Matrix project management agent building method and system based on NLP

Abstract

The invention discloses an NLP-based matrix project management agent building method and system, and the method comprises the steps: extracting entities, attributes and relationships from a preprocessed document, and building a plurality of knowledge graph trees; inputting the multiple knowledge graph trees into a first graph neural network model for aggregation training to generate a target agent, and inputting the multiple knowledge graph trees into a second graph neural network model for aggregation training to generate a utility agent; performing association integration on the target intelligent agent and the utility intelligent agent to obtain an intelligent agent system based on project management; and the intelligent agent system reasones a recommendation scheme of the current project according to the updated project management related document. Project management personnel can obtain the recommendation scheme of the project from the intelligent agent system, decision making on project management based on the recommendation scheme is facilitated, key problems in project management are more specifically solved by introducing AI and advanced data analysis technologies, and a more comprehensive and intelligent solution is provided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method and system for building a matrix project management intelligent agent based on NLP. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] In modern project management practice, the success of a project usually depends on the effective management and control of key elements such as project tasks, progress, quality and results. However, as the scale and complexity of projects increase, traditional project management methods have encountered challenges such as inefficient data processing, delayed information updates, and unreasonable resource allocation. These problems limit the efficiency of project management, hinder the realization of real-time monitoring and intelligent decision-making, and thus affect the overall performance and success rate of the project. Currently, projects generally face the following challenges:

[0004] (1) Whether a new project can quickly extract project requirements and provide the company with a forecast report on the feasibility of the project has become a key technology for enterprises to enhance their competitiveness. (2) Whether project managers can quickly find suitable personnel to participate in the project is a common problem for many companies. (3) Whether project quotations and solutions can be provided in a timely manner is a reflection of the competitiveness of the enterprise. (4) Whether the quality issues and product competitiveness of the project can be provided in real time is the key to the company's success in market competition. (5) Although existing project management software provides automation and decision-making support functions to a certain extent, most systems lack the ability to deeply analyze project data and intelligently process them. Especially when processing large amounts of unstructured data (such as meeting minutes, email communications, change requests, etc.), these systems are often unable to effectively extract key information, resulting in limited intelligence in project management.

[0005] For example, the invention patent "An Intelligent Scientific Research Project Management Method and System" (publication number: CN118690995A) provides a good framework for automation, decision support, efficiency improvement and quality improvement, but lacks a specific description of specific demand extraction, intelligent recommendation, quotation and solution generation, real-time analysis and unstructured data processing. The patent mentions that it can automatically execute multiple project management processes and reduce manual operations and errors, but does not explicitly mention the function of intelligent recommendation of team members. Although the system can provide decision support, it does not specify how to achieve personnel matching; it can also provide predictions and decision support based on historical data and real-time information, but it does not explicitly mention how to generate project quotations and solutions. Although the system can improve resource utilization efficiency and management benefits, it does not specify how to plan time schedules and work hour forecasts; it also does not explicitly mention how to process large amounts of unstructured data. Although the system can improve resource utilization efficiency and management benefits, it does not specify how to extract key information.

[0006] The goal of the existing information project management system is to achieve dynamic management of projects, improve project management efficiency and project success rate. However, the existing system still has shortcomings in solving specific problems, especially in the application of AI technology, unstructured data processing, real-time analysis and push, etc. Summary of the invention

[0007] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for building a matrix project management intelligent agent based on NLP. The intelligent agent is built to generate an Agent customer service assistant based on project management. Project managers and project-related members can consult the Agent on solution issues or assign tasks for processing.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] In a first aspect, the present invention provides a method for building a matrix project management agent based on NLP, comprising:

[0010] Obtain project management related documents, and pre-process the documents to obtain pre-processed project management related documents;

[0011] Extract entities, attributes and relationships from the preprocessed project management related documents to construct multiple knowledge graph trees;

[0012] The multiple knowledge graph trees are input into the first graph neural network model for aggregate training to generate a target agent, and the multiple knowledge graph trees are input into the second graph neural network model for aggregate training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data;

[0013] Associating and integrating the target agent and the utility agent to obtain an agent system based on project management;

[0014] The intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

[0015] According to a further technical solution, the project management related documents include requirement documents, project task orders, outline design documents, and detailed design documents.

[0016] A further technical solution is to construct a knowledge graph tree by using NLP technology to extract entities, attributes and relationships from preprocessed documents, and construct enterprise project graph trees, employee skill graph trees, enterprise project quality graph trees, employee working hours graph trees, and enterprise project results graph trees. The specific formula is:

[0017] G i =g(E i ,R i )

[0018] Among them, G i represents the knowledge graph tree in the i-th place, E i Represents an entity set, R i Represents a set of relations, and g() represents the construction process of the knowledge graph tree.

[0019] Further technical solutions, the process of constructing the knowledge graph tree is as follows:

[0020] Use the pre-trained BERT model to extract entities, attributes, and relationships from project management related documents;

[0021] The extracted entities, attributes and relations are represented as triples, which are expressed as triples of knowledge;

[0022] Fusing triples of knowledge, including entity merging and relation normalization;

[0023] According to the results of knowledge fusion, based on different attribute types, entities and relationships, we construct the enterprise project map tree, employee skill map tree, enterprise project quality map tree, employee working hours map tree and enterprise project results map tree.

[0024] A further technical solution is that the target rule is the task or activity that contributes the most to the project target and has the greatest time completion, and the utility rule is to maximize the resource utilization of the project and achieve the best project benefits.

[0025] In a further technical solution, the first graph neural network model and the second graph neural network model adopt the GraphSAGE model, and perform regular sampling according to the target rule-based sampler and the utility rule-based sampler, respectively, and train the aggregation function based on the entities, attributes and relationships obtained by sampling; the specific formula is expressed as follows:

[0026] O new =φ(D internal ,D external )

[0027] Among them, O new represents the target or utility agent, D internal Represents the knowledge graph entity attributes and relationships, D external represents the external knowledge set, and φ represents the goal or utility rule.

[0028] A further technical solution is to use the CAMEL framework to jointly associate the target agent and the utility agent. By specifying the target agent and the utility agent as two different tasks, they are collaboratively completed through the CAMEL dialogue manager in the dialogue and execution stages.

[0029] In a second aspect, the present invention provides a matrix project management agent building system based on NLP, comprising:

[0030] The data acquisition module is configured to: acquire project management related documents, and pre-process the documents to obtain pre-processed project management related documents;

[0031] A knowledge graph tree construction module is configured to: extract entities, attributes and relationships from the preprocessed project management related documents to construct multiple knowledge graph trees;

[0032] An agent generation module is configured to: input the multiple knowledge graph trees into a first graph neural network model for aggregation training to generate a target agent, and input the multiple knowledge graph trees into a second graph neural network model for aggregation training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data;

[0033] An agent integration module is configured to: associate and integrate the target agent and the utility agent to obtain an agent system based on project management;

[0034] The intelligent agent reasoning module is configured as follows: the intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

[0035] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for building a matrix project management intelligent agent based on NLP as described in the first aspect.

[0036] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for building a matrix project management intelligent agent based on NLP as described in the first aspect are implemented.

[0037] One or more of the above technical solutions have the following beneficial effects:

[0038] The present invention uses advanced natural language processing (NLP) technology to extract useful information from the unstructured data of various project materials of the enterprise, construct a project knowledge graph, and then define target-based sampling rules and utility-based sampling rules according to needs and establish corresponding samplers. The GraphSAGE model is used to perform aggregation training on them to complete the construction of the model Agent, and then the needs, working hours, personnel planning, and approximate completion progress of the newly input project can be evaluated.

[0039] The present invention can more accurately evaluate project requirements, give more accurate working hours, reasonably plan personnel arrangements, and summarize project completion for existing matrix project management through Agent. This method should be able to use advanced natural language processing (NLP) technology to extract useful information from unstructured data, build a project knowledge graph, and implement agent-based project management on this basis. By building five kinds of knowledge graph trees (enterprise project tree, employee skill tree, enterprise project quality tree, employee working time tree, enterprise project achievement tree) based on various documents (requirements documents, project task book, outline design, in-depth design, user needs confirmation, detailed design, iterative development documents, working time statistics, and results data), two intelligent entities are formed by the five knowledge graph trees according to the two key indicator rules based on goals and utility. The specific formation method is to select the corresponding aggregation function through the GraphSAGE model to complete the aggregation training based on the above rule sampler, and then generate the corresponding two intelligent agent agents. Through a unified dialogue window, the two agents work together to generate an Agent customer service assistant based on project management, and then complete the answer to the user's input questions. After that, project managers and project-related members can consult the Agent on solution issues or issue tasks for processing, and can also obtain recommended solutions for the project from the intelligent agent system, which is convenient for making decisions on project management based on the recommended solutions.

[0040] With the help of the development of artificial intelligence, the present invention can summarize reusable requirements and analyze the differences with new project requirements based on the company's historical project information, including various documents of similar projects in the past; by integrating the data in the existing OA management system into this management model, we can identify the personnel suitable for this project and intelligently recommend the corresponding team member combination; through the analysis of historical document materials and the current OA management system by this model, we can plan the corresponding time schedule and working hour forecast, which is convenient for business personnel to communicate with customers in time and negotiate project quotations; this patent uses relevant models to perform statistical analysis on the operation and maintenance logs of historical projects and customer feedback on product quality, and can generate a predicted life curve for the project, and accurately push relevant information to relevant personnel, for example, prompting product managers which products need to be iteratively updated, or which new products have appeared on the market, and our product competitiveness is weakening; by introducing AI and advanced data analysis technology, key issues in project management are solved more specifically, providing more comprehensive and intelligent solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0042] Figure 1It is a flowchart of a method for building a matrix project management agent according to an embodiment of the present invention;

[0043] Figure 2 This is a flow chart of building a knowledge graph according to an embodiment of the present invention;

[0044] Figure 3 It is a flow chart of the intelligent agent model training according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0046] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0047] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0048] In project management, the main problems faced by enterprises include: quickly refining new project requirements and providing feasibility forecast reports to enhance competitiveness; project managers need to quickly find suitable personnel to participate in the project, which is usually a problem for many enterprises; timely providing project quotations and solutions to enhance market competitiveness; real-time monitoring of project quality issues and product competitiveness to win in market competition; and the lack of in-depth analysis and intelligent processing capabilities of existing project management software when processing large amounts of unstructured data (such as meeting records, email communications, change requests, etc.), resulting in limited intelligent level of project management. The present invention, by introducing artificial intelligence and data analysis technology, can summarize reusable requirements, intelligently recommend suitable team members, plan time schedule and man-hour forecasts, generate a predicted life curve for the project, and deeply analyze unstructured data, thereby comprehensively improving the efficiency of project management and the competitiveness of enterprises.

[0049] Embodiment 1

[0050] like Figure 1 As shown, this embodiment discloses a method and system for building a matrix project management agent based on NLP, and the method includes the following steps:

[0051] S1: Obtain project management related documents, and pre-process the documents to obtain pre-processed project management related documents;

[0052] In this embodiment, project management related documents are collected, including data such as requirement documents, project task orders, outline design documents, and detailed design documents.

[0053] The collected documents are cleaned, formatted and standardized to facilitate the subsequent information extraction using natural language processing (NLP) technology. The specific formula is:

[0054] P clean =f(P raw ,T clean )

[0055] Among them, P clean represents the cleaned document set, P raw represents the original document set, T clean Indicates the processing rules for cleaning, formatting and standardization. Function f() represents the process of manually cleaning the document, including clean According to the rules, existing documents are cleaned, formatted, standardized, error detected and corrected, extracted, integrated, and quality controlled.

[0056] S2: extracting entities, attributes and relationships from the preprocessed project management related documents and constructing multiple knowledge graph trees;

[0057] In this embodiment, NLP technology is used to extract entities, attributes and relationships from preprocessed documents to construct five knowledge graph trees, including enterprise project graph tree, employee skill graph tree, enterprise project quality graph tree, employee working hours graph tree, and enterprise project results graph tree. The specific formula is:

[0058] G i =g(E i ,R i )

[0059] Among them, G i represents the knowledge graph tree in the i-th place, E i Represents an entity set, R i Represents a set of relations.

[0060] The processing of function g() is mainly divided into the following steps:

[0061] (1) Use the pre-trained BERT model. The BERT model is a language representation model based on the Transformer model. Fine-tuning is used to adapt the relationship to complete the extraction task.

[0062] (2) Specifically, from P clean Identify named entities in a document collection, such as names of people, places, organizations, etc. Relationship extraction: Extract relationships between entities from text to determine relationships between projects and employees, between projects, and between employees, such as project leaders, departments to which employees belong, etc. Attribute extraction: Extract attribute information of entities from text, such as attributes between project nodes and employee nodes, such as project budgets, employee skill levels, etc.

[0063] (3) The extracted entities, relationships, and attributes are represented as triples, and the knowledge of the represented triples is stored in a knowledge base using a graph database storage method. The graph database is significantly more efficient in associative queries than a traditional relational database. This embodiment uses the Neo4J graph database storage method.

[0064] (4) The extracted knowledge is integrated, including entity merging and relationship normalization, to ensure the consistency and accuracy of the knowledge graph. First, the fuzzy topic clustering algorithm (FADDIS) model is used to insert entity words into new upper and lower levels, discover and conceptualize newly discovered entities, and find the position of new entities in the graph by minimizing the penalty function. The penalty function mainly consists of three parts: head subjects, gaps, and offshoots. Among them, head subjects represent the intermediate nodes (concept nodes) in the graph that the fuzzy subsets finally correspond to, gaps represent the positions that can be inserted, and offshoots represent the leaf nodes that should be under the head subjects but are inserted in the wrong position. The fuzzy subset H is obtained by the Parsimonious Generalization of Fuzzy Sets algorithm. By minimizing head subjects, gaps, and offshoots, the fuzzy subset is correctly integrated into the existing graph and the overall accuracy of the graph structure is guaranteed. Then, the triple relationship is dynamically adjusted as the position of the entity word changes. During the entity alignment process, the triple relationship will also be adjusted accordingly. The attribute representations learned from the attribute triples can fall into the same vector space. Even if the attributes come from different knowledge graphs, the attribute strings are basically similar, so the character representations can be learned from the attribute strings. Then, using the attribute character representations, the entity structure representations are brought into the same vector space, so that the entity embedding can obtain the similarity between entities from the two knowledge graphs, and then dynamically adjust the triple relationship.

[0065] (5) Based on the results of knowledge fusion, five types of knowledge graph trees are constructed according to different attribute types, entities and relationships: enterprise project graph tree, employee skill graph tree, enterprise project quality graph tree, employee working hours graph tree, and enterprise project results graph tree.

[0066] S3: Input the multiple knowledge graph trees into the first graph neural network model for aggregate training to generate a target agent, and input the multiple knowledge graph trees into the second graph neural network model for aggregate training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data;

[0067] In this embodiment, clear target rules and utility rules are set for the target agent model and utility agent model to be generated. The target-based sampling rule (target rule) is: the task or activity that contributes the most to the project goal and has the greatest time completion. The utility-based sampling rule (utility rule) is: the project maximizes resource utilization and has the best project benefits. Then set the sampler according to these rules.

[0068] The goal rule is mainly focused on achieving specific goals. In this embodiment, it is mainly used to identify key information in documents and determine the goal rules of the project, such as project goals, milestones, budget and resource requirements, quality standards, etc. By analyzing historical project data, the feasibility of new projects can be learned and predicted. For the utility rule, by evaluating the utility of different project plans, decision support is provided to project managers based on factors such as risk, cost and potential benefits. In fact, the main purpose is to set the required samplers so that the subsequent GraphSAGE model can perform regular sampling according to the given samplers.

[0069] The participating entities (nodes) and the relationships between entities include project nodes (project ID, project name, project leader, start time, end time, budget, actual cost, project status, cost-effectiveness, time efficiency, quality assessment indicators) and employee nodes (employee ID, name, position, department, skill label, skill level, historical project experience, work efficiency, work quality).

[0070] The first graph neural network model GraphSAGE uses a sampler based on target rules to sample in multiple knowledge graph trees, obtains entities, attributes and relationships (relationships between entities) based on target rules for aggregate training, and generates the target agent Agent1; the second graph neural network model GraphSAGE uses a sampler based on utility rules to sample in multiple knowledge graph trees, obtains entities, attributes and relationships based on utility rules for aggregate training, and generates the utility agent Agent2. The specific formula is as follows:

[0071] O new =φ(D internal ,D external )

[0072] Among them, O new represents the new target or utility agent, D internal Represents the knowledge graph entity attributes and relationships, D external represents the external knowledge set, and φ represents the goal or utility rule.

[0073] The constructed enterprise project atlas tree, employee skill atlas tree, enterprise project quality atlas tree, employee working hours atlas tree, and enterprise project results atlas tree are aggregated and trained through the graph neural network model GraphSAGE, such as Figure 3 As shown, the specific process is:

[0074] Obtain project management related documents of historical projects and preprocess the documents to obtain preprocessed project management related documents; extract entities, attributes and relationships from the preprocessed project management related documents to construct enterprise project atlas tree, employee skill atlas tree, enterprise project quality atlas tree, employee working hours atlas tree and enterprise project achievement atlas tree;

[0075] Using various knowledge graph trees as training data, the first graph neural network model and the second graph neural network model sample nodes and neighbors in the knowledge graph according to the target rule and the utility rule respectively;

[0076] Through forward propagation, loss calculation, and back propagation to update the parameters of the agent model, it is determined whether the model has converged. If so, the model performance is evaluated. If the evaluation results show that the model performance is good, the training is terminated. If the model performance is not good, the sampling rules are adjusted. Specifically, the model training process is implemented through four key steps: forward propagation, loss calculation, back propagation, and parameter update. First, forward propagation involves passing the input data through each layer of the model until the prediction result is output. Next, the loss calculation stage uses the loss function to measure the difference between the predicted output and the true label. Subsequently, back propagation uses the gradient of the loss function to pass backward from the output layer to the input layer to calculate the gradient of the parameters of each layer. Finally, the parameter update stage adjusts the weights and biases of the model based on these gradients and optimization algorithms to reduce losses and improve model performance. This process is repeated in each training cycle until the model achieves satisfactory performance on the training data.

[0077] Define the sampling rule G respectively i and U i , namely, the goal rule and the utility rule;

[0078] According to the target rule, a target-based sampler is set, and according to the utility rule, a utility-based sampler is set. The first graph neural network model GraphSAGE and the second graph neural network model GraphSAGE perform regular sampling according to the target rule-based sampler and the utility rule-based sampler, respectively, and train aggregation functions based on the entities, attributes, and relationships obtained by sampling. Specifically, in the graph neural network model, based on the sampler, a suitable entity in the knowledge graph is searched, and the next entity is searched downward from this entity, and finally an entity set (entity, relationship, attribute) that meets the sampling rule path is formed, and the parameters required for the aggregation function are obtained. The aggregation function can use the mean aggregation function, the PoolingAggregator aggregator, etc.

[0079] The project management-related documents of historical projects are mainly used to train the initial model, that is, to generate the initial target agent and utility agent. During training, the model learns the project plan based on the data of historical projects, including selecting appropriate team members, project time schedule, project working hours, project completion time, etc. Later, during the user interaction process, the newly entered documents will complete the construction of the knowledge graph and the reasoning of the plan according to the agent system, and then answer the user's questions and recommend solutions for the user.

[0080] S4: associating and integrating the target agent and the utility agent to obtain an agent system based on project management;

[0081] In this embodiment, two different Agent agents, namely, target agent and utility agent, are formed based on the constructed knowledge graph tree and different graph neural network models (i.e., sampling rules based on goals and sampling rules based on utility). The target agent and utility agent work together through the association based on goals and utility. The collaboration between the target agent and the utility agent mainly depends on the different value orientations of the project, such as asking questions related to the goal or questions related to the project benefits, by filtering at the front end.

[0082] The specific formula is expressed as

[0083] A i =h(G i ,O i ,U i )

[0084] Among them, A i represents the i-th agent, G i Represents the corresponding knowledge graph tree, O i represents the target set, U i Indicates utility management. The h() function represents the GraphSAGE training process, such as forward propagation, calculation, back propagation, and updating model parameters.

[0085] In this embodiment, the target agent Agent1 and the utility agent Agent2 are integrated into an Agent customer service assistant based on project management (agent system based on project management), so that project managers and related members can consult solutions or assign tasks through the Agent.

[0086] The CAMEL framework is used to effectively associate the target agent Agent1 and the utility agent Agent2. By specifying the tasks, the two agents are assigned to two different tasks, and then the dialogue and execution stages are completed through the CAMEL dialogue manager.

[0087] The specific formula is:

[0088] A integrated =A 1 ⊕A 2

[0089] Among them, A integrated represents an integrated agent system, and ⊕ represents the integrated operation of the agents.

[0090] S5: The intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

[0091] In this embodiment, the agent system continuously manages and analyzes the information based on the newly collected data, optimizes and adjusts the parameters of the agent model by evaluating the effectiveness of the behavior, and thus achieves the goal more efficiently. Figure 3 As shown, the parameters are adjusted by comparing the training results of the intelligent model with the historical project results. Specifically, for the two intelligent agents based on the current goals and utilities, the task completion degree of question answering, task completion time, resource allocation efficiency, resource utilization, cost prediction accuracy, communication response time, project progress tracking, project delay rate, etc. are calculated.

[0092] The specific formula is:

[0093] S optimized =ψ(S cueernt ,D new )

[0094] Among them, S optimized represents the optimized model parameters, S cueernt represents the current model parameters, D new represents the newly collected data, and ψ represents the parameter adjustment optimization process, that is, judging whether the parameters are converged and adjusted.

[0095] The specific process is as follows: the user inputs the project document that needs to be analyzed, and the project management-based intelligent system reasons about the document, mainly based on the two rules of goal and utility. The reasoning process is relatively fast. The user consults about the current project through a dialog box, such as how many people are needed for the current project, which people are more suitable, how many working hours are estimated, etc. The intelligent system gives a recommended plan for the current project.

[0096] Embodiment 2

[0097] This embodiment discloses a matrix project management agent building system based on NLP, including:

[0098] The data acquisition module is configured to: acquire project management related documents, and pre-process the documents to obtain pre-processed project management related documents;

[0099] A knowledge graph tree construction module is configured to: extract entities, attributes and relationships from the preprocessed project management related documents to construct multiple knowledge graph trees;

[0100] An agent generation module is configured to: input the multiple knowledge graph trees into a first graph neural network model for aggregation training to generate a target agent, and input the multiple knowledge graph trees into a second graph neural network model for aggregation training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data;

[0101] An agent integration module is configured to: associate and integrate the target agent and the utility agent to obtain an agent system based on project management;

[0102] The intelligent agent reasoning module is configured as follows: the intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

[0103] Embodiment 3

[0104] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0105] Embodiment 4

[0106] The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method of embodiment 1 are performed.

[0107] The steps involved in the apparatus of the above embodiments 3 and 4 correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0108] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0110] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for building a matrix project management agent based on NLP, characterized in that: include: Obtain project management related documents, and pre-process the documents to obtain pre-processed project management related documents; Extract entities, attributes and relationships from the preprocessed project management related documents to construct multiple knowledge graph trees; The multiple knowledge graph trees are input into the first graph neural network model for aggregate training to generate a target agent, and the multiple knowledge graph trees are input into the second graph neural network model for aggregate training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data; Associating and integrating the target agent and the utility agent to obtain an agent system based on project management; The intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

2. A method for building a matrix project management agent based on NLP as claimed in claim 1, characterized in that: The project management related documents include requirement documents, project task orders, outline design documents, and detailed design documents.

3. A method for building a matrix project management agent based on NLP as claimed in claim 1, characterized in that: The specific steps of constructing the knowledge graph tree are as follows: using NLP technology to extract entities, attributes and relationships from pre-processed documents, and constructing enterprise project graph trees, employee skill graph trees, enterprise project quality graph trees, employee working hours graph trees, and enterprise project results graph trees; The specific formula is: G i =g(E i ,R i ) Among them, G i represents the knowledge graph tree in the i-th place, E i Represents an entity set, R i Represents a set of relations, and g() represents the construction process of the knowledge graph tree.

4. A method for building a matrix project management agent based on NLP as claimed in claim 3, characterized in that: The specific process of constructing a knowledge graph tree is as follows: Use the pre-trained BERT model to extract entities, attributes, and relationships from project management related documents; The extracted entities, attributes and relations are represented as triples, which are expressed as triples of knowledge; Fusing triples of knowledge, including entity merging and relation normalization; According to the results of knowledge fusion, based on different attribute types, entities and relationships, we construct the enterprise project map tree, employee skill map tree, enterprise project quality map tree, employee working hours map tree and enterprise project results map tree.

5. A method for building a matrix project management agent based on NLP as claimed in claim 1, characterized in that: The target rule is the task or activity that contributes most to the project target and has the greatest time completion, and the utility rule is to maximize the resource utilization and the best project benefits.

6. A method for building a matrix project management agent based on NLP as claimed in claim 1, characterized in that: The first graph neural network model and the second graph neural network model adopt the GraphSAGE model, and perform regular sampling according to the target rule-based sampler and the utility rule-based sampler, respectively, and train the aggregation function based on the entities, attributes and relationships obtained by sampling; the specific formula is expressed as follows: The new =φ(D internal ,D external ) Among them, O new represents the target or utility agent, D internal Represents the knowledge graph entity attributes and relationships, D external represents the external knowledge set, and φ represents the goal or utility rule.

7. A method for building a matrix project management agent based on NLP as claimed in claim 1, characterized in that: The CAMEL framework is used to jointly associate the target agent and the utility agent. By specifying the target agent and the utility agent as two different tasks, they are collaboratively completed through the CAMEL dialogue manager in the dialogue and execution stages.

8. A matrix project management agent building system based on NLP, characterized in that: include: The data acquisition module is configured to: acquire project management related documents, and pre-process the documents to obtain pre-processed project management related documents; A knowledge graph tree construction module is configured to: extract entities, attributes and relationships from the preprocessed project management related documents to construct multiple knowledge graph trees; An agent generation module is configured to: input the multiple knowledge graph trees into a first graph neural network model for aggregation training to generate a target agent, and input the multiple knowledge graph trees into a second graph neural network model for aggregation training to generate a utility agent; the first graph neural network model uses a sampler established based on a target rule to sample multiple knowledge graph data, and the second graph neural network model uses a sampler established based on a utility rule to sample multiple knowledge graph data; An agent integration module is configured to: associate and integrate the target agent and the utility agent to obtain an agent system based on project management; The intelligent agent reasoning module is configured as follows: the intelligent agent system infers a recommended solution for the current project based on updated project management related documents.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for building a matrix project management intelligent agent based on NLP as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for building a matrix project management intelligent agent based on NLP as described in any one of claims 1-7 are implemented.

Images