Multi-agent-based product knowledge base construction method and system

Abstract

The invention relates to the technical field of artificial intelligence, and provides a multi-agent-based product knowledge base construction method and system, and the method comprises the steps: obtaining original data from different data sources through a plurality of data collection agents, and obtaining first data through the analysis and structural processing of a knowledge extraction agent; and inputting the first data into a reasoning agent, and generating second data through rule reasoning or neural network prediction. And then sending the second data to the user terminal through the user interaction agent, and storing the second data in the knowledge database in combination with feedback data. And finally, storing the knowledge database in a graph structure, and outputting a product knowledge base by taking the product code as a unique identifier. The invention further discloses a system for the method, the method and the system can perform parallel acquisition on the multivariate format data in the whole process of product research and development and the like, convert original data into structured knowledge, realize dynamic optimization in combination with user feedback, and finally present in a graph structure and take a product code as a unique identifier. And normalization and traceability of knowledge are ensured.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for constructing a product knowledge base based on multiple agents. Background Technology

[0002] In the entire product lifecycle management, from concept design to market withdrawal, knowledge serves as the core driving force, and its circulation efficiency and reusability directly determine product competitiveness. Knowledge carriers are highly heterogeneous, encompassing 3D models and experimental data from the R&D phase, process cards and equipment maintenance records from the production phase, user profiles and competitor analysis reports from the marketing process, and formats ranging from structured databases to unstructured audio, images, and free text. Product knowledge is highly dynamic; with iterative market demands, technological innovations, and supply chain changes, the effectiveness of existing knowledge needs to be verified in real time. However, traditional knowledge bases relying on manual maintenance often lag behind actual business changes.

[0003] CN120277140A discloses a loosely coupled knowledge management system and its construction method. The system comprises the following layers: Resource Layer: This layer collects knowledge from departments and personal computers, including structured and unstructured data. It converts unstructured data into structured data for initial storage and designs interfaces for data collection and synchronization. Resource Integration Layer: This layer integrates and processes the data from the resource layer, unifying and classifying it to provide standardized data formats for the upper layers. Resource Repository Layer: This layer stores explicit knowledge and mined tacit knowledge within the enterprise according to a knowledge repository classification directory, forming a structured knowledge repository. Service Layer: This layer provides knowledge service functions including knowledge management, unified retrieval, expert consultation, knowledge community, and a learning and training platform. Application Layer: Based on the functions of the service layer and combined with business scenarios, this layer automates and contextualizes knowledge management, constructs job knowledge maps and business process knowledge maps, and supports talent development and business empowerment. Presentation Layer: This layer presents knowledge in a visual way, facilitating user access and use of knowledge, and supports employees in independently creating and flexibly modifying process knowledge maps.

[0004] CN120219117A describes an enterprise knowledge base management system and method. It includes portal display, business processing, basic configuration, and business support, interconnected through standardized interfaces and processes to form a complete and efficient knowledge management system. The portal display serves as the primary interface for user interaction with the system, providing multiple portal types to meet the needs of different user groups. Business processing is the core of knowledge base management, responsible for the entire lifecycle management of documents, including creation, publication, updating, borrowing, retrieval, and obsolescence. Basic configuration handles the system's basic settings and process configurations, ensuring the system's flexibility and scalability. Business support provides technical support and services for the system's stable operation and efficient management.

[0005] CN120297354A discloses a multi-agent collaborative system and method based on spatial computing and multimodal AI fusion. The multi-agent collaborative system includes: a multimodal data acquisition and fusion unit, configured to acquire various real-time data from the construction site, and use multimodal AI algorithms to clean, standardize, and semantically fuse all acquired data to generate a unified semantic association model; the acquired data includes at least one or more of images, text, spatial location, and environmental parameters from the construction site; a spatial computing unit, configured to interact with the multimodal data acquisition and fusion unit, and based on the data provided by the multimodal data acquisition and fusion unit, construct a three-dimensional virtual environment of the construction site through a spatial computing engine, forming a real-time synchronous mapping of the construction site, and simulating and predicting dynamic changes during the construction process; and an agent behavior management unit, configured to be able to... The system integrates real-time environmental data provided by the fusion unit to perceive dynamic changes at the construction site and collaborates with the spatial computing unit to construct a high-precision 3D virtual environment, performing real-time synchronous mapping of the construction process. The agent behavior management unit is also configured to work in conjunction with a multi-agent collaboration unit, using reinforcement learning algorithms and distributed decision-making models to allocate tasks and plan behaviors for multiple agents. Furthermore, it can configure agents to autonomously adjust task paths, collaboration modes, and resource allocation based on real-time data and dynamic needs at the construction site. The multi-agent collaboration unit, based on real-time environmental data provided by the multimodal data acquisition and fusion unit, collaborates with the spatial computing unit to construct a 3D virtual environment of the construction site, performing synchronous mapping between the physical world and the digital model. The agent behavior management unit is also configured to enable real-time information sharing and task collaboration between agents via communication protocols.

[0006] CN118540693A discloses a method for dynamic management of vehicle network security resources based on transfer learning and multi-agent game theory. The method includes: collecting real-time data on vehicle status, traffic conditions, and environmental information; performing multimodal data fusion on the collected real-time data; treating nodes in the vehicle network system as agents and resource exchange between agents as a game, constructing a multi-agent game model; solving the game and combining the fused data to determine a dynamic resource allocation strategy among the agents.

[0007] Traditional knowledge bases that rely on manual maintenance often lag behind actual business changes. They struggle to collect and update diverse knowledge formats, such as structured databases, unstructured audio and video, and free text, across the entire process of R&D, production, and marketing in real time. Relying on manual maintenance can easily lead to knowledge obsolescence. There is an urgent need to build an intelligent product knowledge base that can adaptively process heterogeneous information, connect cross-domain knowledge, and achieve dynamic updates. Summary of the Invention

[0008] Long-term practice has revealed that with the accelerating pace of product updates, traditional methods struggle to achieve unified access and parsing of diverse knowledge formats, including structured databases, unstructured audio and video, and free text, across the entire product development, production, and marketing process. Differences in terminology across departments lead to information gaps in key knowledge, hindering the reuse of experience, causing recurring errors, and resulting in lagging dynamic knowledge management and an inability to respond to market demands in real time.

[0009] In view of this, the present invention aims to propose a method for constructing a product knowledge base based on multi-agent intelligence, comprising:

[0010] Step S1: Raw data is obtained by multiple data acquisition agents from different data sources. The raw data is then input into the knowledge extraction agent, which performs structured processing after data parsing to obtain the first data.

[0011] Step S2: Input the first data into the reasoning agent, and the reasoning agent obtains the second data based on rule-based reasoning or neural network prediction of the first data;

[0012] Step S3: Send the second data to the user terminal through the user interaction agent, and store the second data in the knowledge database in combination with the feedback data of the user terminal;

[0013] Step S4: The knowledge database is stored in the form of a graph structure. The nodes, relationships and attributes of knowledge entities are defined by the ontology management agent. The product code is used as the unique identifier of the knowledge database to output the product knowledge base.

[0014] Preferably, the product knowledge base is version controlled and monitored in real time by a knowledge update agent. If the nodes and / or relationships and / or attributes in the product knowledge base are updated, version control needs to be performed by the knowledge update agent.

[0015] Preferably, if there are conflicts in the nodes and / or relationships and / or attributes in the product knowledge base, then the cross-validation agent shall perform the verification.

[0016] Preferably, the raw data collected by multiple data acquisition agents includes different data formats. After standardizing and normalizing the raw data of different data formats, the data is input into the knowledge extraction agent and parsed into knowledge triples.

[0017] This invention also discloses a product knowledge base system based on the above-described multi-agent product knowledge base construction method, the product knowledge base system comprising,

[0018] Multiple data acquisition agents are used to collect raw data from different data sources. Each data acquisition agent collects raw data from one source and can send the raw data to the knowledge extraction agent.

[0019] A knowledge extraction agent is used to receive raw data sent from the knowledge extraction agent, parse the raw data, and output structured first data.

[0020] An inference agent is used to receive the first data, perform rule-based inference or neural network-based prediction on the first data, and output the second data.

[0021] The user interaction intelligent agent is used to receive the second data and send the second data to the user terminal according to the user interaction request information; and to store the feedback data from the user terminal in the knowledge database after matching the second data one by one.

[0022] The knowledge database is used to receive and store data sent from the user interaction agent. It is stored in a structured graph structure and outputs the product knowledge base with the product code as the unique identifier.

[0023] Preferably, the product knowledge base system further includes a knowledge update intelligent agent for version control and real-time monitoring of the product knowledge base. If nodes and / or relationships and / or attributes in the product knowledge base are updated, version control needs to be performed through the knowledge update intelligent agent.

[0024] Preferably, the product knowledge base system further includes a cross-validation agent, which is used for data conflict identification and information verification if there are conflicts in nodes and / or relationships and / or attributes in the product knowledge base.

[0025] Preferably, the knowledge database includes a graph structure generation module, used to map the data in the knowledge database into graph nodes, edges, and attribute labels to construct a knowledge graph.

[0026] The present invention also discloses an electronic device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0027] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the multi-agent product knowledge base construction method described above.

[0028] The present invention provides a machine-readable storage medium storing instructions that cause a machine to execute the multi-agent product knowledge base construction method described above.

[0029] This invention discloses a multi-agent product knowledge base construction method. Through steps S1-S4, multiple data acquisition agents first obtain raw data from different data sources. This raw data is then parsed and structured by a knowledge extraction agent to obtain first data. The first data is then input into a reasoning agent, which generates second data through rule-based reasoning or neural network prediction. Subsequently, the second data is sent to the user terminal via a user interaction agent and, combined with terminal feedback data, is stored in a knowledge database. Finally, the knowledge database is stored in a graph structure, with the ontology management agent defining the nodes, relationships, and attributes of knowledge entities, and outputting the product knowledge base using the product code as a unique identifier. Multiple data acquisition agents collect structured databases, unstructured audio / video, and free text data in a parallel manner throughout the entire product development, production, and marketing process, achieving unified access. This avoids the bottleneck of a single acquisition node, and the multi-source data coverage reduces information blind spots, acquiring both product production data and user feedback data, providing more comprehensive raw data for subsequent knowledge extraction. Each agent effectively solves the problem of information gaps caused by differences in terminology systems across departments, leading to key knowledge gaps between different departments. This invention also discloses a system for the aforementioned method of constructing a product knowledge base based on multi-agent intelligence. This method and system enhance the compatibility and flexibility of data collection, reducing the failure rate due to differences in data sources. The transformation from raw data to structured knowledge, combined with user feedback for dynamic optimization, ultimately presents the knowledge in a graph structure with product codes as unique identifiers, ensuring the standardization and traceability of the knowledge.

[0030] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0031] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0032] In the attached diagram:

[0033] Figure 1 This is a schematic diagram of a method for constructing a product knowledge base based on multiple agents according to one embodiment of the present invention;

[0034] Figure 2 This is a business logic diagram of a multi-agent product knowledge base construction method according to one embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram of node relationships in a multi-agent product knowledge base construction method according to one embodiment of the present invention;

[0036] Figure 4 This is a schematic diagram of the agent structure in a multi-agent product knowledge base construction method according to one embodiment of the present invention. Detailed Implementation

[0037] The specific embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the embodiments described herein are only for explaining and illustrating the technical solutions of the present invention, and do not constitute any limitation on the scope of protection of the present invention.

[0038] To facilitate a thorough understanding of the technical solutions of this invention by those skilled in the art, the technical solutions will be comprehensively and clearly described below in conjunction with the accompanying drawings of the embodiments. It should be understood that the described embodiments are merely some examples of the technical solutions of this invention, and not all of them. Based on the embodiments disclosed in this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0039] Furthermore, in the specification, claims, and drawings of this invention, terms such as "first," "second," and "third" are primarily used to distinguish similar technical features, rather than to limit a specific order or sequence of execution. In suitable scenarios, the data referred to by these terms can be interchanged to meet the different application requirements of the embodiments of this invention. Meanwhile, the terms "comprising," "having," and their derivatives are intended to cover non-exclusive combinations of technical elements. For example, a method, system, product, or device comprising multiple steps or units is not limited to the explicitly listed steps or units, but also includes other steps or units not explicitly listed but inherently part of the technical solution.

[0040] With the rapid pace of product updates, traditional methods struggle to achieve unified access and parsing of diverse knowledge formats, including structured databases, unstructured audio / video, and free text, across the entire product development, production, and marketing process. Cross-departmental knowledge gaps arise due to differences in terminology, leading to obstacles in experience reuse, repetitive errors, and lagging dynamic knowledge management, hindering real-time responses to market demands. This invention provides a method for constructing a product knowledge base based on multi-agent systems, such as... Figure 1-3 As shown, the method for constructing a product knowledge base based on multi-agent agents includes,

[0041] Step S1: Raw data is obtained by multiple data acquisition agents from different data sources. The raw data is then input into the knowledge extraction agent, which performs structured processing after data parsing to obtain the first data.

[0042] Step S2: Input the first data into the reasoning agent, and the reasoning agent obtains the second data based on rule-based reasoning or neural network prediction of the first data;

[0043] Step S3: Send the second data to the user terminal through the user interaction agent, and store the second data in the knowledge database in combination with the feedback data of the user terminal;

[0044] Step S4: The knowledge database is stored in the form of a graph structure. The nodes, relationships and attributes of knowledge entities are defined by the ontology management agent. The product code is used as the unique identifier of the knowledge database to output the product knowledge base.

[0045] The multi-agent product knowledge base construction method, through steps S1-S4, first involves multiple data acquisition agents obtaining raw data from different data sources. This raw data is then parsed and structured by a knowledge extraction agent to obtain the first data. Next, the first data is input into a reasoning agent, which generates second data through rule-based reasoning or neural network prediction. Subsequently, the second data is sent to the user terminal via a user interaction agent and, combined with terminal feedback data, is stored in the knowledge database. Finally, the knowledge database is stored in a graph structure, with the ontology management agent defining the nodes, relationships, and attributes of knowledge entities, and outputting the product knowledge base using the product code as a unique identifier. Multiple data acquisition agents collect structured databases, unstructured audio / video, and free text data in a parallel manner throughout the entire product development, production, and marketing process, achieving unified access. This avoids the bottleneck of a single acquisition node, and multi-source data coverage reduces information blind spots, acquiring both product production data and user feedback data, providing more comprehensive raw data for subsequent knowledge extraction. Each agent effectively solves the problem of information gaps caused by differences in terminology systems across departments, leading to key knowledge gaps between different departments. This method enhances the compatibility and flexibility of data collection, reducing the failure rate caused by differences in data sources. It transforms raw data into structured knowledge, dynamically optimizing it based on user feedback, and ultimately presents it in a graph structure with product codes as unique identifiers, ensuring the standardization and traceability of knowledge.

[0046] The data acquisition agent is responsible for scraping raw data from various data sources, such as databases, text files, web pages, and sensors from different departments within an enterprise. It can process multi-source data in parallel, improving acquisition efficiency. The agent incorporates multiple data interfaces, such as database APIs, web crawler protocols, and file parsers, supporting integration with different types of data sources, including MySQL databases, PDF documents, and sensor data streams, ensuring format compatibility. Tasks are allocated based on preset rules, controlling multi-threaded parallel acquisition to avoid data conflicts (i.e., repeated scraping from the same data source). Preliminary checks are performed on data integrity and validity, such as checking the correct date format, and filtering obviously invalid data, such as null values ​​and garbled characters.

[0047] The knowledge extraction agent parses the raw data, performing tasks such as data cleaning, deduplication, format conversion, and structuring. This includes extracting key information, dividing fields, and transforming unstructured and semi-structured data into structured data (field-value pairs with a unified format), ultimately outputting the first data. For example, if the data source is an unstructured product manual or a semi-structured production record table, information such as "product name," "specification," and "production batch" is extracted to form structured data containing these fields. The data cleaning module performs deduplication; ideally, hash-based comparison of duplicate data is used. Noise reduction removes irrelevant characters, such as residual code from web page advertisements. Units and formats are standardized, for example, converting "5000 mAh" to "5000mAh." The first data, structured data (e.g., a JSON / CSV file containing fields such as "product name," "battery capacity," and "user review keywords"), is input into the inference agent. This data is passed through the inference agent's input interface, automatically matching the required field information to the module. For example, the rule engine requires a "battery capacity" field, and the neural network model requires a "user review keyword" field. Data format validation is performed, such as verifying that numeric fields are indeed numbers and that text fields meet length requirements. If any data is missing (e.g., "battery capacity" is empty), a completion mechanism is triggered, which calls historical data from the knowledge base to fill in default values, ensuring the data meets the inference requirements.

[0048] Rule-based reasoning retrieves rules related to the first data field from a pre-defined business rule base. For example: Rule 1: If battery capacity ≥ 4500mAh, then battery life rating = Excellent; if 3000-4499mAh, then battery life rating = Medium; if < 3000mAh, then battery life rating = Poor. Rule 2: If user reviews include "lag" and "RAM ≥ 8GB", then it's determined to be a "system software problem". The field values ​​of the first data are substituted into the rules for judgment. For example: If the first data contains "battery capacity = 5000mAh", matching Rule 1 will output "battery life rating = Excellent".

[0049] In the first set of data, "user review keyword = lag" and "RAM = 12GB" match rule 2, resulting in the output "problem type = system software problem". The output of the reasoning agent includes the generation rule reasoning result, which adds new fields such as "battery life level" and "problem type".

[0050] The neural network-based prediction process first transforms the unstructured / semi-structured fields in the initial data. Text fields, such as "user review keywords = blurry photos, severe overheating," are converted into word vectors through the "feature processing submodule." Ideally, semantic vectors are generated based on the BERT model. Numerical / categorical fields, such as "production batch = 20250624-B" and "screen size = 6.7 inches," are normalized and encoded into an input format recognizable by the model.

[0051] The pre-trained model is invoked; more preferably, a graph neural network is used for fault prediction, and a classification model is used for risk assessment, inputting processed features and historical correlation data from a knowledge database. For example, fault records from the same batch of products. The model inputs the word vector for "blurry photo" plus the feature for "camera fault" from historical data, outputting a prediction result of "potential fault probability: camera module = 70%, software algorithm = 30%". The generated neural network prediction result includes fields such as "potential fault probability" and "risk level".

[0052] After receiving the second data output by the inference agent, the user interaction agent first converts it into a display format supported by the target user's terminal, such as a PC for R&D personnel, a mobile app for after-sales personnel, or a large screen system in the production workshop. For example, it might be an editable table on a PC, a simple card on a mobile device, and a visual chart on a large screen. For instance, the second data "Product Name = Model X, Battery Life Rating = Excellent, Problem Type = Hardware Compatibility Issue, Potential Failure Probability = 65%" is converted into a table with field explanations on a PC and simplified into a text card "Model X: Excellent Battery Life, Hardware Compatibility Risk" on a mobile device. Based on the user role, for example, R&D personnel can view the complete data, while after-sales personnel only need the "Problem Type" and "Failure Probability." Sensitive or irrelevant information is filtered through the permission management submodule of the "Data Display Module" to ensure the accuracy and security of the data display.

[0053] Through the terminal interface, using HTTP protocol and other methods, and according to preset rules, such as real-time push and scheduled summary push, the adapted second data is sent to the user terminal. For example, real-time push is for urgent information, such as "potential fault probability ≥ 80%", which is immediately pushed via terminal pop-up, SMS, or APP notification. Scheduled push is for regular data, such as daily product quality summaries, which are sent to the "Data Center" section of the terminal at fixed times. After receiving the data, the user terminal automatically returns a "read", "unread", or "reception failed" status signal to the user interaction agent.

[0054] The user terminal's feedback data provides dropdown selection boxes for specific fields. For example, a dropdown selection box for "Problem Type" includes options such as "Confirm," "Correct to: Software Problem," or "Further Verification Required." For complex situations, text input boxes and image uploads are provided, such as fault cause analysis, fault screenshots, and voice messages. Furthermore, in a more preferred embodiment of the invention, it also includes operation behavior feedback, automatically recording user interaction behaviors such as "Click to view 'Potential Fault Probability' details" or "Modify 'Problem Type' field," as implicit feedback data. After the user submits feedback, the terminal sends the feedback information, including user ID, operation time, and feedback content, to the user interaction agent via an encrypted channel. The user interaction agent sends the fused complete data, including the second data, structured feedback data, and associated identifiers, to the knowledge database. The knowledge database's access module first converts the data into the format required for graph structure storage. For example, merging data such as "Product Code = FXF20250624, Problem Type = Software Problem - User Correction, Feedback Provider = R&D Department - Engineer Fang" is converted into graph node attributes. The "Problem Type" for the value "FXF20250624" is updated to "Software Problem," and a "Feedback Provider" attribute is added. Core information requiring long-term storage is filtered from the second set of data and user feedback data, including basic product information such as name, code, parameters, and entity association information such as component composition and fault association. User feedback information undergoes final cleaning, such as removing duplicate feedback records and verifying field consistency. Based on the storage granularity of the data feature planning graph, "Product" is used as the core node, associated with auxiliary nodes such as "Component," "Fault," and "User," ensuring a clear hierarchical relationship. For example, "Product-Component" has an "Inclusion" relationship, and "Product-Fault" has an "Existence" relationship. Extended fields are reserved for potential additions such as a "Supply Chain" node.

[0055] The knowledge database transforms from scattered, fused data into a structured knowledge base centered on product codes and using a graph structure as its carrier. This preserves the complex relationships between entities while ensuring the uniqueness and traceability of knowledge, providing precise knowledge support for decision-making throughout the product lifecycle. Nodes define the type and scope of knowledge entities, relationships describe the logical connections between nodes, and attributes supplement the characteristics and details of the nodes. The definition of intelligent agents is managed by the ontology: for example, a product node, such as "Smartphone Model X," represents the core knowledge entity. Component nodes, such as Snapdragon 8 Gen3 and camera module, represent the components of the product. Fault nodes, such as software algorithm bugs or hardware compatibility issues, represent abnormal states of the product. A graph structure is constructed using the product code as a unique identifier for storage. Figure 3 As shown, unique identifier binding sets the product code, such as FXF20250624, as the primary key attribute of the product node and establishes a mapping through all associated nodes. For example, the component node Snapdragon8Gen3 is bound to FXF20250624 through the "belongs to" relationship, and the fault node software algorithm BUG is bound to FXF20250624 through the "association" relationship, ensuring that all nodes can ultimately be traced back to the core product code.

[0056] To ensure the dynamic iteration capability of the product knowledge base and adapt to real-time changes in product information, strict version control is implemented to guarantee the accuracy and reliability of knowledge. In a more preferred embodiment of this invention, a knowledge update agent performs version control and real-time monitoring of the product knowledge base. If nodes and / or relationships and / or attributes in the product knowledge base are updated, version control is performed through the knowledge update agent. Each update of a node, relationship, or attribute corresponds to a unique version number, supporting historical state backtracking; for example, querying the knowledge status of a product code in August 2023. Real-time monitoring of update operations ensures that newly added or modified content conforms to the rules defined by the ontology management agent, such as node attribute formats and relationship type specifications, preventing illegal data from intruding into the knowledge base. The knowledge base is dynamically updated along with the product lifecycle, such as iteration upgrades and fault repairs, ensuring that knowledge is synchronized with the actual product status and providing the latest knowledge support for R&D, after-sales, and other stages. If the update passes the verification, a new version number will be automatically generated, using the format of base version + timestamp + update sequence number. For example, if the original version is V20250729, the first update on August 1st will be V20250801-1. The base version is associated with the product code, such as version FXF20250624, which will be prefixed with it to ensure unique identification binding.

[0057] To efficiently resolve various conflicts in the knowledge base and further optimize the knowledge system through conflict cases, thereby improving the reliability and business adaptability of the product knowledge base, in a more preferred embodiment of this invention, if conflicts occur in nodes and / or relationships and / or attributes within the product knowledge base, a cross-validation agent is used for verification. The camera module pixel attribute of product code FXF20250624 has both 50 million and 64 million pixels. The production BOM table shows 50 million pixels (90% confidence level), while the e-commerce details page shows 64 million pixels (60% confidence level). The EXIF ​​information of user-taken sample photos shows 50 million pixels (80% confidence level), and the supplier's official website parameters show 50 million pixels (95% confidence level).

[0058] After comprehensive credibility calculation during cross-validation, the score for 50 million was 0.3×90+0.2×0+0.2×80+0.3×95=83.5, which is significantly higher than the score for 64 million, which was 0.3×0+0.2×60+0.2×0+0.3×0=12. The knowledge base attribute was corrected to 50 million, the e-commerce details page data was marked as promotional error, and the validation rules were updated to prioritize pixel values ​​from supplier official website data.

[0059] To enable unified processing of unstructured product manuals, semi-structured after-sales work orders, and structured production records, thus resolving the data silo problem, standardization and normalization eliminate format noise, allowing the knowledge extraction agent to accurately identify key information and reduce parsing errors caused by format differences.

[0060] By using unified naming rules, relation types, and attribute formats, the parsed triples are ensured to conform to the ontology definition, avoiding the confusion of multiple expressions of the same relation in the subsequent knowledge base, thus laying a standardized foundation for knowledge graph construction. The standardized structured data can be directly called by the algorithm of the knowledge extraction agent, reducing repetitive format adaptation work and improving triple parsing efficiency. In a more preferred embodiment of this invention, the raw data collected by multiple data acquisition agents includes different data formats. After standardizing and normalizing the raw data of different formats, it is input into the knowledge extraction agent and parsed into knowledge triples. For example, using an image recognition model to extract text information from an image, such as a camera module model CM-2025, it is converted into a key-value pair format {component: camera module, model: CM-2025}. The standardized and normalized data is input into the knowledge extraction agent, which generates triples (subject-relation-object) through a combination of rules and models. For structured / semi-structured data, triples are extracted based on a preset template. For example, the tabular data "Product Name: Model-X, Processor: Snapdragon 8 Gen3" is converted into the triple (Model-X, equipped with, Snapdragon 8 Gen3). The key-value pair "Camera Module - Supplier: XX Technology" is converted into the triple (Camera Module, Supplier, XX Technology).

[0061] This invention also discloses a product knowledge base system based on the above-described multi-agent product knowledge base construction method, the product knowledge base system comprising,

[0062] Multiple data acquisition agents are used to collect raw data from different data sources. Each data acquisition agent collects raw data from one source and can send the raw data to the knowledge extraction agent.

[0063] A knowledge extraction agent is used to receive raw data sent from the knowledge extraction agent, parse the raw data, and output structured first data.

[0064] An inference agent is used to receive the first data, perform rule-based inference or neural network-based prediction on the first data, and output the second data.

[0065] The user interaction intelligent agent is used to receive the second data and send the second data to the user terminal according to the user interaction request information; and to store the feedback data from the user terminal in the knowledge database after matching the second data one by one.

[0066] The knowledge database is used to receive and store data sent from the user interaction agent. It is stored in a structured graph structure and outputs the product knowledge base with the product code as the unique identifier.

[0067] In this product knowledge base system, multiple data acquisition agents collect raw data from different data sources and send it to a knowledge extraction agent. The knowledge extraction agent parses the raw data and outputs structured first data. A reasoning agent receives the first data and outputs second data through rule-based reasoning or neural network-based prediction. A user interaction agent receives the second data, sends it to the user terminal according to user interaction requests, and stores the user terminal's feedback data in the knowledge database after matching it with the second data. The knowledge database stores data in a structured graph structure and ultimately outputs the product knowledge base using the product code as a unique identifier. This system significantly improves the compatibility and flexibility of data acquisition by adapting to multi-source heterogeneous data, effectively reducing the risk of acquisition failures caused by differences in data source formats and protocols. In the data processing stage, the system achieves accurate transformation from raw data to structured knowledge and forms a dynamic optimization loop based on user feedback. Finally, it constructs a knowledge network in graph structure form, using the product code as a globally unique identifier, comprehensively ensuring the standardization and full lifecycle traceability of the knowledge system.

[0068] Data acquisition agent, knowledge extraction agent, reasoning agent, user interaction agent, knowledge update agent, and cross-validation agent are all AI agents, and their structures are as follows: Figure 4 As shown, the AI ​​agent is based on a large model. The Planning module's main function is for the agent to break down large tasks into sub-tasks and plan the execution process. The agent reflects on and analyzes the task execution process to decide whether to continue or terminate the task. The Memory module includes short-term and long-term memory. Short-term memory refers to the context during task execution, generated and temporarily stored during sub-task execution, and cleared after task completion. Long-term memory is information retained for a long time, generally referring to external knowledge bases, usually stored and retrieved using vector databases. The Tool module equips the agent with tool APIs, such as calculators, search tools, code executors, and database query tools. With these tool APIs, the agent can interact with the physical world and solve practical problems.

[0069] To ensure the dynamic iteration capability of the product knowledge base and adapt to real-time changes in product information, strict version control is used to ensure the accuracy and reliability of the knowledge. In a more preferred embodiment of the invention, the product knowledge base system further includes a knowledge update agent for version control and real-time monitoring of the product knowledge base. If nodes and / or relationships and / or attributes in the product knowledge base are updated, version control is required through the knowledge update agent.

[0070] To monitor and capture updates to nodes, relationships, and attributes in real time, ensuring accurate recording of every change, a version control mechanism is used to construct a complete version hierarchy for the knowledge base, enabling rapid tracing from the initial version to any updated version. Simultaneously, strict version management avoids knowledge chaos caused by collaborative updates from multiple users or the import of data from multiple sources, ensuring the consistency and integrity of different versions of the knowledge base. In a more preferred embodiment of this invention, the product knowledge base system also includes a cross-validation agent, used for data conflict identification and information verification if conflicts occur in nodes and / or relationships and / or attributes within the product knowledge base.

[0071] To break down the isolation between data in structured storage, this invention uses nodes to represent knowledge entities, edges to reflect relationships between entities, and attribute tags to supplement entity features. This allows scattered information to form an interconnected organic whole, intuitively presenting a knowledge network throughout the product's lifecycle and facilitating rapid tracing of deep relationships between entities. Simultaneously, the graph structure makes knowledge querying and expansion more flexible and efficient. It allows for quick location of relevant edges and attributes through nodes and easy addition of new nodes, edges, or attributes to adapt to dynamic knowledge growth. This provides a structured and interconnected foundation for subsequent knowledge graph-based reasoning and analysis, significantly enhancing the utilization value of knowledge and the system's intelligence level. In a more preferred embodiment, the knowledge database includes a graph structure generation module for mapping data in the knowledge database into graph nodes, edges, and attribute tags to construct a knowledge graph.

[0072] This invention provides an electronic device, comprising at least one processor; and

[0073] A memory communicatively connected to the at least one processor; wherein,

[0074] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the above-described method for constructing a product knowledge base based on a multi-agent agent.

[0075] The present invention provides a machine-readable storage medium storing instructions that cause a machine to execute the multi-agent product knowledge base construction method described above.

[0076] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0077] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0078] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a product knowledge base based on multi-agent agents, characterized in that, The method for constructing a product knowledge base based on multi-agent intelligence includes: Step S1: Raw data is obtained by multiple data acquisition agents from different data sources. The raw data is then input into the knowledge extraction agent, which performs structured processing after data parsing to obtain the first data. Step S2: Input the first data into the reasoning agent, and the reasoning agent obtains the second data based on rule-based reasoning or neural network prediction of the first data; Step S3: Send the second data to the user terminal through the user interaction agent, and store the second data in the knowledge database in combination with the feedback data of the user terminal; Step S4: The knowledge database is stored in the form of a graph structure. The nodes, relationships and attributes of knowledge entities are defined by the ontology management agent. The product code is used as the unique identifier of the knowledge database to output the product knowledge base.

2. The method for constructing a product knowledge base based on multiple agents according to claim 1, characterized in that, In step S4, the product knowledge base is version controlled and monitored in real time by the knowledge update agent. If the nodes and / or relationships and / or attributes in the product knowledge base are updated, version control needs to be performed by the knowledge update agent.

3. The method for constructing a product knowledge base based on multiple agents according to claim 2, characterized in that, If there are conflicts in the nodes and / or relationships and / or attributes in the product knowledge base, then the cross-validation agent will perform the verification.

4. The method for constructing a product knowledge base based on multiple agents according to any one of claims 1-3, characterized in that, The raw data collected by multiple data acquisition agents includes different data formats. After standardizing and normalizing the raw data of different data formats, the data is input into the knowledge extraction agent and parsed into knowledge triples.

5. A product knowledge base system based on the multi-agent product knowledge base construction method according to any one of claims 1-4, characterized in that, The product knowledge base system includes, Multiple data acquisition agents are used to collect raw data from different data sources. Each data acquisition agent collects raw data from one source and can send the raw data to the knowledge extraction agent. A knowledge extraction agent is used to receive raw data sent from the knowledge extraction agent, parse the raw data, and output structured first data. An inference agent is used to receive the first data, perform rule-based inference or neural network-based prediction on the first data, and output the second data. The user interaction intelligent agent is used to receive the second data and send the second data to the user terminal according to the user interaction request information; The feedback data from the user terminal is matched one-to-one with the second data and then stored in the knowledge database; The knowledge database is used to receive and store data sent from the user interaction agent. It is stored in a structured graph structure and outputs the product knowledge base with the product code as the unique identifier.

6. The product knowledge base system according to claim 5, characterized in that, The product knowledge base system also includes a knowledge update intelligent agent, which is used to perform version control and real-time monitoring of the product knowledge base. If the nodes and / or relationships and / or attributes in the product knowledge base are updated, version control needs to be performed through the knowledge update intelligent agent.

7. The product knowledge base system according to claim 5, characterized in that, The product knowledge base system also includes a cross-validation agent, which is used for data conflict identification and information verification if there are conflicts in nodes and / or relationships and / or attributes in the product knowledge base.

8. The product knowledge base system according to claim 5, characterized in that, The knowledge database includes a graph structure generation module, which is used to map the data in the knowledge database into graph nodes, edges, and attribute labels to construct a knowledge graph.

9. Electronic devices, including: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the multi-agent product knowledge base construction method as described in any one of claims 1-4.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to execute the multi-agent product knowledge base construction method of the present invention as described in any one of claims 1-4.

Images