Intelligent decision-making auxiliary system based on AI intelligent agent

Abstract

The invention provides an intelligent decision-making auxiliary system based on an AI agent, and the system comprises six modules: a demand input module, a data collection module, a graph construction module, a model training module, an intelligent decision-making module, and a visualization module: after a user submits a decision-making demand through the demand input module, the data collection module automatically collects related data, and constructs a domain knowledge graph through the graph construction module; the model training module is used for training adaptive machine learning models for different decision-making tasks; the intelligent decision-making module generates a decision-making scheme through an AI intelligent agent in combination with knowledge graph retrieval and model reasoning, finally the visualization module presents the decision-making scheme through a visualization interface, the system creatively fuses the knowledge graph and the machine learning technology, can deal with multiple types of complex decision-making problems, adapts to dynamic decision-making scenes, and compared with traditional manual decision making, the decision-making efficiency is greatly improved. The decision-making efficiency and quality are obviously improved, and human factor interference is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of AI agents, and particularly to an intelligent decision-making assistance system based on AI agents. Background Art

[0002] In today's complex and rapidly changing business environment, decision-makers are faced with the challenge of a vast amount of information and need to make accurate and timely decisions within a limited time. With the development of information technology, people have started to use decision-making systems based on data analysis to process the vast amount of information and provide decision-making assistance for decision-makers. Traditional decision-making systems, especially those that do not rely on advanced artificial intelligence technologies, have the following main drawbacks: First, they lack flexibility. Traditional decision-making systems are usually based on preset rules and logics, and once these rules are set, it is difficult to adapt to new or changed environments. Second, their ability to handle complex problems is limited. Faced with complex decision-making problems, the system is difficult to extract useful information from complex data, resulting in poor decision-making effects. Third, they highly rely on expert knowledge. Traditional systems often require experts in relevant fields to formulate rules and highly rely on the knowledge and experience of experts. If the expert knowledge is insufficient or outdated, the decision-making quality of the system will be greatly affected. Fourth, the data utilization efficiency is low. Traditional systems may not be able to effectively process and analyze a large amount of data, thus unable to discover patterns and trends in the data, resulting in low data utilization efficiency. Fifth, the data security is poor. When processing a vast amount of data, there is a lack of effective protection for data security, which is prone to data leakage and abuse. Sixth, the transparency and interpretability are low. Users do not understand the decision-making principle and progress of the system. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent decision-making assistance system based on AI agents, which provides real-time, efficient, and accurate decision-making support for users by integrating technologies such as machine learning and knowledge graphs.

[0004] To achieve the above-mentioned invention purpose, the present invention provides an intelligent decision-making assistance system based on AI agents, and the system includes:

[0005] A requirement input module, which is used for users to input decision requirement information and determine a decision problem according to the decision requirement information;

[0006] A data collection module, which is used to determine a data source associated with the decision problem, collect raw data from the data source through a data interface, and preprocess the raw data;

[0007] A graph construction module, which is used to determine a decision domain according to the decision problem, determine a knowledge system according to the decision domain, and extract entities and relationships from the raw data based on the knowledge system to construct a knowledge graph;

[0008] A model training module, which is used to build an adapted machine learning model for different types of decision-making tasks, pre-train the machine learning model with raw data, and adjust the parameters of the machine learning model;

[0009] An intelligent decision-making module, which is used to search for relevant information in the knowledge graph according to the decision-making problem through an AI agent, input the search results into the machine learning model, and output a decision-making plan;

[0010] A visualization module, which is used to provide an input window for the requirement input module and display the decision-making plan.

[0011] Furthermore, the raw data includes structured data and unstructured data, and preprocessing the raw data includes data cleaning, data transformation, and integration processing.

[0012] Furthermore, the graph construction module specifically includes:

[0013] A subject construction sub-module, which is used to determine the decision-making field according to the decision-making problem, obtain the knowledge system of the corresponding field according to the decision-making field, extract ontologies and concepts according to the knowledge system, and abstract and transform the knowledge norms of the decision-making field into a standardized expression;

[0014] A knowledge extraction sub-module, which is used to extract information related to the decision-making problem from the preprocessed raw data, perform named entity recognition and relationship extraction on the extracted information, and convert it into triple data;

[0015] A graph storage sub-module, which is used to store the triple data in a graph database.

[0016] Furthermore, the data collection module is specifically used to perform the following operations:

[0017] Send a data acquisition request to the data source through the data interface;

[0018] Receive the processing feedback information of the data source for the data acquisition request and parse it;

[0019] If the processing feedback information indicates that the requested data to be acquired is confidential data and the raw data cannot be directly acquired, a distributed learning instruction is sent to the graph construction module and the model training module.

[0020] Furthermore, after the graph construction module receives the distributed learning instruction, the following operations are performed:

[0021] Package a toolkit, which includes a named entity recognition tool, a relationship extraction tool, and a data normalization tool, determine the data source for providing the raw data, and send the packaged toolkit to the data source through the data interface;

[0022] The data source processes the original data locally using a toolkit to obtain the recognition results of the named entity recognition tool and the extraction results of the relationship extraction tool, and inputs the recognition results and extraction results into the data normalization tool to obtain triple data;

[0023] The triple data is sent to the knowledge graph construction module through a data interface to construct a knowledge graph.

[0024] Furthermore, after the model training module receives the distributed learning instruction, the following operations are performed:

[0025] Construct a scaled model, determine the data source for providing the original data, and send the basic information of the scaled model, including the model structure and model parameters, to the data source that can provide the original data through a data interface;

[0026] The data source obtains the basic information of the scaled model through a data interface to construct a local model, inputs the original data into the local model for training, and optimizes the local model parameters to obtain optimized model parameters;

[0027] The data source feeds back the optimized model parameters to the model training module through a data interface. The model training module aggregates the optimized model parameters of different data sources and updates the global model according to the aggregated optimized model parameters. Through multiple rounds of iterative optimization, the final global model is obtained.

[0028] Furthermore, when the data source inputs the original data into the local model for training, the following operations are performed:

[0029] Judge the confidentiality level of the original data and determine the confidentiality level coefficient according to the confidentiality level of the original data;

[0030] Calculate the Gaussian noise distribution according to the confidentiality level coefficient and the query function sensitivity. Add Gaussian perturbation to the original data according to the Gaussian noise distribution. The query function is used to query the original data, and the query function sensitivity is used to measure the maximum difference between the target original data and the similar data.

[0031] Furthermore, the visualization module specifically includes:

[0032] A material acquisition sub-module, which is used to determine the decision scenario according to the decision task type and obtain the virtual scenario materials and AI agent virtual images corresponding to the decision scenario from a preset material library;

[0033] An animation rendering sub-module, which is used to add the AI agent virtual image to the virtual scenario according to the decision task process information, and add animation effects to the virtual scenario and the AI agent virtual image to generate a decision task process animation scene;

[0034] A display sub-module for displaying an animation scene of the decision-making task process to the user;

[0035] The decision-making task process information includes the construction process and real-time construction progress of the graph construction module, the training process and real-time training progress of the model training module, and the decision-making process and real-time decision-making progress of the intelligent decision-making module.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. The system collects data related to the decision-making problem based on the decision-making requirement information input by the user, extracts entities and relationships from the original data according to the decision-making field to construct a knowledge graph, and constructs a machine learning model adapted thereto based on the decision-making task type. The AI intelligent agent searches for relevant information in the knowledge graph according to the decision-making problem and inputs it into the machine learning model, so as to obtain a decision-making plan, enabling the present invention to handle different types of complex decision-making problems, adapt to complex and changeable decision-making scenarios, and provide flexible and reliable decision-making support for decision-makers;

[0038] 2. Compared with the traditional manual decision-making process, the system can greatly shorten the decision-making time, improve the decision-making efficiency, and avoid the interference of human factors on the decision-making result, and can improve the objectivity and accuracy of the decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only the preferred embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0040] Figure 1 It is a schematic diagram of the overall structure of an intelligent decision-making assistance system based on an AI intelligent agent provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following describes the principles and features of the present invention with reference to the drawings. The listed embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0042] Refer to Figure 1 , this embodiment provides an intelligent decision-making assistance system based on an AI intelligent agent. The system includes a requirement input module, a data collection module, a graph construction module, an intelligent decision-making module, and a visualization module.

[0043] Among them, the requirement input module is used for users to input decision requirement information and determine decision problems according to the decision requirement information. The decision requirement information is used to describe in which field the user needs to obtain what kind of decision support to achieve what purpose. Exemplarily, determining decision problems according to the decision requirement information can be realized through natural language processing technology.

[0044] The data collection module is used to determine data sources associated with the decision problem, collect raw data from the data sources through a data interface, and preprocess the raw data.

[0045] The data sources should be closely related to the decision problem. Exemplarily, the data sources can be core business systems such as ERP and CRM within an enterprise, or external professional data providers, such as authoritative market research institutions, financial data platforms, and real-time updated news on the Internet.

[0046] Exemplarily, preprocessing the raw data includes systematically cleaning, transforming, and integrating the collected raw data. The cleaning link aims to remove noise and duplicate information in the data and purify the data quality; the transformation process unifies the data into a standard format, such as standardizing the date format, to ensure data consistency; the integration step integrates data from different sources into a data warehouse for convenient subsequent unified analysis and management.

[0047] Exemplarily, the raw data includes structured data and unstructured data. Custom-developed adapted data interfaces are developed for different types of data sources. For structured databases, SQL language is used to efficiently extract the required data; for unstructured web information, web crawler technology is used to accurately capture it, and stable docking with external data providers is achieved through API.

[0048] The knowledge graph construction module is used to determine the decision field according to the decision problem, determine the knowledge system according to the decision field, and extract entities and relationships from the raw data based on the knowledge system to construct a knowledge graph.

[0049] The model training module is used to construct an adapted machine learning model for different types of decision tasks, pre-train the machine learning model with the raw data, and adjust the parameters of the machine learning model.

[0050] The intelligent decision module is used to search for relevant information in the knowledge graph according to the decision problem through an AI agent, input the search results into the machine learning model, and output a decision plan.

[0051] The visualization module is used to provide an input window for the requirement input module and display the decision plan.

[0052] The system provided in this embodiment collects data related to decision-making problems based on the decision-making requirement information input by the user, extracts entities and relationships from the original data according to the decision-making field to construct a knowledge graph, constructs a machine learning model adapted to it based on the decision-making task type, and the AI agent searches for relevant information in the knowledge graph according to the decision-making problem to input into the machine learning model, so as to obtain a decision-making plan, enabling the system to process different types of complex decision-making problems, adapt to complex and changeable decision-making scenarios, and provide flexible and reliable decision-making support for decision-makers.

[0053] As a possible implementation manner, the graph construction module specifically includes a subject construction sub-module, a knowledge extraction sub-module, and a graph storage sub-module.

[0054] Among them, the subject construction sub-module is used to determine the decision-making field according to the decision-making problem, obtain the knowledge system of the corresponding field according to the decision-making field, extract the ontology and concepts according to the knowledge system, and abstract and convert the knowledge specification of the decision-making field into a standardized expression.

[0055] In this implementation manner, the subject construction sub-module realizes the expression, sharing, and reuse of knowledge by abstracting and describing the domain knowledge system. For the knowledge in the knowledge system, it can be expressed in a standardized manner through the Resource Description Framework. The core of the Resource Description Framework is to adopt a triple data model to achieve accurate description and standardized expression of resources. The triple data model includes a subject, a predicate, and an object, which can accurately express various relationships and information in different industries, so as to realize the structured description and unified representation of resources in the decision-making field.

[0056] The knowledge extraction sub-module is used to extract information related to the decision-making problem from the pre-processed original data, perform named entity recognition and relationship extraction on the extracted information, and convert it into triple data.

[0057] In this implementation manner, in the process of extracting information related to the decision-making problem, the knowledge extraction sub-module extracts key contents such as entities, relationships, and attributes, and then sorts the extracted contents into structured data in a unified triple format according to the subject structure to ensure the standardization and unity of the data.

[0058] The extraction of entities is realized through named entity recognition technology, and its goal is to accurately identify entities with specific meanings from the data and classify the entities according to their attributes.

[0059] The goal of relationship extraction is to further extract the semantic relationships between these entities after identifying the entities from the data. Exemplarily, the extraction of relationships can be realized through the HanLP toolkit.

[0060] In this embodiment, during the knowledge extraction process, it is also necessary to merge entities with different names but the same meaning. Exemplarily, a method of calculating character similarity can be used to calculate the similarity of entity names. By comparing the similarity of character sequences, it is determined whether different names correspond to the same entity. If so, these names are normalized to a unique entity object to ensure that each entity in the knowledge graph has a unique name identifier, thereby eliminating the redundancy and non-uniqueness of the knowledge graph.

[0061] The graph storage sub-module is used to store the triple data into a graph database.

[0062] Exemplarily, the graph database can use Neo4j. Compared with traditional relational databases, Neo4j shows entities and relationships more intuitively and can be queried in a visual way, which is more convenient for management.

[0063] As another alternative embodiment, the data collection module is specifically used to perform the following operations:

[0064] S11. Send a data acquisition request to the data source through the data interface.

[0065] S12. Receive the processing feedback information of the data source for the data acquisition request and parse it.

[0066] In this embodiment, after receiving the data acquisition request, the data source judges the request content to determine whether it can respond to the data acquisition request, and then generates processing feedback information based on the judgment result of whether it can respond and feeds it back to the data collection module through the data interface.

[0067] S13. If the processing feedback information indicates that the requested data to be acquired is confidential data and the original data cannot be directly acquired, send a distributed learning instruction to the graph construction module and the model training module.

[0068] In this embodiment, for the content-sensitive confidential data, the data source cannot directly provide downloads externally. In this case, the data collection module cannot directly obtain the original data through the data interface and save it locally. It is necessary to send a distributed learning instruction to the graph construction module and the model training module to realize the construction of the knowledge graph and the model training when the data cannot be saved locally.

[0069] As a further possible embodiment, after the graph construction module receives the distributed learning instruction, it performs the following operations:

[0070] S21. Package the toolkit, which includes a named entity recognition tool, a relationship extraction tool, and a data normalization tool. Determine the data source for providing the original data, and send the packaged toolkit to the data source through the data interface.

[0071] S22. The data source uses a toolkit to process the original data locally, obtains the recognition results of the named entity recognition tool and the extraction results of the relationship extraction tool, and inputs the recognition results and extraction results into the data normalization tool to obtain triple data.

[0072] In this embodiment, the named entity recognition tool is used to identify and extract entity information from the local data of the data source. The relationship extraction tool is used to further identify and extract the relationship information between different entities from the local data of the data source. The data normalization tool is used to organize and obtain triple data based on the entity information extracted by the named entity recognition tool and the relationship information extracted by the relationship extraction tool.

[0073] S23. Send the triple data to the graph construction module through the data interface to construct a knowledge graph.

[0074] In this embodiment, when the data collection module cannot save the original data locally, the graph construction module encapsulates a toolkit including the named entity recognition tool, the relationship extraction tool, and the data normalization tool, and transmits it to the data source through the data interface. The data source uses the tools in the toolkit to process the original data locally, thereby obtaining triple data, and transmits it back to the graph construction module through the data interface, so that the graph construction module can obtain triple data to construct a knowledge graph without directly obtaining the original data, thus meeting the requirements of data privacy management.

[0075] Meanwhile, after the model training module receives the distributed learning instruction, the following operations are performed:

[0076] S31. Construct a scaled model, determine the data source for providing the original data, and send the basic information of the scaled model, including the model structure and model parameters, to the data source that can provide the original data through the data interface.

[0077] S32. The data source obtains the basic information of the scaled model through the data interface to construct a local model, inputs the original data into the local model for training, and optimizes the local model parameters to obtain optimized model parameters.

[0078] S33. The data source feeds back the optimized model parameters to the model training module through the data interface. The model training module aggregates the optimized model parameters of different data sources, updates the global model according to the aggregated optimized model parameters, and obtains the final global model through multiple rounds of iterative optimization.

[0079] In this embodiment, when the data collection module fails to save the original data locally, the model training module constructs a scaled model and sends the basic information of the scaled model to the data source through a data interface, enabling the data source to construct a model locally according to the basic information of the scaled model, input the original data into the local model for training, optimize the parameters of the local model, obtain the optimized model parameters, and feedback them to the model training module. The model training module updates the global model by summarizing the optimized parameters fed back by each data source until the model converges. Thus, the model training module can train the model with the original data without directly obtaining and saving the original data, meeting the model training requirements and data privacy protection requirements.

[0080] As a further optional embodiment, when the data source inputs the original data into the local model for training, the following operations are performed:

[0081] S41. Determine the confidentiality level of the original data and determine the confidentiality level coefficient according to the confidentiality level of the original data.

[0082] In this embodiment, for the original data with different privacy protection requirements, corresponding confidentiality levels are assigned. The privacy protection requirements are positively correlated with the confidentiality levels, that is, the greater the privacy protection requirements, the higher the corresponding confidentiality level. Different confidentiality levels correspond to different confidentiality level coefficients.

[0083] S42. Calculate the Gaussian noise distribution according to the confidentiality level coefficient and the query function sensitivity, and add Gaussian perturbation to the original data according to the Gaussian noise distribution. The query function is used to query the original data, and the query function sensitivity is used to measure the maximum difference between the target original data and the approximate data.

[0084] In this embodiment, data privacy protection is achieved by adding Gaussian perturbation to the original data, and the calculation of Gaussian perturbation is affected by the confidentiality level of the original data, so as to adapt to different levels of data protection requirements.

[0085] As another optional embodiment, the visualization module specifically includes a material acquisition sub-module, an animation rendering sub-module, and a display sub-module.

[0086] Among them, the material acquisition sub-module is used to determine the decision scenario according to the decision task type and obtain the virtual scene materials and AI agent virtual images of the corresponding decision scenario from the preset material library.

[0087] The animation rendering sub-module is used to add the AI agent virtual image to the virtual scene according to the decision task process information, and add animation effects to the virtual scene and the AI agent virtual image to generate a decision task process animation scene.

[0088] The display sub-module is used to display the animation scene of the decision-making task process to the user.

[0089] The decision-making task process information includes the construction process and real-time construction progress of the graph construction module, the training process and real-time training progress of the model training module, and the decision-making process and real-time decision-making progress of the intelligent decision-making module.

[0090] In this embodiment, the visualization module determines the corresponding decision-making scene based on the type of the decision-making task, and obtains the corresponding virtual scene material and AI intelligent agent virtual image material from the material library based on this. After obtaining the materials, the animation rendering sub-module adds the AI intelligent agent virtual image to the virtual scene according to the decision-making task process information, and adds animation effects to the virtual scene and the AI intelligent agent virtual image to generate the animation scene of the decision-making task process, and displays it to the user. In the animation scene of the decision-making task process, the AI intelligent agent virtual image will perform actions and operate virtual devices in the decision-making scene according to the decision-making task process information to intuitively show the whole process and real-time progress of the decision-making task, thereby improving the transparency and interpretability of the system, and enabling the user to intuitively understand the decision-making process of the AI intelligent agent.

[0091] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent decision-making assistance system based on an AI agent, characterized in that, The system includes: A requirement input module for a user to input decision requirement information and determine a decision problem according to the decision requirement information; A data collection module for determining a data source associated with the decision problem, collecting raw data from the data source through a data interface, and preprocessing the raw data; A knowledge graph construction module for determining a decision domain according to the decision problem, determining a knowledge system according to the decision domain, and extracting entities and relationships from the raw data based on the knowledge system to construct a knowledge graph; A model training module for constructing an adapted machine learning model for different decision task types, pre-training the machine learning model with the raw data, and adjusting the parameters of the machine learning model; An intelligent decision-making module for an AI agent to search for relevant information in the knowledge graph according to the decision problem, input the search results into the machine learning model, and output a decision-making plan; A visualization module for providing an input window of the requirement input module and displaying the decision-making plan.

2. An intelligent decision-making assistance system based on an AI agent according to claim 1, characterized in that, The raw data includes structured data and unstructured data. Preprocessing the raw data includes data cleaning, data transformation, and integration processing.

3. An intelligent decision-making assistance system based on an AI agent according to claim 1, characterized in that The knowledge graph construction module specifically includes: A subject construction sub-module for determining a decision domain according to the decision problem, obtaining a knowledge system of the corresponding domain according to the decision domain, extracting ontologies and concepts according to the knowledge system, abstracting the knowledge specification of the decision domain, and converting it into a standardized expression; A knowledge extraction sub-module for extracting information related to the decision problem from the preprocessed raw data, performing named entity recognition and relationship extraction on the extracted information, and converting it into triple data; A knowledge graph storage sub-module for storing the triple data in a graph database.

4. An intelligent decision-making assistance system based on an AI agent according to claim 1, characterized in that, The data collection module is specifically used to perform the following operations: Send a data acquisition request to the data source through the data interface; Receive the processing feedback information of the data source for the data acquisition request and parse it; If the processing feedback information indicates that the requested data to be acquired is confidential data and the raw data cannot be directly acquired, send a distributed learning instruction to the knowledge graph construction module and the model training module.

5. An intelligent decision-making assistance system based on an AI agent according to claim 4, characterized in that, After the knowledge graph construction module receives the distributed learning instruction, perform the following operations: Package a tool kit, which includes a named entity recognition tool, a relationship extraction tool, and a data normalization tool. Determine the data source for providing the raw data, and send the packaged tool kit to the data source through the data interface; The data source uses the tool kit to process the raw data locally, obtains the recognition result of the named entity recognition tool and the extraction result of the relationship extraction tool, inputs the recognition result and the extraction result into the data normalization tool, and obtains triple data; Send the triple data to the knowledge graph construction module through the data interface to construct a knowledge graph.

6. An intelligent decision-making assistance system based on an AI agent according to claim 4, characterized in that, After the model training module receives the distributed learning instruction, perform the following operations: Construct a scaled model, determine the data source for providing the raw data, and send the basic information of the scaled model to the data source that can provide the raw data through the data interface. The basic information of the scaled model includes the model structure and model parameters; The data source obtains the basic information of the scaled model through the data interface to construct a local model, inputs the original data into the local model for training, optimizes the local model parameters, and obtains optimized model parameters; The data source feeds back the optimized model parameters to the model training module through the data interface. The model training module aggregates the optimized model parameters of different data sources, updates the global model according to the aggregated optimized model parameters, and obtains the final global model through multiple rounds of iterative optimization.

7. An intelligent decision-making assistance system based on an AI agent according to claim 6, characterized in that, When the data source inputs the original data into the local model for training, the following operations are performed: Judge the confidentiality level of the original data, and determine the confidentiality level coefficient according to the confidentiality level of the original data; Calculate the Gaussian noise distribution according to the confidentiality level coefficient and the query function sensitivity, and add Gaussian perturbation to the original data according to the Gaussian noise distribution. The query function is used to query the original data, and the query function sensitivity is used to measure the maximum difference between the target original data and the approximate data.

8. An intelligent decision-making assistance system based on an AI agent according to claim 1, characterized in that, The visualization module specifically includes: The material acquisition sub-module is used to determine the decision scenario according to the decision task type, and obtain the virtual scene materials and AI agent virtual images of the corresponding decision scenario from the preset material library; The animation rendering sub-module is used to add the AI agent virtual image to the virtual scene according to the decision task process information, and add animation effects to the virtual scene and the AI agent virtual image to generate a decision task process animation scene; The display sub-module is used to display the decision task process animation scene to the user; The decision task process information includes the construction process and real-time construction progress of the graph construction module, the training process and real-time training progress of the model training module, and the decision process and real-time decision progress of the intelligent decision module.

Images