Knowledge layer and intelligence layer learning method and system of netted DIKWP model

Abstract

The invention provides a learning method and system for a knowledge layer and an intelligent layer of a netted DIKWP model. According to the method, a knowledge graph (including semantic association new information, knowledge reasoning extension and probability screening confirmation) is dynamically constructed and updated through a knowledge layer, high-level decision reasoning is performed on the knowledge graph through an intelligent layer, and mechanisms such as introspection meta-learning and analogy reasoning are fused to generate a solution. A 'BUG' strategy is introduced into the system, mild cognitive deviation is intentionally added in intelligence layer reasoning to stimulate creative thinking, and meanwhile, bidirectional feedback between an intelligence layer and a knowledge layer is established, so that decision output can reversely perfect a knowledge base, and closed-loop adaptive learning is realized. The method and the system support integrated deployment of software and hardware, can be compatible and combined with a large-scale pre-training model, and have wide application prospects in the fields of judicial decision assistance, personalized education and the like.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and cognitive computing technology, and in particular to a learning method and system for the knowledge layer and wisdom layer of a mesh DIKWP model. Background Technology

[0002] In the fields of artificial intelligence and knowledge management, the commonly used DIKW model divides data, information, knowledge, and wisdom into a bottom-up hierarchical structure. The traditional DIKW pyramid assumes that information and knowledge can be extracted through the hierarchical processing of data, ultimately leading to intelligent decision-making. However, this static hierarchical model suffers from insufficient flexibility: real-world knowledge bases are often pre-built and difficult to update dynamically, and high-level decision-making processes lack mechanisms for adaptive adjustments based on specific goals and contexts.

[0003] To overcome the limitations of the DIKW model, an extended DIKWP model has been proposed in recent years, adding an "Intention" layer at the top of the data-information-knowledge-wisdom system. This five-layer model (Data→Information→Knowledge→Wisdom→Purpose) emphasizes intention-driven cognitive processes, achieving a closed-loop iteration from data perception to decision execution. By introducing the intention layer, AI systems can dynamically allocate lower-level resources based on user or environmental goals, improving the targeting and flexibility of decision-making. However, current implementations of the higher layers of the DIKWP model (especially the knowledge and wisdom layers) remain incomplete, and the industry lacks effective methods to tightly integrate knowledge acquisition with intelligent reasoning and merge it with intention-driven mechanisms.

[0004] Existing knowledge representation and reasoning technologies also have some shortcomings. Knowledge graphs, as an important form of knowledge layer implementation, have been widely used in systems such as search engines and intelligent question answering. However, the construction of traditional knowledge graphs usually relies on offline organization and manual input, lacking the ability to update knowledge in real time with changes in environment and data. On the other hand, although many artificial intelligence decision-making systems (such as large-scale pre-trained language models, LLM) implicitly contain a large amount of knowledge in their parameters, this knowledge is stored in a distributed representation, making it very difficult to update or precisely control specific knowledge content. In addition, the reasoning mechanism of large models is like a black box, lacking a transparent knowledge reasoning chain, making the decision-making process difficult to interpret, thus limiting its credibility in professional applications.

[0005] In high-level intelligent decision-making, cutting-edge methods such as meta-learning and analogical reasoning have been proposed to enhance AI's generalization ability and human-like thinking. Meta-learning enables AI to learn "how to learn" from past tasks, quickly adjusting its strategies when facing new problems; analogical reasoning mimics the human process of inference through similar cases, helping AI transfer existing knowledge in unfamiliar situations. However, current mainstream AI systems rarely combine meta-learning or analogical reasoning with explicit knowledge bases, resulting in a lack of flexible reasoning and rapid adaptation to new tasks at their intelligence layer, making it difficult to achieve truly human-like high-level cognition.

[0006] Furthermore, traditional AI systems typically strive for logical rigor and definitive results, minimizing errors or biases. However, in creative research, a moderate degree of "cognitive bias" can actually be a catalyst for innovation. Some researchers have proposed the "BUG theory," advocating for allowing intentional, slight imperfections in AI decision-making processes to help the system break free from existing patterns and arrive at unexpected new solutions. In other words, deliberately introducing subtle random perturbations or biases can help algorithms escape local optima and expand problem-solving approaches. Most existing AI algorithms do not consider this mechanism; overly rigorous reasoning processes may actually limit creativity.

[0007] Another pressing issue is the inter-layer feedback learning mechanism. Most current AI architectures operate unidirectionally, processing knowledge step-by-step from lower-level perception to higher-level decision-making, stopping once a result is output, lacking feedback optimization of previous stages. If omissions or errors exist in the knowledge layer, higher-level decisions do not prompt the knowledge base to self-correct, causing system performance to be consistently limited by the initial knowledge quality. Especially in complex and changing environments, systems lacking feedback mechanisms struggle to correct errors and acquire new knowledge in a timely manner, impacting the accuracy and robustness of decisions. Therefore, a bidirectional interactive mechanism between upper and lower layers is needed, enabling higher-level intelligent decisions to guide the acquisition and updating of lower-level knowledge, forming a self-evolving closed-loop learning system.

[0008] Finally, with the rapid development of artificial intelligence technology, effectively deploying the aforementioned intelligent mechanisms in practical applications is also a challenge. On the one hand, it is necessary to consider integrated hardware and software deployment, that is, to utilize the powerful computing resources of the cloud while achieving low-latency real-time inference on local devices or dedicated chips to meet the performance requirements of different scenarios. On the other hand, new methods should be compatible with current mainstream large-scale models, leveraging their powerful perception and language capabilities while compensating for their shortcomings such as slow knowledge updates and uninterpretable reasoning. However, existing technologies lack targeted designs for hardware-software collaboration and integration with large-scale models, making it difficult to efficiently implement some innovative algorithmic ideas in industrial applications.

[0009] In summary, existing technologies lack a complete solution capable of achieving dynamic knowledge graph generation, high-level intelligent decision-making (including introspective learning and analogical reasoning), introducing creative bias mechanisms, and inter-layer bidirectional feedback optimization, while also considering hardware and software deployment and large-scale model integration. To address these issues, this invention provides a learning method and system for the knowledge and intelligence layers of a mesh DIKWP model, thus overcoming the shortcomings of existing technologies. Summary of the Invention

[0010] The main objective of this invention is to address the shortcomings of the aforementioned background technology and provide a learning method and system for the knowledge layer and intelligence layer of a mesh DIKWP model. This invention aims to: dynamically generate and update the knowledge graph to maintain the real-time performance and integrity of the knowledge base; achieve adaptive decision-making and reasoning of the intelligence layer based on knowledge, possessing human-like intelligent characteristics such as introspective meta-learning and analogical reasoning; introduce a suitable cognitive bias mechanism to stimulate the creativity of the AI ​​system; establish a bidirectional feedback channel between the knowledge layer and the intelligence layer to continuously optimize overall performance; and design a system architecture adapted to integrated hardware and software deployment and large-scale model collaboration, thereby comprehensively improving the knowledge acquisition capability, reasoning and decision-making level, and application flexibility of the artificial intelligence system.

[0011] To achieve the above objectives, this invention provides a learning method and system for the knowledge layer and intelligence layer of the DIKWP model. In terms of overall architecture, the system comprises two core components: a knowledge layer module and an intelligence layer module, which work collaboratively through intent-driven and feedback mechanisms. The technical solution of this invention is summarized below in conjunction with the functions of each component:

[0012] First, the dynamic knowledge graph generation mechanism of the knowledge layer: The knowledge layer of this invention is used to process and elevate the data provided by the information layer into a structured knowledge graph. Specifically, the system obtains pre-processed semantic data from the information layer, performs structured parsing on the data, extracts knowledge nodes and their relational edges, and continuously constructs and expands the knowledge graph. During the knowledge formation process, mechanisms such as semantic association, probabilistic verification, and knowledge reasoning are employed to ensure the accuracy and relevance of knowledge acquisition. Driven by user or environmental intent, the system semantically associates new information with existing knowledge and performs reasoning deduction, dynamically adding or updating nodes and relationships in the knowledge graph; and through probabilistic evaluation, it filters out high-confidence associations, removes noise or unreliable new knowledge, and ensures that the content included in the knowledge base is semantically consistent and reasonable.

[0013] Secondly, the learning and decision-making mechanism of the intelligence layer: The intelligence layer comprehensively utilizes the knowledge graph provided by the knowledge layer at a higher level to complete decision-making and reasoning tasks in complex situations. The intelligence layer employs an introspective meta-learning strategy and a human-inspired analogical reasoning mechanism: First, it abstracts existing domain knowledge and historical experience, summarizing them into general strategy templates; when faced with new problems, it uses semantic mapping to project the problem onto relevant nodes and subgraphs in the knowledge graph, deducing solutions by analogy with existing cases and knowledge. In this process, the intelligence layer fully considers the decision-making context and user intent, ensuring that the reasoning process conforms to human cognitive habits. For example, in a judicial judgment context, the intelligence layer can analogically match similar cases and relevant legal provisions in the knowledge graph to deduce a reference judgment; in an educational context, the intelligence layer can generate personalized teaching strategies based on students' knowledge graphs. By combining the "contextual understanding" and "intent perception" mechanisms in human cognition, the decision-making results of the intelligence layer are closer to actual needs and possess interpretability.

[0014] Furthermore, the introduction of the "BUG" creative bias mechanism: This invention incorporates an intentional cognitive bias strategy (i.e., the BUG mechanism) into the reasoning process of the intelligent layer. The intelligent layer allows for the introduction of moderate random perturbations or imperfect matches during decision-making reasoning to prevent the algorithm from falling into rigid logical frameworks, thereby stimulating creative solution approaches. Specifically, the system selectively adds slight random factors or atypical associations to the knowledge reasoning chain, prompting the intelligent layer to explore unconventional solution paths. This "intentional defect" mechanism helps AI escape the local optima traps that traditional algorithms may fall into, generating unexpected new insights and improving the creativity and diversity of problem-solving.

[0015] Furthermore, the invention employs a two-way feedback optimization mechanism between the knowledge layer and the wisdom layer: emphasizing the closed-loop interaction between knowledge acquisition and intelligent decision-making. While outputting decision solutions, the wisdom layer verifies and evaluates the evidence provided by the knowledge layer. If it discovers incomplete or logically flawed reasoning links in the knowledge graph, the wisdom layer generates feedback information, sending it back to the knowledge layer and even the information layer to prompt targeted supplementation of new data or correction of existing knowledge. This two-way feedback between layers enables the system to continuously improve itself: high-level decisions drive the evolution of underlying knowledge, and new knowledge, in turn, enhances the quality of subsequent decisions. Through multiple iterations, the system's overall cognitive ability continuously strengthens, forming a dynamic learning architecture with interconnected networks.

[0016] Finally, the integrated hardware and software deployment and large-scale model compatibility design: While realizing the aforementioned intelligent functions, the architecture of this invention specifically considers deployment and compatibility requirements. On the one hand, the intelligent layer decision-making module can be flexibly deployed on cloud servers or local dedicated hardware: cloud deployment can utilize powerful computing resources to handle complex reasoning, while local deployment (such as on edge devices or AI chips) meets real-time requirements. On the other hand, the system can seamlessly integrate with mainstream large-scale pre-trained models (such as large language models): leveraging the superior perceptual understanding capabilities of large models to provide richer input to the information layer, or providing the knowledge graph and reasoning results generated by this invention to large models to improve their contextual understanding and reasoning accuracy. Through this design, the method and system of this invention can quickly absorb and utilize the latest large-scale model results and adapt to different hardware and software environments, achieving efficient deployment and iterative upgrades in various applications such as judicial auxiliary decision-making, intelligent educational assessment, and intelligent customer service.

[0017] Compared with the prior art, the beneficial effects of the present invention are reflected in the following aspects:

[0018] Dynamic knowledge updates: By introducing a dynamic knowledge graph generation mechanism, the system can continuously expand and update the knowledge base as the environment and data change, overcoming the shortcomings of traditional static knowledge bases that are difficult to maintain, and enabling AI to always master the latest relevant knowledge.

[0019] Intelligent Adaptive Decision Making: The intelligent layer employs techniques such as introspective meta-learning and analogical reasoning, which can flexibly adjust decision-making strategies according to new problems. It can also quickly generalize solutions even in the absence of a large amount of training data, greatly improving the AI's autonomous decision-making ability in unfamiliar situations.

[0020] Enhanced Creativity: By introducing intentional biases through the bug mechanism, the system's decisions are no longer confined to existing patterns, enabling it to explore unconventional paths. Therefore, it exhibits stronger problem-solving capabilities in tasks requiring innovative thinking. Compared to algorithms that solely pursue rigor, the method of this invention is more creative and diverse.

[0021] Closed-loop continuous learning: Two-way feedback between the knowledge layer and the wisdom layer enables the system to form a self-improving closed loop. High-level feedback can promptly correct errors or deficiencies in the knowledge layer, avoiding the accumulation and stagnation of errors, ensuring that system performance continuously improves with use, and exhibiting higher robustness and reliability.

[0022] Flexible deployment and strong compatibility: This invention's system supports collaborative deployment on the cloud and terminals, capable of handling complex calculations while meeting real-time response requirements. Furthermore, its design is compatible with large models, enabling the system to combine the advantages of symbolic knowledge reasoning and deep learning. It possesses both interpretable reasoning chains and a vast amount of underlying knowledge, significantly enhancing the system's practicality and application scope. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system architecture for the collaborative work of the knowledge layer and the wisdom layer in this invention.

[0024] Figure 2 This is a schematic diagram of the process of dynamically constructing a knowledge graph in the knowledge layer of this invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments, but the scope of protection of the present invention is not limited thereto.

[0026] like Figure 1 As shown, this invention proposes a knowledge layer-intelligence layer learning system architecture based on the DIKWP model. The system mainly consists of two parts: a knowledge layer module and an intelligence layer module, with mechanisms for intent input and feedback control. The knowledge layer module (e.g., a knowledge graph database) receives data input from the information layer and performs knowledge generation and updating; the intelligence layer module (e.g., an intelligent decision engine) uses the knowledge provided by the knowledge layer to perform high-level reasoning and decision-making, while simultaneously transmitting feedback information back to the knowledge layer for optimization. The two modules work collaboratively under the intent driven by the user or environment: the intelligence layer acquires relevant knowledge from the knowledge layer based on the intent and provides decision results; when insufficient knowledge is detected, it requests the knowledge layer to improve the knowledge in the corresponding domain through the feedback mechanism. The entire system forms a network-like interactive cognitive architecture: a positive transmission from data to knowledge to intelligence, and a negative feedback from intelligent decision-making to knowledge updates, achieving a closed-loop learning process similar to human cognition. Figure 1 A schematic diagram illustrating the framework of collaborative operation at various levels of the system of this invention is shown.

[0027] As above Figure 1 As shown, the user's goal or the environment's intent serves as the driving input for the intelligent layer's decision-making, while the knowledge graph of the knowledge layer provides the background knowledge support needed for decision-making. While generating decision / action outputs, the intelligent layer returns knowledge gaps discovered during the cognitive process to the knowledge layer through feedback channels, thereby prompting the knowledge layer to update and improve the corresponding knowledge, achieving closed-loop optimization between layers.

[0028] Next reference Figure 2 The flowchart illustrating the dynamic generation of knowledge graphs in the knowledge layer of this invention is as follows. The knowledge layer module implements a series of steps to gradually transform raw data into structured knowledge:

[0029] like Figure 2 As shown, the knowledge layer performs semantic-level association and understanding of the data from the information layer, gradually extracting new knowledge and integrating it into the knowledge graph. The process includes the following main steps:

[0030] Semantic preprocessing and association: The system acquires raw data from the information layer (e.g., pre-processed information such as text and sensor signals), identifies key information units, and performs semantic matching and association with entities and relationships in the existing knowledge graph. For example, if the information layer inputs a new legal text, the system will identify the relevant legal clauses, subject actions, etc., find the corresponding legal concept nodes in the knowledge graph, and establish a link between the two.

[0031] Knowledge Reasoning and Expansion: Building upon semantic association, the knowledge layer uses reasoning modules to deduce and expand new information, extracting potential implicit knowledge. For example, based on causal rules in existing knowledge graphs, it can infer the legal areas that new regulations might affect, or deduce new relationships behind the data based on industry experience. In this process, the system may add new nodes (new concepts or instances) or new relational edges, making the knowledge implicit in the original data explicit.

[0032] Probabilistic screening and verification: For candidate knowledge obtained through reasoning, the system evaluates its reliability through probability assessment or confidence calculation. Only when the confidence level of a new piece of knowledge reaches a predetermined threshold is it considered valid. Conversely, if the probability of a reasoning association is low, it is not included in the knowledge graph. This step is equivalent to quality control of new knowledge, filtering out noise or erroneous inferences to ensure the accuracy and consistency of the knowledge base.

[0033] Knowledge Graph Update: New knowledge, after passing the above screening and confirmation, is formally integrated into the knowledge graph database. The system adds new nodes and relationships to the graph according to a standard knowledge representation format and maintains the consistency of the global topology. If new knowledge conflicts with or is redundant with existing knowledge, the system will label or merge it. After the update is completed, the knowledge graph has undergone a dynamic expansion, improving both its structure and content, and providing more comprehensive new knowledge for subsequent intelligent layer decision-making.

[0034] Through the above process, the knowledge layer can continuously evolve as information input changes, maintaining the freshness and accuracy of the knowledge graph. It is worth mentioning that this dynamic construction process can be executed automatically during system operation, enabling online learning and updates, thus ensuring that the intelligence layer always bases its decision-making on the latest and most comprehensive knowledge.

[0035] The method and system of the present invention will be described below in conjunction with specific application scenarios.

[0036] This embodiment describes the application of the present invention in the judicial field. Based on the knowledge-intelligence layer learning system of the present invention, an intelligent judicial auxiliary decision-making platform can be constructed, providing judges, lawyers, and others with case analysis and judgment suggestions.

[0037] In this platform, the information layer can draw from legal document databases, case libraries, and real-time collected case information. The knowledge layer utilizes this data to construct a legal knowledge graph: nodes include legal provisions, legal concepts, historical cases, and rulings, etc., and nodes are connected by relational edges, such as reflecting the hierarchical relationship between legal provisions, the association between cases and applicable legal provisions, and the correspondence between case facts and legal elements. Through a dynamic knowledge generation mechanism, newly promulgated laws and the latest precedents are promptly parsed and added to the knowledge graph; if new types of cases arise in the judicial process, the knowledge layer will also expand corresponding new knowledge nodes through reasoning (such as new classifications of criminal methods).

[0038] The intelligent layer acts as an intelligent legal advisor in judicial assistance. When encountering a new case, the intelligent layer first receives input regarding the specific circumstances of the case and the intended judicial goal (e.g., the desired charge and sentencing range). Then, the intelligent layer retrieves relevant knowledge from the legal knowledge graph, including precedents of similar cases, relevant legal provisions and their interpretations, sentencing guidelines, etc. Using an introspective strategy, the intelligent layer uses past case-handling approaches to similar cases as templates, comparing them with the current case context. Through analogical reasoning, it infers the applicable legal provisions and reference judgments from existing similar cases in the knowledge graph. For example, the intelligent layer might identify similarities between the current case and a historical case in terms of modus operandi and circumstances, and thus, by analogy with the judgment in that case, suggest applying the same legal provisions and providing a similar sentencing range. Simultaneously, the intelligent layer incorporates the thought models of human judges, considering the special circumstances and social impact of the case (situational factors) to ensure the recommendations are more reasonable.

[0039] During the decision-making process, the expert layer also employs a "bug mechanism" to avoid blindly adhering to existing precedents and sticking to conventions. Specifically, when using analogical reasoning, the expert layer consciously introduces different perspectives, such as considering unconventional legal interpretations or citing cross-disciplinary legal principles when dealing with complex cases, thereby providing innovative defenses or judicial viewpoints. This appropriate "divergent thinking" helps judges comprehensively weigh the facts of a case and avoid overlooking potentially important factors.

[0040] Ultimately, the intelligence layer generates supplementary judgment suggestions for the case, including: possible charges and legal provisions, referenced past cases and their judgments, suggested sentencing ranges, and explanations of the reasons for these suggestions. The system presents this suggestion to judges or case handlers for decision-making reference. If the intelligence layer discovers gaps in key knowledge during analysis (e.g., the legal interpretation of a specific circumstance is missing from the knowledge graph), it will prompt the knowledge layer to conduct targeted supplementary learning through a feedback mechanism—such as retrieving relevant interpretations from a legal database or requesting experts to input new knowledge—and subsequently update the knowledge graph to make a more accurate assessment of the case. Through this process, this invention achieves a deep integration of knowledge and intelligence in the judicial field, significantly improving the efficiency of analyzing complex cases and the objectivity and fairness of judgments.

[0041] This embodiment describes a typical application scenario of the present invention in the field of education. Based on the system of the present invention, an intelligent personalized teaching system can be realized, providing students with learning plans and tutoring feedback tailored to their individual needs.

[0042] In this system, the information layer's data sources include textbooks and lesson plans, question banks, students' historical learning records, assessment scores, and real-time classroom feedback. The knowledge layer then constructs a knowledge graph for the education field. The nodes of this knowledge graph cover knowledge points, concepts, and skill modules for each subject, and record each student's level of mastery of related knowledge (which can be viewed as each student having a dynamic sub-knowledge graph). The relationships between nodes represent the dependencies between knowledge points (e.g., "calculus" depends on "basic function knowledge"), the knowledge module level to which a knowledge point belongs, and the mastery relationship between students and knowledge points (level of mastery, common mistakes, etc.). As teaching progresses, new test data and learning behavior data from students are updated in real-time to their knowledge graphs through a dynamic knowledge generation mechanism, reflecting their latest learning status.

[0043] In the personalized learning system, the intelligence layer acts as an intelligent tutor. When a student begins learning or a teacher develops a teaching plan, the intelligence layer receives the learning objectives (such as passing an exam or mastering a chapter) as input and extracts a subset of the student's relevant domain knowledge graph from the knowledge layer. The intelligence layer first employs an introspective meta-learning strategy, summarizing effective teaching strategy templates based on the learning trajectories of a large number of past students. For example, it discovers that certain types of students find that strengthening their knowledge of functions first when learning calculus yields better results, or that a certain teaching sequence is more effective in improving comprehension. These teaching patterns extracted from historical data are stored as strategy templates in the intelligence layer.

[0044] For each student, the intelligent layer combines their individual knowledge graph and learning style, using analogical reasoning to select the most suitable solution from the strategy template library for adjustment and optimization, generating a personalized learning plan. For example, if the intelligent layer finds that a student has a weak understanding of abstract concepts and prefers learning through examples, the system will recommend more intuitive cases and exercises, and select relevant basic concepts from the knowledge graph for review and reinforcement. If a student repeatedly answers incorrectly on a certain knowledge point, the intelligent layer will draw parallels with other similar students to infer that the student may have a gap in prior knowledge, thus adjusting the teaching sequence to address the corresponding foundational knowledge first. The entire plan will be presented to students and teachers in the form of a schedule or learning path, listing the learning content, sequence, key points, and accompanying exercises.

[0045] In the teaching process, the intelligent layer of this invention also embodies scene awareness and creative inspiration functions. For example, when students spend a long time thinking about a certain problem, the intelligent layer will provide some hints or unexpected guiding questions based on the BUG mechanism, prompting students to think about the problem from different angles. As another example, in science inquiry courses, the system may deliberately pose a question with paradoxes or minor errors, allowing students to try to correct them, thereby training their critical thinking and creativity. These intentionally designed "minor deviations" can stimulate students' learning interest and innovative awareness, which is more effective than the traditional method of simply imparting standard answers.

[0046] As teaching progresses, the intelligent layer continuously monitors students' learning outcomes. If it finds that a step in the teaching plan has not achieved the expected results (for example, a student still has not mastered a certain knowledge point), the intelligent layer will adjust its strategy in a timely manner (feedback to the knowledge layer requesting an update to the student's knowledge mastery model), such as increasing the number of practice questions for that knowledge point or changing the explanation method. Through this real-time feedback, the teaching plan can be dynamically optimized, truly achieving individualized instruction and addressing learning gaps.

[0047] This invention, when applied in educational settings, enables a student-centered, personalized learning process. The system not only imparts knowledge but also continuously adjusts its teaching strategies based on students' cognitive states and feedback, simulating an experienced tutor guiding each student. This significantly improves learning efficiency and effectiveness while reducing the burden on teachers in developing personalized teaching plans.

[0048] It should be noted that the method and system of this invention are also applicable to intelligent decision support in other fields, such as intelligent customer service (providing accurate and humanized answers to customers through knowledge graphs and intelligent decision-making) and medical diagnosis (assisting doctors in diagnosis through analogical reasoning using patient information and medical knowledge graphs). In different fields, the knowledge layer will construct a knowledge graph corresponding to the relevant industry, and the intelligence layer will perform domain-specific decision reasoning based on this graph. Those skilled in the art can make various modified applications based on the above principles, which will not be elaborated here.

Claims

1. A knowledge layer and wisdom layer learning method of a mesh DIKWP model, characterized in that, The method comprises: obtaining original data after semantic processing and constructing an initial knowledge graph; using a knowledge layer to semantically associate the data with existing knowledge and extend the knowledge through reasoning, and dynamically updating the knowledge graph; using a wisdom layer to perform decision reasoning on the updated knowledge graph to generate a solution for the current task; introducing a predetermined degree of random disturbance or cognitive bias in the decision reasoning process of the wisdom layer to promote the generation of creative solutions; and performing bidirectional feedback adjustment on the knowledge layer and the information layer according to the decision requirements and results of the wisdom layer to iteratively optimize the knowledge graph and the decision strategy.

2. The method of claim 1, wherein the dynamically updating a knowledge graph comprises: semantically associating new information with existing nodes in the knowledge graph; obtaining one or more candidate new knowledge nodes or relationships through knowledge reasoning; screening the candidate new knowledge based on a preset probability threshold to confirm the credibility thereof; and integrating the screened new knowledge into the knowledge graph to expand and update the knowledge base.

3. The method of claim 1, wherein the wisdom layer decisional inference comprises: abstracting and generalizing existing knowledge and historical experience to form general strategy templates (meta-knowledge); mapping a new problem to relevant nodes and subgraphs in the knowledge graph, calling a matched strategy template for reasoning, performing analogical reasoning based on similar cases and rules in the knowledge graph to generate a solution to the new problem, and evaluating and adjusting the solution in combination with the current context and intention to make the decision result meet the expected goal.

4. The method of claim 1, wherein the introduction of a predetermined degree of random disturbance or cognitive bias refers to: in the analogical reasoning or strategy selection process of the wisdom layer, randomly selecting an alternative similar case or introducing an imperfectly matched knowledge link to avoid the decision process from falling into a single path, thereby stimulating diverse solution ideas.

5. The method of claim 1, wherein the bidirectional feedback adjustment comprises: When the wisdom layer detects that there is a knowledge gap or contradiction related to decision-making in the knowledge graph, feedback information is generated and sent to the knowledge layer or the information layer to trigger new data acquisition and knowledge update for the knowledge gap; and when the knowledge layer structure changes or new knowledge is added, the wisdom layer is notified to update the decision strategy or parameters to adapt to the latest knowledge base state.

6. A knowledge layer and wisdom layer learning system of a mesh DIKWP model, characterized in that, The system comprises: a knowledge layer module for receiving data input from the information layer and performing semantic association, knowledge reasoning and probability filtering processing to construct and update a knowledge graph; a wisdom layer module for obtaining knowledge from the knowledge graph and performing decision reasoning to generate a decision solution through introspective meta-learning and analogical reasoning; the wisdom layer module is pre-provisioned with a mechanism allowing random bias to produce creative solutions during decision reasoning; and a feedback control unit for feeding back knowledge update requirements generated by the wisdom layer module during reasoning to the knowledge layer module to dynamically optimize the synergistic effect of the knowledge graph and the decision strategy.

7. The system of claim 6, wherein the knowledge layer module comprises: a semantic association submodule for semantically matching the data of the information layer with existing entities and relationships in the knowledge graph; a knowledge reasoning submodule for deducing new knowledge nodes or relationships based on the association results; a screening submodule for screening the candidate new knowledge based on a preset probability threshold to confirm the credibility thereof; and a feedback control unit for feeding back knowledge update requirements generated by the wisdom layer module during reasoning to the knowledge layer module to dynamically optimize the synergistic effect of the knowledge graph and the decision strategy. A probability evaluation submodule is configured to calculate the reliability of the new knowledge and filter out the knowledge meeting the confidence requirement; and a knowledge base updating submodule is configured to integrate the filtered new knowledge into the knowledge graph database and maintain the consistency of the knowledge base.

8. The system of claim 6, wherein the intelligence layer module comprises: A strategy abstraction submodule is configured to abstract a general strategy template based on historical decision-making experience; A problem mapping submodule is configured to convert a current problem to be solved into a target representation in the knowledge graph; An analogy reasoning submodule is configured to search for a similar knowledge mode or case to the problem to be solved on the knowledge graph and generate a preliminary solution; a bias introduction submodule is configured to apply random disturbance or select an alternative reasoning path in the decision-making process according to a preset rule; and a decision output submodule is configured to form a final decision scheme by comprehensively integrating the analogy reasoning result and current situational information and output the result.

Images