agent, thereby improving the accuracy and efficiency of task execution.

[0112] Based on the above task matching method, Fig. 9 As shown, some embodiments of the present application also provide a task matching system, including:

[0113] An acquisition unit, used for acquiring a target task;

[0114] A construction unit, used for constructing a matching model, wherein the matching model is used for processing the target task;

[0115] A processing unit, configured to input the target task, data information, and a first prompt word into a matching model, so as to output a task decomposition step through the matching model, wherein the data information is information of a target person, and the first prompt word is used to guide the matching model to output the task decomposition step;

[0116] The acquisition unit is further used to acquire a second prompt word, where the second prompt word is used to guide the matching model to output a matching result;

[0117] The matching unit is used to input the task decomposition steps and data information into the matching model, so as to output the matching result through the matching model, so as to execute the matching result through the target person.

[0118] The present application provides a task matching method and system, the method comprising: obtaining a target task, constructing a matching model, the matching model being used to process the target task; then inputting the target task, data information and a first prompt word into the matching model, so as to output the task decomposition steps through the matching model, the data information being the information of the target person, and the first prompt word being used to guide the matching model to output the task decomposition steps; then obtaining a second prompt word, and inputting the task decomposition steps, data information and the second prompt word into the matching model, so as to output the matching result through the matching model, and execute the matching result through the target person. The method can split and match tasks through different prompt words and the language processing capability of the matching model to solve the problem of low task allocation accuracy or poor task execution efficiency.

[0119] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A task matching method, characterized in that: include: Get data information; Constructing an intelligent agent, wherein the intelligent agent is used to define data information of a target person, wherein the data information includes first data information and second data information, wherein the first data information is information related to the target person; and the second data information is information related to the professional skills and work experience of the target person; Get the target task; Constructing a matching model, wherein the matching model is used to process the target task; Inputting the target task, data information and a first prompt word into a matching model, so as to output a task decomposition step through the matching model, wherein the first prompt word is used to guide the matching model to output the task decomposition step; Decomposing the task into steps to generate a task list; Analyze the task types of the task list; generating a second prompt word according to the task type and the first data information, wherein the second prompt word is used to guide the matching model to output a matching result; Inputting the task list, the first data information and the second prompt word into the matching model to output a matching result through the matching model; generating a third prompt word according to the second data information, wherein the third prompt word is used to guide the matching model to output an evaluation result; Inputting the second data information and the third prompt word into the matching model to output an evaluation result through the matching model, wherein the evaluation result includes a first type of result, a second type of result, and a third type of result; Inputting the evaluation result and the matching result into the matching model to output a task description through the matching model, wherein the task description includes a first-category description, a second-category description, and a third-category description; The step of inputting the evaluation result and the matching result into the matching model so as to output a task description through the matching model includes: If the evaluation result is a first-category result, a fourth prompt word is generated, the first-category result, the matching result and the fourth prompt word are input into the matching model, so that the matching model outputs a first-category description, and the fourth prompt word is used to guide the matching model to output the first-category description; If the evaluation result is a second type of result, outputting a second type of description through the matching model, wherein the second type of description is a task decomposition step; If the evaluation result is a third-category result, a fifth prompt word is generated, and the third-category result, the matching result and the fifth prompt word are input into the matching model to output a third-category description through the matching model, the complexity of the first-category description is greater than that of the second-category description, and the complexity of the second-category description is greater than that of the third-category description, and the fifth prompt word is used to guide the matching model to output a third-category description.

2. The task matching method according to claim 1, characterized in that: The constructing of the matching model comprises: Obtain a pre-trained model and a data set, wherein the data set is text data in the same field as the target task; The pre-trained model is fine-tuned using the data set to generate a matching model.

3. The task matching method according to claim 2, characterized in that: The step of inputting the target task, data information and the first prompt word into a matching model to output a task decomposition step through the matching model includes: If the target task cannot be obtained through the data set, embedding a language processing model in the matching model to obtain an embedding model; Converting the target task into a first vector by the embedding model, and converting the data set into a second vector; Calculating the cosine similarity of the first vector and the second vector to obtain a reference task, where the reference task is text data in the data set that has the highest similarity to the target task; The reference task is input into the matching model to output task decomposition steps through the matching model.

4. The task matching method according to claim 3, characterized in that: The step of inputting the target task, data information and the first prompt word into a matching model to output a task decomposition step through the matching model includes: Generate a first prompt word according to the target task and the reference task; The target task, data information and the first prompt word are input into the matching model so as to output the task decomposition steps through the matching model.

5. The task matching method according to claim 3, characterized in that: The calculating the cosine similarity of the first vector and the second vector to obtain a reference task includes: Performing an average pooling operation on the first vector and the second vector to calculate an average value, where the average value is an average value of the first vector and the second vector; Calculating the cosine similarity of the first vector and the second vector according to the average value; If the cosine similarity of the second vector is greater than or equal to the similarity threshold, the text data corresponding to the second vector is used as a reference task.

6. A task matching system, characterized in that: include: An acquisition unit is used to acquire data information; construct an intelligent agent, wherein the intelligent agent is used to define the data information of a target person, wherein the data information includes first data information and second data information, wherein the first data information is information related to the target person; and the second data information is information related to the professional skills and work experience of the target person; and obtaining target tasks; A construction unit, used for constructing a matching model, wherein the matching model is used for processing the target task; a processing unit, configured to input the target task, data information, and a first prompt word into a matching model, so as to output a task decomposition step through the matching model, wherein the first prompt word is used to guide the matching model to output the task decomposition step; Matching unit for: Decomposing the task into steps to generate a task list; Analyze the task types of the task list; generating a second prompt word according to the task type and the first data information, wherein the second prompt word is used to guide the matching model to output a matching result; Inputting the task list, the first data information and the second prompt word into the matching model to output a matching result through the matching model; generating a third prompt word according to the second data information, wherein the third prompt word is used to guide the matching model to output an evaluation result; Inputting the second data information and the third prompt word into the matching model to output an evaluation result through the matching model, wherein the evaluation result includes a first type of result, a second type of result, and a third type of result; Inputting the evaluation result and the matching result into the matching model to output a task description through the matching model, wherein the task description includes a first-category description, a second-category description, and a third-category description; The step of inputting the evaluation result and the matching result into the matching model so as to output a task description through the matching model includes: If the evaluation result is a first-category result, a fourth prompt word is generated, the first-category result, the matching result and the fourth prompt word are input into the matching model, so that the matching model outputs a first-category description, and the fourth prompt word is used to guide the matching model to output the first-category description; If the evaluation result is a second type of result, outputting a second type of description through the matching model, wherein the second type of description is a task decomposition step; If the evaluation result is a third-category result, a fifth prompt word is generated, and the third-category result, the matching result and the fifth prompt word are input into the matching model to output a third-category description through the matching model, the complexity of the first-category description is greater than that of the second-category description, and the complexity of the second-category description is greater than that of the third-category description, and the fifth prompt word is used to guide the matching model to output a third-category description.

Images