Knowledge graph-guided agent data full-life-cycle management method and device

Abstract

The invention discloses a knowledge graph-guided agent data full-life-cycle management method and device, and relates to the technical field of agents. The method comprises the following steps: connecting a data management platform to an intelligent agent of a client, and constructing a core knowledge graph; in the data acquisition stage, according to a task request, a core knowledge graph is used for intelligent guidance, and data acquisition is carried out; in the data preprocessing stage, a core knowledge graph is used for performing intelligent treatment on a plurality of original data; in a data storage stage, intelligently organizing a plurality of preprocessed data by using a core knowledge graph; in a data application stage, performing intelligent service by using the updated core knowledge graph according to task information; and in a data archiving stage, performing intelligent processing on the secondarily updated core knowledge graph. The problems that in the prior art, data collection blindness is high, data management cost is high, data organization is disordered, data application and data management are disjointed, and a data archiving strategy is rigid are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent agent technology, and in particular to a knowledge graph-guided method and apparatus for the full lifecycle management of intelligent agent data. Background Technology

[0002] With the rapid development of artificial intelligence technology, various intelligent agents (such as intelligent customer service, autonomous driving systems, personalized recommendation engines, and industrial robots) have been widely applied to all aspects of social production and life. The performance and intelligence level of these intelligent agents highly depend on the support of high-quality, large-scale, and diverse data. However, the data generated by intelligent agents during their operation is characterized by its wide range of sources, complex structure, dynamic evolution, and uneven value density, posing a significant challenge to data management.

[0003] Traditional data management methods typically employ a siloed or isolated management model, optimizing for a specific stage of the data lifecycle (such as collection, storage, and analysis), lacking a holistic and coherent approach. This leads to a series of problems:

[0004] 1) Data collection is highly unpredictable and difficult to accurately match the dynamic needs of intelligent agents;

[0005] 2) Data governance is costly, relies heavily on manual rules and scripts, and is inefficient and prone to errors;

[0006] 3) The data is poorly organized, lacks effective semantic relationships between data, and makes it difficult to achieve cross-source and cross-domain knowledge discovery and reuse;

[0007] 4) Data application and data management are disconnected, making it impossible to dynamically evaluate and optimize agent data based on the actual usage behavior of the agent;

[0008] 5) Rigid data archiving strategies fail to effectively identify and dispose of low-value data, resulting in wasted storage resources and bloated data assets.

[0009] Therefore, how to provide a solution that enables intelligent, automated, and semantic management of data throughout its entire lifecycle, and guides the entire lifecycle management process of agent data from collection, governance, storage, application to archiving, in order to improve the utilization efficiency and value of agent data and support the continuous learning and evolution of agents, is a technical problem that urgently needs to be solved. Summary of the Invention

[0010] This invention provides a knowledge graph-guided method and apparatus for managing the entire lifecycle of intelligent agent data, which solves the problems of high data collection blindness, high data governance costs, chaotic data organization, disconnect between data application and data management, and rigid data archiving strategies in existing technologies.

[0011] In a first aspect, embodiments of the present invention provide a knowledge graph-guided method for managing the entire lifecycle of intelligent agent data, the method comprising:

[0012] Connect the data management platform to the client's intelligent agent, and build the corresponding core knowledge graph in the cloud server based on the agent's metadata;

[0013] During the data acquisition phase, based on the agent's task request, the core knowledge graph is used for intelligent guidance to collect data from the data source and obtain a number of raw data.

[0014] In the data preprocessing stage, core knowledge graphs are used to intelligently manage several raw data sets to obtain several preprocessed data sets.

[0015] During the data storage phase, the core knowledge graph is used to intelligently organize several preprocessed data to obtain an updated core knowledge graph.

[0016] In the data application phase, based on the agent's task information, the updated core knowledge graph is used to provide intelligent services, support the agent's tasks, and record the agent's data usage behavior to obtain a second-updated core knowledge graph.

[0017] During the data archiving stage, the core knowledge graph updated twice is intelligently processed to eliminate low-value data entities, resulting in a core knowledge graph updated three times.

[0018] The technical solution provided in this application has at least the following beneficial effects:

[0019] For the first time, knowledge graphs are used as the core engine, running through the entire lifecycle of data collection, governance, storage, application, and archiving. This forms a data-driven, knowledge-guided, and self-optimizing closed-loop management system, breaking down the stage barriers of traditional intelligent agent data management. In the data collection stage, knowledge graphs are used to understand task requirements, and combined with improved optimization algorithms to plan the optimal collection path, achieving a shift from "blind collection" to "precise collection on demand," significantly improving data collection efficiency and quality. In the data governance and storage stages, the semantic capabilities of knowledge graphs are used for data cleaning, labeling, and association, establishing effective semantic relationships between data and integrating heterogeneous data. Transforming data into structured and semantic knowledge entities greatly enhances data usability and knowledge discovery capabilities, reduces data governance costs and error rates, and improves data governance efficiency. In the data application stage, it not only provides precise knowledge services to intelligent agents but also feeds back the agents' usage behavior to the knowledge graph, enabling the knowledge graph to dynamically evolve according to actual application scenarios and achieve continuous value-added of intelligent agent data. In the data archiving stage, based on the multi-dimensional information accumulated in the knowledge graph, it performs value prediction and intelligent processing of data entities, realizing lean management of intelligent agent data, saving storage costs, and ensuring the "health" of intelligent agent data.

[0020] In one alternative implementation, the data management platform is connected to the client's intelligent agent, and based on the agent's application scenario information, a corresponding core knowledge graph is constructed on a cloud server, including:

[0021] Log in to the data management platform on the client side and connect the data management platform to the intelligent agent stored on the client's local storage;

[0022] The data management platform is used to collect metadata of the intelligent agent and upload the metadata to the cloud server through an encrypted and secure channel;

[0023] Based on the agent's metadata, a corresponding core knowledge graph is constructed on a cloud server.

[0024] In one alternative implementation, a corresponding core knowledge graph is constructed on a cloud server based on the agent's metadata, including:

[0025] Define a multi-dimensional ontology layer for the core knowledge graph on a cloud server;

[0026] Structured, semi-structured, and unstructured data are extracted from the metadata of the intelligent agent and populated into the multi-dimensional ontology layer to obtain the initial core knowledge graph.

[0027] A dynamic evolution mechanism is constructed for the initial core knowledge graph to obtain the final core knowledge graph.

[0028] In one alternative implementation, during the data acquisition phase, based on the agent's task request, a core knowledge graph is used for intelligent guidance to acquire data from the data source, obtaining several raw data, including:

[0029] During the data acquisition phase, a data management platform is used to collect task requests from intelligent agents and upload these requests to a cloud server via an encrypted and secure channel.

[0030] In the cloud server, a pre-trained entity recognition model is used to parse the task request to obtain the request entity, and the data entity and intelligent agent entity associated with the request entity are queried in the core knowledge graph to obtain the data requirement;

[0031] The core knowledge graph is used to query the data source network associated with the data requirement, and an improved optimization algorithm is used to find the optimal combination of data collection paths in the data source network.

[0032] Based on the optimal combination of data acquisition paths, cloud servers are used to collect data from different data sources to obtain a number of raw data.

[0033] In one alternative implementation, the core knowledge graph is queried to find the data source network associated with the data requirement, and an improved optimization algorithm is used to find the optimal combination of data collection paths in the data source network, including:

[0034] The core knowledge graph is used to query the data source network associated with the data requirement; the data source network includes several data sources corresponding to several data items to be collected in the data requirement.

[0035] The data acquisition paths from several data sources are combined and encoded into individual vectors for the ISBO algorithm, and the ISBO population parameters and maximum number of iterations are set.

[0036] Based on the ISBO population parameters, the initial ISBO population is obtained by initializing using the Tent chaotic mapping sequence; each ISBO individual in the ISBO population corresponds to a candidate data acquisition path combination.

[0037] The fitness function is used to obtain the fitness value of each initial ISBO individual, and the initial ISBO individual with the best fitness value is taken as the optimal solution.

[0038] By introducing the Levy flight strategy and convergence factor, the initial ISBO population is explored by building a pavilion or decorating a pavilion, and an updated first ISBO population is obtained.

[0039] Using the theft probability, several first ISBO individuals are randomly selected from the updated first ISBO population to simulate theft behavior, resulting in several updated second ISBO individuals.

[0040] A dynamic reversal mechanism is introduced, in which several first ISBO individuals are randomly selected from the updated first ISBO population with reversal probability to perform orientation, thereby obtaining several updated third ISBO individuals.

[0041] Using the fitness function, obtain the fitness values ​​of the first, second, and third ISBO individuals for each update, and update the ISBO individual with the best fitness value as the optimal solution;

[0042] When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, the iterative update of the ISBO population is terminated, and the optimal solution of the current iteration is output.

[0043] Decode the individual vector of the ISBO individual corresponding to the optimal solution to obtain the optimal combination of data acquisition paths in the data source network.

[0044] In one alternative implementation, during the data preprocessing stage, a core knowledge graph is used to intelligently manage several raw data sets, resulting in several preprocessed data sets, including:

[0045] In the data preprocessing stage, an entity recognition model is used to extract entities from each piece of raw data to obtain several material entities. These material entities are then associated with the corresponding data entities in the core knowledge graph to obtain several associated data entities.

[0046] Based on the data meta-model and cleaning rules defined in the core knowledge graph, several original data sets with associated data entities are cleaned, deduplicated, and formatted to obtain several standard data sets with associated data entities.

[0047] Based on the domain knowledge defined in the core knowledge graph, semantic annotation and entity linking are performed on all related data entities of several standard data to obtain several preprocessed data with annotation information.

[0048] In one alternative implementation, during the data storage phase, a core knowledge graph is used to intelligently organize several preprocessed data sets to obtain an updated core knowledge graph, including:

[0049] During the data storage phase, the associated data entities and annotation information of each preprocessed data are used as new data entities and updated to the core knowledge graph.

[0050] Based on the graph structure of the core knowledge graph, semantic relationships are established between the preprocessed data, and each preprocessed data is vectorized to obtain a semantic index.

[0051] The semantic index is added to the corresponding new data entity in the core knowledge graph to obtain the updated core knowledge graph.

[0052] In one alternative implementation, during the data application phase, based on the agent's task information, an updated core knowledge graph is used to provide intelligent services, offering task support to the agent and recording the agent's data usage behavior, resulting in a second-updated core knowledge graph, including:

[0053] During the data application phase, a data management platform is used to collect task information of intelligent agents and upload the task information to the cloud server through an encrypted and secure channel.

[0054] In the cloud server, an entity recognition model is used to extract entities from the task information to obtain several task entities. Then, the data entities and knowledge paths associated with the task entities are queried in the core knowledge graph to obtain a knowledge subgraph.

[0055] Based on the knowledge subgraph and historical feedback information, a pre-built intelligent service decision model is used to generate intelligent service decisions. The knowledge subgraph and intelligent service decisions are then sent to the client's intelligent agent through the data management platform to provide task support for the intelligent agent.

[0056] The data management platform records the data usage behavior of intelligent agents on the knowledge subgraph based on intelligent service decisions, and uploads the data usage behavior to the cloud server through an encrypted and secure channel;

[0057] In the cloud server, data usage behavior is used as a new entity relationship for all data entities in the knowledge subgraph, and the new entity relationship is added to the corresponding data entity in the core knowledge graph to obtain a second-updated core knowledge graph.

[0058] In one alternative implementation, during the data archiving phase, the core knowledge graph updated twice undergoes intelligent processing to eliminate low-value data entities, resulting in a core knowledge graph updated three times, including:

[0059] During the data archiving phase, the value evaluation criteria for all data entities in the core knowledge graph that is updated twice are collected.

[0060] Based on the value assessment criteria, a pre-built value prediction model is used to predict the future value of each data entity.

[0061] Based on the preset archiving strategy and compliance requirements, data entities whose future value is below the value threshold are archived and destroyed, and the status of the corresponding data entities in the second-updated core knowledge graph is updated to obtain the third-updated core knowledge graph.

[0062] Secondly, embodiments of the present invention provide a knowledge graph-guided intelligent agent data lifecycle management device for implementing an intelligent agent data lifecycle management method. The device includes:

[0063] The knowledge graph construction unit is used to connect the data management platform to the client's intelligent agent and construct the corresponding core knowledge graph in the cloud server based on the agent's metadata.

[0064] The data acquisition unit is used to collect data from the data source by using the core knowledge graph for intelligent guidance based on the task request of the intelligent agent during the data acquisition phase, and obtain a number of raw data.

[0065] The data preprocessing unit is used to intelligently manage several raw data using a core knowledge graph during the data preprocessing stage, resulting in several preprocessed data.

[0066] The data storage unit is used to intelligently organize several preprocessed data using the core knowledge graph during the data storage stage to obtain an updated core knowledge graph.

[0067] The data application unit is used to provide intelligent services to the intelligent agent based on the agent's task information during the data application phase, using the updated core knowledge graph, to provide task support to the intelligent agent, and to record the agent's data usage behavior to obtain a second-updated core knowledge graph.

[0068] The data archiving unit is used to intelligently process the core knowledge graph that has been updated twice during the data archiving stage, eliminating low-value data entities and obtaining the core knowledge graph that has been updated three times.

[0069] A third aspect of this invention provides an electronic device, which includes:

[0070] At least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0071] The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0072] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention;

[0074] Figure 2 This is a flowchart illustrating the steps of a knowledge graph-guided intelligent agent data lifecycle management method provided in an embodiment of the present invention.

[0075] Figure 3 This is a schematic diagram of the functional units of a knowledge graph-guided intelligent agent data lifecycle management device provided in an embodiment of the present invention. Detailed Implementation

[0076] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0077] The present invention will be further described below with reference to the accompanying drawings.

[0078] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0079] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0080] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0081] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for a knowledge graph-guided intelligent agent data lifecycle management device.

[0082] exist Figure 1 In the illustrated electronic device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the electronic program of the knowledge graph-guided intelligent agent data life cycle management device stored in the memory 1005 through the processor 1001, and executes the knowledge graph-guided intelligent agent data life cycle management method provided in the embodiment of the present invention.

[0083] Reference Figure 2 The embodiments of the present invention provide a knowledge graph-guided method for managing the entire lifecycle of intelligent agent data, the method comprising:

[0084] S201: Connect the data management platform to the client's intelligent agent and construct the corresponding core knowledge graph in the cloud server based on the agent's metadata;

[0085] S202: During the data acquisition phase, based on the agent's task request, the core knowledge graph is used for intelligent guidance to acquire data from the data source and obtain some raw data.

[0086] S203: In the data preprocessing stage, core knowledge graphs are used to intelligently manage several raw data to obtain several preprocessed data.

[0087] S204: In the data storage stage, the core knowledge graph is used to intelligently organize several preprocessed data to obtain an updated core knowledge graph.

[0088] S205: In the data application phase, based on the agent's task information, the updated core knowledge graph is used to provide intelligent services, provide task support to the agent, and record the agent's data usage behavior to obtain a second-updated core knowledge graph.

[0089] S206: During the data archiving stage, the core knowledge graph updated twice is intelligently processed to eliminate low-value data entities and obtain the core knowledge graph updated three times.

[0090] The technical solution provided in this application has at least the following beneficial effects:

[0091] For the first time, knowledge graphs are used as the core engine, running through the entire lifecycle of data collection, governance, storage, application, and archiving. This forms a data-driven, knowledge-guided, and self-optimizing closed-loop management system, breaking down the stage barriers of traditional intelligent agent data management. In the data collection stage, knowledge graphs are used to understand task requirements, and combined with improved optimization algorithms to plan the optimal collection path, achieving a shift from "blind collection" to "precise collection on demand," significantly improving data collection efficiency and quality. In the data governance and storage stages, the semantic capabilities of knowledge graphs are used for data cleaning, labeling, and association, establishing effective semantic relationships between data and integrating heterogeneous data. Transforming data into structured and semantic knowledge entities greatly enhances data usability and knowledge discovery capabilities, reduces data governance costs and error rates, and improves data governance efficiency. In the data application stage, it not only provides precise knowledge services to intelligent agents but also feeds back the agents' usage behavior to the knowledge graph, enabling the knowledge graph to dynamically evolve according to actual application scenarios and achieve continuous value-added of intelligent agent data. In the data archiving stage, based on the multi-dimensional information accumulated in the knowledge graph, it performs value prediction and intelligent processing of data entities, realizing lean management of intelligent agent data, saving storage costs, and ensuring the "health" of intelligent agent data.

[0092] In one alternative implementation, the data management platform is connected to the client's intelligent agent, and based on the agent's application scenario information, a corresponding core knowledge graph is constructed on a cloud server, including:

[0093] S2011: The intelligent agent that logs into the data management platform on the client and connects the data management platform to the client's local storage;

[0094] For example, users or administrators log in to the data management platform on a client (such as a mobile terminal or server) and establish a connection between the platform and a locally running intelligent agent (such as an intelligent recommendation system in the e-commerce field) through application programming interfaces (APIs), software development kits (SDKs), etc.

[0095] S2012: Use a data management platform to collect metadata of intelligent agents and upload the metadata to a cloud server through an encrypted secure channel;

[0096] In this embodiment, the metadata includes, but is not limited to: the application scenario of the intelligent agent (such as e-commerce recommendation), functional description, data interface specifications, initial data source information, domain terminology, etc. The collected metadata is uploaded to the cloud server through encrypted secure channels such as SSL / TLS to ensure transmission security.

[0097] S2013: Based on the agent's metadata, construct the corresponding core knowledge graph in the cloud server;

[0098] In this embodiment, the core function of the core knowledge graph is to transform fragmented and heterogeneous descriptive information about intelligent agents into a structured, semantic, and machine-understandable "digital blueprint" or "cognitive center".

[0099] Core knowledge graphs lay the foundation for intelligent management, enabling a leap from "data" to "knowledge":

[0100] Effect description: Traditional metadata is isolated and flat information. After building a core knowledge graph, this information is organized into an interconnected network. This makes it no longer a simple "storage" of data, but "understanding" data. The core knowledge graph knows what data an agent needs, where this data comes from, and how they affect each other. This is the fundamental premise for all subsequent "intelligent" guidance (such as intelligent collection and intelligent governance).

[0101] For example, instead of seeing an "API address" string, one can understand that "this API is used to obtain 'user profile' data, and the 'user profile' data is necessary for the 'e-commerce recommendation' agent to complete the 'personalized recommendation' task";

[0102] Core knowledge graphs enable accurate understanding and forward-looking planning of data needs:

[0103] Effect description: Knowledge graphs directly link the agent's "task objectives" with "data requirements". When an agent proposes a new task, it can accurately understand the underlying data requirements through graph reasoning, and even predict the data that may be needed in the future, thereby turning passive response into proactive planning.

[0104] For example, when there is a path in the graph such as "(Agent A) - Execution → (Task X) - Need → (Data Y)", once Agent A is ready to execute Task X, it can immediately and accurately locate Data Y without performing complex queries or manual configuration.

[0105] Core knowledge graphs break down data silos and enable semantic interconnection across data sources:

[0106] Effect description: Metadata typically describes data from different sources (such as databases, APIs, and files). Knowledge graphs connect entities and relationships from these heterogeneous data sources through a unified semantic model to form a global view, which provides a "navigation map" for subsequent cross-source data collection, governance, and integration.

[0107] For example, the graph can associate the "user ID" in the MySQL database with the "user behavior log" in MongoDB and the "user tag" returned by the third-party API under the same "user" entity, making it possible to build a complete 360-degree view of the user;

[0108] Core knowledge graphs enable automation and collaboration at every stage of the entire lifecycle:

[0109] Effect description: The core knowledge graph acts as a "central brain," providing a unified context and decision-making basis for all subsequent stages, such as data collection, governance, storage, application, and archiving. This makes each stage no longer an isolated operation, but a closely coordinated and mutually reinforcing automated process.

[0110] For example, the collection phase plans the optimal path based on the knowledge graph, the governance phase uses the graph's meta-model and cleaning rules for automated processing, the application phase provides accurate knowledge services based on the knowledge graph, and the archiving phase uses the accumulated usage frequency and value assessment in the knowledge graph for intelligent processing; all these automated processes rely on the high-quality knowledge graph initially constructed.

[0111] The core knowledge graph enhances scalability and adaptability:

[0112] Effect description: When a new intelligent agent is connected or a new data source is added, the system only needs to extract its metadata and integrate it into the existing knowledge graph. Due to the good scalability of the graph, the entire management system can smoothly adapt to changes without large-scale reconstruction of the core architecture. The dynamic evolution mechanism further ensures that the system can "keep pace with the times" and continuously learn and grow.

[0113] In one alternative implementation, a corresponding core knowledge graph is constructed on a cloud server based on the agent's metadata, including:

[0114] S20131: Define a multi-dimensional ontology layer for the core knowledge graph in a cloud server;

[0115] First, a multi-dimensional ontology layer is defined, which defines core concepts such as agents, data, tasks, and data sources, along with their attributes and relationships. For example, the "agent" entity is defined to have an "application scenario" attribute, the "data" entity to have "source" and "format" attributes, and a "requirement" relationship is defined between "agents" and "data." Then, structured (e.g., configuration files), semi-structured (e.g., JSON / XML logs), and unstructured (e.g., functional description documents) information is extracted from metadata and populated into the ontology layer to form an initial knowledge graph. Finally, a dynamic evolution mechanism is established for this graph, such as setting rules to automatically update the graph structure when new data is added or new entity relationships are discovered, resulting in the final core knowledge graph.

[0116] In this embodiment, the multi-dimensional ontology layer is the skeleton of the core knowledge graph, defining the allowed entity types, relation types, their attributes, and constraints in the graph. To support the data management of the intelligent agent, the ontology layer must be multi-dimensional, covering at least the following four dimensions:

[0117] 1) Agent dimension:

[0118] Entity types: intelligent agents, capability modules (such as perception modules, decision-making modules, and execution modules), and algorithm models (such as YOLOv5, Bidirectional Encoder Representations from Transformers (BERT) models, and Proximal Policy Optimization (PPO) models).

[0119] Relationship types: Agent - owns → Capability module, Capability module - runs → Algorithm model, Agent - executes → Task;

[0120] Attributes: Agent ID, version, state (running, idle, faulty); Algorithm model accuracy, latency, and resource consumption;

[0121] 2) Data Dimensions:

[0122] Entity types: datasets, data items (single data entries), data sources (such as sensors, APIs, databases), and data schemas;

[0123] Relationship types: Data item - belongs to → dataset, dataset - originates from → data source, data item - follows → data schema, data item - passes through → processing flow;

[0124] Attributes: data format, size, creation time, quality score, storage location, and access permissions;

[0125] 3) Task and process dimension:

[0126] Entity types: Task, Subtask, Processing flow, Decision node.

[0127] Relationship types: Task - Decomposed into → Subtask, Subtask - Requires → Dataset, Processing flow - Contains → Decision node, Decision node - Input → Data item, Decision node - Output → Data item;

[0128] Attributes: Task objective, priority, and deadline; average process time and success rate;

[0129] 4) Domain knowledge dimension:

[0130] Entity type: Defined according to the application domain of intelligent agents, such as vehicles, traffic lights, and road signs in the field of autonomous driving; and products, brands, and users in the field of e-commerce recommendations.

[0131] Relationship type: describes the objective relationship between entities in the domain, such as vehicle - located on → road, product - belongs to → brand, user - purchase → product;

[0132] Attributes: Inherent attributes of domain entities, such as the color and speed of a vehicle; the price and inventory of a product;

[0133] Implementation: The ontology layer can be formally defined using Web Ontology Language or Schema languages ​​such as Protobuf, and then loaded and parsed by the knowledge graph construction module;

[0134] S20132: Extract structured, semi-structured, and unstructured data from the agent's metadata and populate it into the multi-dimensional ontology layer to obtain the initial core knowledge graph;

[0135] In this embodiment, in the early stage of map establishment, information needs to be extracted from the agent's metadata or external sources to fill the map;

[0136] Structured data extraction: For agent configuration files (such as JSON, YAML) and database tables, directly use ETL tools or custom scripts to map them to entities and relationships in the graph. For example, map model_name: "BERT-Large" in the configuration file to an attribute of the algorithm model entity.

[0137] Semi-structured / unstructured data extraction:

[0138] Code and Documentation Analysis: Using static code analysis tools and natural language processing technology, extract entities such as capability modules and processing flows, as well as their relationships, from the source code, technical documentation, and API descriptions of the intelligent agent;

[0139] Log data mining: By using log parsing templates and regular expressions, tasks, data items, performance metrics, etc. are extracted from the agent's operation logs, and the temporal relationships between them are established;

[0140] Domain knowledge injection: Utilize existing domain knowledge bases (such as Wikidata, DBpedia) or professional dictionaries, and fill the domain dimension entities in the ontology layer with rich background knowledge through entity linking technology;

[0141] Implementation: This stage can integrate multiple extractors, such as rule-based extractors or pre-trained entity recognition models, to adapt to different types of data sources;

[0142] Entities extracted from different sources may point to the same object in the real world, but their descriptions differ, requiring entity alignment:

[0143] Semantic similarity-based alignment: Calculate the vector similarity of entity names, attributes, and context descriptions (e.g., using cosine similarity), and merge entities with similarity scores higher than a threshold;

[0144] Rule-based alignment: using unique identifiers (such as product ID, user ID) for precise matching;

[0145] Relationship disambiguation: Handling ambiguity in relation types. For example, in "Apple releases a new phone" and "I like to eat apples", the context is used to determine whether "apple" refers to a company entity or a fruit entity, respectively.

[0146] Implementation method: An entity alignment model based on the Graph Neural Network (GNN) algorithm can be used. This model can simultaneously utilize the attribute information of nodes and the structural information of the graph to achieve higher alignment accuracy.

[0147] S20133: Construct a dynamic evolution mechanism for the initial core knowledge graph to obtain the final core knowledge graph;

[0148] In this embodiment, the dynamic evolution mechanism defines how the map responds to changes and feedback, including the following three core loops:

[0149] 1) Real-time update loop:

[0150] Trigger: When an agent performs a new task, processes a new data item, or its internal state changes, a "graph update event" is automatically generated.

[0151] Processing: The core knowledge graph manager listens to these events, parses out new entities, relationships, or attribute changes in real time, and updates them to the graph. For example, when an agent completes an image recognition task, it creates a data item entity in the graph and establishes the relationships of task-generation → data item and data item-containment → recognition result.

[0152] Technology: This can be achieved by using a stream processing framework (such as Apache Flink) combined with the real-time write API of a graph database;

[0153] 2) Feedback-driven optimization loop:

[0154] Trigger: The decision results of the agent in the data application stage (such as task success / failure, user satisfaction rating) serve as feedback signals;

[0155] Processing: This feedback signal was used to correct and optimize the knowledge in the map;

[0156] Correction: If a decision fails, trace back to the data or knowledge path it depends on and reduce the confidence weight of related relationships or entities on that path;

[0157] Strengthen knowledge: If a decision is successful, increase the weight of the relevant path;

[0158] Update model performance: Update the attributes of the corresponding algorithm model entities with the performance of the algorithm model in real tasks (such as accuracy and latency);

[0159] Technology: This loop can be closely integrated with reinforcement learning modules, where the reward signals from reinforcement learning directly drive the adjustment of weights in the graph;

[0160] Regularly explore and discover cycles:

[0161] Trigger: Activates according to a set time period (e.g., daily, weekly);

[0162] Processing: Perform offline analysis and mining on the entire map to discover new and implicit knowledge;

[0163] Link prediction: Using graph embeddings or GNN models, predict possible but undiscovered relationships in a graph and confirm them manually or automatically.

[0164] Community discovery: Discovering tightly connected groups of entities in the graph, which may represent a new concept or a potential task pattern;

[0165] Pattern discovery: Analyze frequently occurring subgraph structures, extract new and more efficient processing flow templates, and add them as new processing flow entities to the graph;

[0166] Technology: This loop is implemented using offline computing frameworks (such as Apache Spark) and graph computing algorithm libraries (such as GraphX);

[0167] The core knowledge graph constructed in this step is no longer a static knowledge base, but a "cognitive center" with perception, learning, memory and reasoning capabilities, which can provide continuous and dynamically optimized intelligent guidance for the full lifecycle management of data by intelligent agents.

[0168] In one alternative implementation, during the data acquisition phase, based on the agent's task request, a core knowledge graph is used for intelligent guidance to acquire data from the data source, obtaining several raw data, including:

[0169] S2021: During the data acquisition phase, the data management platform is used to collect the task requests of the intelligent agent and upload the task requests to the cloud server through an encrypted secure channel;

[0170] For example, when an intelligent agent needs to perform a task (such as "recommend summer menswear for user A"), it generates a task request, which is captured by the data management platform and uploaded to the cloud.

[0171] S2022: In the cloud server, a pre-trained entity recognition model is used to parse the task request to obtain the request entity, and the data entity and intelligent entity associated with the request entity are queried in the core knowledge graph to obtain the data requirement;

[0172] In this embodiment, the cloud server uses a pre-trained entity recognition model (such as the BERT model) to parse the request and extract request entities such as "User A", "Summer", and "Men's Clothing".

[0173] By querying these request entities in the core knowledge graph, the graph may record "User A's" browsing history, the best-selling categories corresponding to "summer", and the product attributes related to "men's clothing". Through graph reasoning, the data requirements for this task can be clarified, such as: needing to collect User A's browsing data for the last 30 days, men's clothing sales data for the last week, and summer fashion element data, etc.

[0174] S2023: Query the data source network associated with the data requirement in the core knowledge graph, and use the improved optimization algorithm to find the optimal combination of data collection paths in the data source network;

[0175] In this embodiment, the data source network includes multiple data sources such as internal databases, third-party APIs, and public datasets; the cloud server collects data from each data source in parallel or serially according to the data collection path combination.

[0176] S2024: Based on the optimal combination of data acquisition paths, use cloud servers to collect data from different data sources to obtain a number of raw data.

[0177] In one alternative implementation, the core knowledge graph is queried to find the data source network associated with the data requirement, and an improved optimization algorithm is used to find the optimal combination of data collection paths in the data source network, including:

[0178] S20231: Query the data source network associated with the data requirement in the core knowledge graph; the data source network includes several data sources corresponding to several data items to be collected in the data requirement;

[0179] S20232: Encode the data collection path combinations of several data sources (such as API address, query statement, authentication method, etc.) into individual vectors of the Improved Satin Bowerbird Optimizer (ISBO) algorithm, and set the ISBO population parameters and maximum number of iterations;

[0180] S20233: Based on the ISBO population parameters, the initial ISBO population is obtained by initializing using the Tent chaotic mapping sequence; each ISBO individual in the ISBO population corresponds to a candidate data acquisition path combination.

[0181] The formula is:

[0182]

[0183] In the formula, The i-th initial ISBO individual in the initial ISBO population; Let i be the i-th chaotic variable; represents the upper and lower bounds of the search space; i represents the ISBO individual indicator.

[0184]

[0185] In the formula, Let be the (i-1)th chaotic variable; compared with random initialization, chaotic initialization can ensure that the population is evenly distributed in the solution space, thus enhancing diversity;

[0186] S20234: Use the fitness function to obtain the fitness value of each initial ISBO individual, and take the initial ISBO individual with the best fitness value as the optimal solution;

[0187] Define a fitness function that comprehensively considers multiple indicators such as data collection cost (e.g., API call fees), time latency, data error rate, and path duplication rate. The formula is as follows:

[0188]

[0189] In the formula, The fitness value of individual X in ISBO; The data collection cost for the alternative data collection path combinations corresponding to ISBO individual X; The time delay for the alternative data collection path combinations corresponding to ISBO individual X; The data error rate for the alternative data collection path combinations corresponding to ISBO individual X; The path duplication rate of the alternative data collection path combinations corresponding to ISBO individual X; The fitness coefficient;

[0190] S20235: Introduce the Levy flight strategy and convergence factor to explore the initial ISBO population by building a pavilion or decorating a pavilion, and obtain an updated first ISBO population.

[0191] The formula is:

[0192]

[0193] In the formula, To explore probability; To explore the minimum and maximum probabilities; t represents the maximum number of iterations; t represents the current number of iterations.

[0194] To explore random numbers (Select behavior pattern within [0, 1]):

[0195] if > To perform exploratory behavior, the formula is:

[0196]

[0197] In the formula, The first ISBO individual updated for the i-th iteration of the pavilion construction exploration at iteration number t+1; Step size factor; is a Levy distribution random number; b is the Levy step size, and b∈[1,2]; This refers to the i-th ISBO individual in iteration t, which is the initial ISBO individual in the initial iteration; The optimal solution for iteration number t; It is a random integer with a value of 1 or 2; It uses element-wise multiplication; it enables ISBO individuals to perform Lévy flight searches around the current optimal solution, which has both randomness and directionality, greatly enhancing the global exploration capability;

[0198] if To execute the attack, the formula is:

[0199]

[0200] In the formula, The first ISBO individual updated for the i-th iteration of the decorative gazebo at iteration number t+1; The convergence factor; A random number between [0, 1];

[0201]

[0202] In the formula, These are the maximum and minimum values ​​of the convergence factor; , To adjust the parameters; It is the hyperbolic tangent function; in the early stages of the algorithm (when t is small), The attack step size is relatively large, which is beneficial for exploring a larger area around the optimal solution, especially in the later stages of the algorithm (when t is relatively large). Smaller attack step size allows for more refined localized development;

[0203] S20236: Randomly select several first ISBO individuals from the updated first ISBO population with theft probability to simulate theft behavior, and obtain several updated second ISBO individuals;

[0204] Generate a random number between [0, 1]. ,if Theft probability If so, then the theft simulation will be executed;

[0205] The formula is:

[0206]

[0207] In the formula, The second ISBO individual updated for the theft behavior at iteration number t+1; is a random number that is uniformly distributed in the interval [−1,1]. For the first ISBO individual updated at the i, k, jth random selection of iteration number t+1; k, j are ISBO individual indicators;

[0208] S20237: Introduce a dynamic reversal mechanism, randomly select several first ISBO individuals from the updated first ISBO population with reversal probability to perform orientation, and obtain several updated third ISBO individuals.

[0209] Generate a random number between [0, 1]. ,if Reverse probability If so, then reverse learning will be performed;

[0210] The formula is:

[0211]

[0212] In the formula, The third ISBO individual is the i-th updated one obtained by dynamic reverse processing at iteration number t+1. The reverse solution for the first ISBO individual, which is randomly selected for the i-th update at iteration number t+1; The fitness function;

[0213]

[0214] In the formula, The reverse center point, determined by the dynamic boundary of the current search space, is the point of iteration t+1. It is a dynamic inverse factor that increases with iteration (inverse exploration is enhanced in later stages).

[0215] S20238: Use the fitness function to obtain the fitness values ​​of the first, second, and third ISBO individuals for each update, and update the updated ISBO individual with the best fitness value as the optimal solution;

[0216] S20239: When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, terminate the iterative update of the ISBO population and output the optimal solution of the current iteration;

[0217] S202310: Decode the individual vector of the ISBO individual corresponding to the optimal solution to obtain the optimal combination of data acquisition paths in the data source network.

[0218] In one alternative implementation, during the data preprocessing stage, a core knowledge graph is used to intelligently manage several raw data sets, resulting in several preprocessed data sets, including:

[0219] S2031: In the data preprocessing stage, an entity recognition model is used to extract entities from each piece of raw data to obtain several material entities, and these material entities are associated with the corresponding data entities in the core knowledge graph to obtain several associated data entities.

[0220] In this embodiment, an entity recognition model is used to extract entities from each piece of original data (such as a user browsing record) to obtain material entities such as "user ID", "product ID", and "browsing time". These material entities are then associated with the data entities defined in the core knowledge graph. For example, the extracted "product ID" is linked to the corresponding "product" entity in the graph.

[0221] S2032: Based on the data meta-model (such as data type and format specifications) and cleaning rules (such as missing value filling and outlier removal) defined in the core knowledge graph, clean, deduplicate, and format several original data sets with associated data entities to obtain several standard data sets with associated data entities.

[0222] S2033: Based on the domain knowledge defined in the core knowledge graph (e.g., "T-shirt" belongs to "top", "top" belongs to "clothing"), perform semantic annotation and entity linking on all related data entities of several standard data to obtain several preprocessed data with annotation information.

[0223] In one alternative implementation, during the data storage phase, a core knowledge graph is used to intelligently organize several preprocessed data sets to obtain an updated core knowledge graph, including:

[0224] S2041: During the data storage stage, the associated data entities and annotation information of each preprocessed data are used as new data entities and updated to the core knowledge graph.

[0225] S2042: Based on the graph structure of the core knowledge graph, establish semantic relationships between preprocessed data. For example, newly collected user browsing data can be associated with the user's historical purchase data and profile data. Each piece of preprocessed data is vectorized (e.g., using Doc2Vec or Sentence-BERT models) to generate low-dimensional dense semantic vectors as semantic indexes.

[0226] S2043: Add semantic indexes to the corresponding new data entities in the core knowledge graph to obtain an updated core knowledge graph with richer content and a more complete structure.

[0227] In one alternative implementation, during the data application phase, based on the agent's task information, an updated core knowledge graph is used to provide intelligent services, offering task support to the agent and recording the agent's data usage behavior, resulting in a second-updated core knowledge graph, including:

[0228] S2051: In the data application phase, the data management platform is used to collect the task information of the intelligent agent (such as "generating a recommendation list for user A"), and the task information is uploaded to the cloud server through an encrypted secure channel;

[0229] S2052: In the cloud server, an entity recognition model is used to extract entities from the task information to obtain several task entities. Then, the data entities and knowledge paths associated with the task entities (such as "User A" → "Preference" → "Brand X", "Summer" → "Fashion" → "Style Y") are queried in the core knowledge graph to obtain highly related knowledge subgraphs.

[0230] S2053: Based on the knowledge subgraph and historical feedback information, a pre-built intelligent service decision model is used to generate intelligent service decisions. The knowledge subgraph and intelligent service decisions are then sent to the client's intelligent agent through the data management platform to provide task support for the intelligent agent.

[0231] In this embodiment, the intelligent service decision model is constructed based on the PPO algorithm, and its state space includes a composite vector composed of a knowledge subgraph and historical feedback information.

[0232] Knowledge Subgraph Vectorization: Using graph neural networks, such as Graph Sampling and Aggregation (GraphSAGE) or Graph Attention Network (GAT), the knowledge subgraph generated for the current task is encoded. GNN can capture the attributes of nodes (such as users, products, and brands) in the graph and the complex relationship structure between them, and finally generate a fixed-dimensional graph embedding vector.

[0233] Vectorization of historical feedback information: Organize historical feedback information related to the current user (such as click-through rate, purchase conversion rate, dwell time, etc. in the past week) into a feature vector;

[0234] Final state vector: The two vectors are concatenated into a complete state vector that represents the current decision-making situation;

[0235] The action space can be enormous (all possible combinations of goods), which is difficult to handle; therefore, it is simplified to a multi-label classification or sequence generation problem. Implementation method:

[0236] The output layer of the PPO policy network can be designed with neurons of the same size as the candidate product pool and use the Sigmoid activation function. The output value of each neuron (between 0 and 1) represents the probability of recommending the corresponding product. During decision-making, the products with the highest probabilities are selected to form the final recommendation list.

[0237] Reward: The immediate feedback signal obtained from the environment after an agent performs an action is used to guide learning, and the design of the reward function is crucial;

[0238] Instant reward: After the agent sends the recommendation list to the user, observe the user's short-term behavior;

[0239] The user clicked on any item in the recommended list;

[0240] The user purchased any item from the recommended list;

[0241] The user did not interact at all; a slight penalty was imposed to encourage exploration.

[0242] Long-term rewards: can be combined with longer-term metrics such as user retention rate and lifetime value, but immediate rewards are easier for PPOs to learn online;

[0243] Policy: The decision model of an agent, that is, a mapping from state to action. In this embodiment, the policy is the PPO model being trained, which receives the state and outputs a probability distribution about the action.

[0244] S2054: Use a data management platform to record the data usage behavior of intelligent agents on the knowledge subgraph based on intelligent service decisions. For example, which items in the recommendation list did user A click on, and which did user A ignore? The data usage behavior is then uploaded to the cloud server through an encrypted secure channel.

[0245] S2055: In the cloud server, data usage behavior (such as "user A clicked on product Z") is used as a new entity relationship for all data entities in the knowledge subgraph, and the new entity relationship is added to the corresponding data entity in the core knowledge graph to obtain a second-updated core knowledge graph.

[0246] In this embodiment, a "click" relationship is added between the "User A" entity and the "Product Z" entity, and a timestamp is recorded. This enables the knowledge graph to reflect the latest and real application effects and achieve secondary updates.

[0247] In one alternative implementation, during the data archiving phase, the core knowledge graph updated twice undergoes intelligent processing to eliminate low-value data entities, resulting in a core knowledge graph updated three times, including:

[0248] S2061: During the data archiving phase, collect the value evaluation criteria for all data entities in the core knowledge graph that are updated twice, either periodically or triggered. These criteria are multi-dimensional and include: the creation time of the data entity, the last access time, the frequency of citation, the weight in the decision model, the data quality score, etc.

[0249] S2062: Based on the value evaluation criteria, use a pre-built value prediction model (such as Extreme Gradient Boosting (XGBoost) model, Long Short-Term Memory (LSTM) network model) to predict the future value of each data entity, that is, the potential value in the future.

[0250] S2063: Based on the preset archiving strategy (such as "not accessed for more than one year and with a predicted value of less than 0.1") and compliance requirements (such as certain data must be retained for 5 years), archive and destroy data entities whose future value is below the value threshold (move to low-cost storage or destroy them), and update the status of these disposed data entities in the second-updated core knowledge graph (such as marking them as "archived" or "destroyed") to obtain a more concise and higher-quality third-updated core knowledge graph, preparing for the next round of data lifecycle management.

[0251] This invention also provides a knowledge graph-guided intelligent agent data lifecycle management device, referring to... Figure 3 The diagram illustrates the functional units of a knowledge graph-guided intelligent agent data lifecycle management device 300 according to the present invention. This device may include the following units:

[0252] The knowledge graph construction unit 301 is used to connect the data management platform to the client's intelligent agent and construct the corresponding core knowledge graph in the cloud server based on the agent's metadata.

[0253] The data acquisition unit 302 is used to collect data from the data source by using the core knowledge graph for intelligent guidance according to the task request of the intelligent agent during the data acquisition stage, and obtain a number of raw data.

[0254] The data preprocessing unit 303 is used to intelligently manage several raw data using a core knowledge graph during the data preprocessing stage to obtain several preprocessed data.

[0255] Data storage unit 304 is used to intelligently organize several preprocessed data using the core knowledge graph during the data storage stage to obtain an updated core knowledge graph.

[0256] The data application unit 305 is used to provide intelligent services to the intelligent agent by using the updated core knowledge graph based on the agent's task information during the data application phase, to provide task support to the intelligent agent, and to record the agent's data usage behavior to obtain a second-updated core knowledge graph.

[0257] Data archiving unit 306 is used to intelligently process the core knowledge graph that has been updated twice during the data archiving stage, eliminate low-value data entities, and obtain the core knowledge graph that has been updated three times.

[0258] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0259] Memory, used to store computer programs;

[0260] The processor, when executing a program stored in memory, implements the knowledge graph-guided intelligent agent data lifecycle management method of the present invention.

[0261] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0262] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0263] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the knowledge graph-guided intelligent agent data lifecycle management method of the embodiments of the present invention.

[0264] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0265] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0266] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0267] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0268] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0269] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A knowledge graph-guided method for managing the entire lifecycle of intelligent agent data, characterized in that, The method includes: Connect the data management platform to the client's intelligent agent, and build the corresponding core knowledge graph in the cloud server based on the agent's metadata; During the data acquisition phase, based on the agent's task request, the core knowledge graph is used for intelligent guidance to collect data from the data source and obtain a number of raw data. In the data preprocessing stage, core knowledge graphs are used to intelligently manage several raw data sets to obtain several preprocessed data sets. During the data storage phase, the core knowledge graph is used to intelligently organize several preprocessed data to obtain an updated core knowledge graph. In the data application phase, based on the agent's task information, the updated core knowledge graph is used to provide intelligent services, support the agent's tasks, and record the agent's data usage behavior to obtain a second-updated core knowledge graph. During the data archiving stage, the core knowledge graph updated twice is intelligently processed to eliminate low-value data entities, resulting in a core knowledge graph updated three times.

2. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 1, characterized in that, Connect the data management platform to the client's intelligent agent, and construct the corresponding core knowledge graph in the cloud server based on the agent's application scenario information, including: Log in to the data management platform on the client side and connect the data management platform to the intelligent agent stored on the client's local storage; The data management platform is used to collect metadata of the intelligent agent and upload the metadata to the cloud server through an encrypted and secure channel; Based on the agent's metadata, a corresponding core knowledge graph is constructed on a cloud server.

3. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 2, characterized in that, Based on the agent's metadata, a corresponding core knowledge graph is constructed in the cloud server, including: Define a multi-dimensional ontology layer for the core knowledge graph on a cloud server; Structured, semi-structured, and unstructured data are extracted from the metadata of the intelligent agent and populated into the multi-dimensional ontology layer to obtain the initial core knowledge graph. A dynamic evolution mechanism is constructed for the initial core knowledge graph to obtain the final core knowledge graph.

4. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 3, characterized in that, During the data acquisition phase, based on the agent's task request, the core knowledge graph is used for intelligent guidance, and data is collected from the data source to obtain several raw data, including: During the data acquisition phase, a data management platform is used to collect task requests from intelligent agents and upload these requests to a cloud server via an encrypted and secure channel. In the cloud server, a pre-trained entity recognition model is used to parse the task request to obtain the request entity, and the data entity and intelligent agent entity associated with the request entity are queried in the core knowledge graph to obtain the data requirement; The core knowledge graph is used to query the data source network associated with the data requirement, and an improved optimization algorithm is used to find the optimal combination of data collection paths in the data source network. Based on the optimal combination of data acquisition paths, cloud servers are used to collect data from different data sources to obtain a number of raw data.

5. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 4, characterized in that, The core knowledge graph is used to query the data source network associated with the data requirement, and an improved optimization algorithm is used to find the optimal combination of data collection paths within the data source network, including: The core knowledge graph is used to query the data source network associated with the data requirement; the data source network includes several data sources corresponding to several data items to be collected in the data requirement. The data acquisition paths from several data sources are combined and encoded into individual vectors for the ISBO algorithm, and the ISBO population parameters and maximum number of iterations are set. Based on the ISBO population parameters, the initial ISBO population is obtained by initializing using the Tent chaotic mapping sequence; each ISBO individual in the ISBO population corresponds to a candidate data acquisition path combination. The fitness function is used to obtain the fitness value of each initial ISBO individual, and the initial ISBO individual with the best fitness value is taken as the optimal solution. By introducing the Levy flight strategy and convergence factor, the initial ISBO population is explored by building a pavilion or decorating a pavilion, and an updated first ISBO population is obtained. Using the theft probability, several first ISBO individuals are randomly selected from the updated first ISBO population to simulate theft behavior, resulting in several updated second ISBO individuals. A dynamic reversal mechanism is introduced, in which several first ISBO individuals are randomly selected from the updated first ISBO population with reversal probability to perform orientation, thereby obtaining several updated third ISBO individuals. Using the fitness function, obtain the fitness values ​​of the first, second, and third ISBO individuals for each update, and update the ISBO individual with the best fitness value as the optimal solution; When the number of iterations reaches the maximum number of iterations or the fitness value of the optimal solution meets the requirements, the iterative update of the ISBO population is terminated, and the optimal solution of the current iteration is output. Decode the individual vector of the ISBO individual corresponding to the optimal solution to obtain the optimal combination of data acquisition paths in the data source network.

6. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 5, characterized in that, In the data preprocessing stage, a core knowledge graph is used to intelligently manage several raw data sets, resulting in several preprocessed data sets, including: In the data preprocessing stage, an entity recognition model is used to extract entities from each piece of raw data to obtain several material entities. These material entities are then associated with the corresponding data entities in the core knowledge graph to obtain several associated data entities. Based on the data meta-model and cleaning rules defined in the core knowledge graph, several original data sets with associated data entities are cleaned, deduplicated, and formatted to obtain several standard data sets with associated data entities. Based on the domain knowledge defined in the core knowledge graph, semantic annotation and entity linking are performed on all related data entities of several standard data to obtain several preprocessed data with annotation information.

7. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 6, characterized in that, During the data storage phase, the core knowledge graph is used to intelligently organize several preprocessed data sets to obtain an updated core knowledge graph, including: During the data storage phase, the associated data entities and annotation information of each preprocessed data are used as new data entities and updated to the core knowledge graph. Based on the graph structure of the core knowledge graph, semantic relationships are established between the preprocessed data, and each preprocessed data is vectorized to obtain a semantic index. The semantic index is added to the corresponding new data entity in the core knowledge graph to obtain the updated core knowledge graph.

8. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 7, characterized in that, In the data application phase, based on the agent's task information, an updated core knowledge graph is used to provide intelligent services, offering task support to the agent and recording the agent's data usage behavior, resulting in a second-updated core knowledge graph, including: During the data application phase, a data management platform is used to collect task information of intelligent agents and upload the task information to the cloud server through an encrypted and secure channel. In the cloud server, an entity recognition model is used to extract entities from the task information to obtain several task entities. Then, the data entities and knowledge paths associated with the task entities are queried in the core knowledge graph to obtain a knowledge subgraph. Based on the knowledge subgraph and historical feedback information, a pre-built intelligent service decision model is used to generate intelligent service decisions. The knowledge subgraph and intelligent service decisions are then sent to the client's intelligent agent through the data management platform to provide task support for the intelligent agent. The data management platform records the data usage behavior of intelligent agents on the knowledge subgraph based on intelligent service decisions, and uploads the data usage behavior to the cloud server through an encrypted and secure channel; In the cloud server, data usage behavior is used as a new entity relationship for all data entities in the knowledge subgraph, and the new entity relationship is added to the corresponding data entity in the core knowledge graph to obtain a second-updated core knowledge graph.

9. The knowledge graph-guided intelligent agent data lifecycle management method according to claim 8, characterized in that, During the data archiving phase, the core knowledge graph updated twice undergoes intelligent processing to eliminate low-value data entities, resulting in a core knowledge graph updated three times, including: During the data archiving phase, the value evaluation criteria for all data entities in the core knowledge graph that is updated twice are collected. Based on the value assessment criteria, a pre-built value prediction model is used to predict the future value of each data entity. Based on the preset archiving strategy and compliance requirements, data entities whose future value is below the value threshold are archived and destroyed, and the status of the corresponding data entities in the second-updated core knowledge graph is updated to obtain the third-updated core knowledge graph.

10. A knowledge graph-guided intelligent agent data lifecycle management device, used to implement the intelligent agent data lifecycle management method as described in any one of claims 1-9, characterized in that, The device includes: The knowledge graph construction unit is used to connect the data management platform to the client's intelligent agent and construct the corresponding core knowledge graph in the cloud server based on the agent's metadata. The data acquisition unit is used to collect data from the data source by using the core knowledge graph for intelligent guidance based on the task request of the intelligent agent during the data acquisition phase, and obtain a number of raw data. The data preprocessing unit is used to intelligently manage several raw data using a core knowledge graph during the data preprocessing stage, resulting in several preprocessed data. The data storage unit is used to intelligently organize several preprocessed data using the core knowledge graph during the data storage stage to obtain an updated core knowledge graph. The data application unit is used to provide intelligent services to the intelligent agent based on the agent's task information during the data application phase, using the updated core knowledge graph, to provide task support to the intelligent agent, and to record the agent's data usage behavior to obtain a second-updated core knowledge graph. The data archiving unit is used to intelligently process the core knowledge graph that has been updated twice during the data archiving stage, eliminating low-value data entities and obtaining the core knowledge graph that has been updated three times.

Images