Mathematical semantic mapping method and system between intent and dikwps

Abstract

The application relates to artificial intelligence and knowledge engineering, and proposes a method and system for bidirectional semantic mapping between data-information-knowledge-wisdom-intention (DIKWP) multi-layers. By constructing a semantic tensor field and using category theory to define cross-layer functor mapping, combined with the "intention vectorization-knowledge graph embedding" algorithm, the high-level intention is projected to the low-level data / knowledge. And the reverse semantic path search is designed to back up the candidate intention from the observed data. The system sets up a semantic closed loop consistency detection and correction mechanism to ensure the equivalence of layer conversion and the reliability of reasoning. An executable interface is provided to output semantic sub-graph, explanation chain and score, which is suitable for explainable AI, controllable large model and cognitive robot scenes, and improves the relevance, safety and robustness.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence and knowledge engineering, and relates to a semantic mapping method and system based on a cognitive semantic model. Specifically, this invention addresses the inter-layer semantic transformation within a five-layer cognitive model (DIKWP model) comprising Data, Information, Knowledge, Wisdom, and Purpose, providing a mathematical semantic mapping method and system for mapping and transforming semantic structures between the "Purpose layer" and the layers "Data, Information, Knowledge, and Wisdom". Background Technology

[0002] The classic DIKW model describes the process of gradually extracting information from raw data, forming knowledge, and ultimately gaining wisdom. However, the traditional DIKW model lacks a characterization of the intention layer, failing to reflect the driving role of subjective goals on lower-level semantics in the cognitive process. Therefore, in recent years, the DIKWP model, which extends the "Purpose" layer to the top of the DIKW model, has been proposed. This is the "Data-Information-Knowledge-Wisdom-Intention" model, such as... Figure 1 As shown, the DIKWP model emphasizes an intention-driven cognitive transformation process, arguing that the acquisition and application of knowledge should serve specific goals or intentions. By introducing a Purpose layer into the cognitive framework, the DIKWP model aligns the goals of AI decision-making with human values, forming a closed-loop structure from underlying data to high-level intentions.

[0003] Existing large-scale models (such as large language models LLM) and knowledge graph systems are mostly data-driven or statistically driven, lacking explicit modeling of subjective intent, resulting in insufficient ability to handle personalized semantics and non-statistical problems. Due to the lack of semantic representation of subjective intent, AI in traditional data centers struggles to adjust its behavior in a timely manner under uncertain environments (when information is incomplete, inaccurate, or inconsistent). In contrast, the DIKWP model provides an intent-driven semantic network mechanism that can connect intent with underlying data, information, knowledge, and wisdom, thus potentially overcoming the aforementioned limitations of existing models. For example, a intent graph can be used to represent the relationships between an organization's goals, strategies, and activities. When combined with data graphs, information graphs, knowledge graphs, and wisdom graphs, it can form a DIKWP semantic graph, achieving higher-level cognition and decision support at the semantic level.

[0004] However, to fully leverage the advantages of the DIKWP model, a key challenge lies in establishing a rigorous mapping relationship between the highest-level intent semantics and the lower-level data / information / knowledge / wisdom semantics, enabling semantic content to be transformed between different levels without losing key meaning. Traditional semantic graphs and knowledge bases lack such inter-level transformation mechanisms, resulting in fragmented semantics at different levels and difficulty in achieving coherent reasoning. For example, in the absence of intent-driven mechanisms, the knowledge extracted from knowledge graphs struggles to determine which information is most useful for a specific goal; conversely, given a set of discrete data and information, it is difficult to infer its underlying intent or purpose. Some existing research proposes using tools such as category theory and formal logic to rigorously define and transform semantics in order to ensure semantic equivalence or consistency. Meanwhile, with the development of semantic embedding and knowledge graph embedding technologies, it has become possible to transform symbolic knowledge into computable vector representations. These advances provide a foundation for establishing semantic mappings mathematically and achieving a unified representation of multi-level semantics.

[0005] In summary, there is an urgent need for a new technical solution that can compress, abstract, interpret, and invert semantic information across the multiple layers of DIKWP based on semantic mathematics principles. Such a solution should accurately map high-level intents to demands for low-level data / knowledge, and also be able to infer possible intent hypotheses from specific data and facts, ultimately forming a semantic closed loop to verify the consistency of the overall logic. Furthermore, the solution should provide open interfaces for easy developer access, enabling transparent presentation of the reasoning process and meeting the requirements of interpretable, secure, and controllable AI applications. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a mathematical semantic mapping method and system for intention and DIKWP multi-layers. Through this invention, a bidirectional semantic mapping channel can be established in the DIKWP cognitive model, both top-down (Purpose→Data / Information / Knowledge / Wisdom) and bottom-up (Data→Information→Knowledge→Wisdom→Purpose), enabling the compression, abstraction, interpretation, and inversion of multi-layer semantics, thereby constructing an intention-driven closed-loop semantic reasoning framework.

[0007] The core solution of this invention includes the following mutually cooperating parts:

[0008] 1. Construction of the DIKWP Semantic Tensor Field Model: This invention introduces a "semantic tensor field" representation method for the semantic units at each layer of the DIKWP model. Tensor structures are used to represent the basic semantic units and their relationships in the data layer, information layer, knowledge layer, intelligence layer, and intent layer, capturing their high-dimensional structural features. Simultaneously, category theory formalization tools are used to define the mapping relationships between semantic objects at different levels. The principle of semantic isometry (equivalence) is followed to design transformation operators between tensors, ensuring that semantic content maintains semantic equivalence or distance without distortion during inter-layer transformations. This semantic tensor field model provides a unified mathematical foundation, enabling semantics at each layer to be represented and operated on within the same formal system, such as... Figure 2 As shown.

[0009] 2. "Intent Vectorization - Knowledge Graph Embedding" Transformation Algorithm: This invention proposes an algorithmic process for converting high-level intent semantics into low-level semantic representations. First, through an "intent aggregation" operation, the abstract intent elements of the Purpose layer are extracted and represented as a high-dimensional semantic vector (i.e., the intent is vectorized). Then, using knowledge graph embedding technology, this intent vector is mapped to an embedding vector space containing multi-layered semantic information. This embedding space integrates the graph structure of data, information, and knowledge layers, so that an intent vector corresponds to a set of related nodes and edges in the knowledge graph. Through this algorithm, abstract intents can be projected into operable vector forms, thereby recalling content related to that intent from the knowledge layer or data layer. For example, for a given intent vector, the system can find several semantically similar knowledge entities and data instances in the knowledge graph, which can be used as the basis for further reasoning or decision-making. The process is as follows: Figure 3 As shown.

[0010] 3. Reverse Mapping and Semantic Path Back-Inference Mechanism: This invention designs a bottom-up reverse reasoning method, enabling the system to infer possible high-level intent hypotheses from specific data and information. The basic process is as follows: Given a set of facts and their constraints at the data or information layer, the system first uses the causal relationships at the knowledge and wisdom layers to find the potential target set that these facts may serve. Then, it scores each potential Purpose hypothesis using a semantic evaluation function, selecting the best interpretation that best matches the observed data. This process is similar to backward planning or abductive reasoning in artificial intelligence—that is, finding hypothetical intents that can explain the observed facts through backward reasoning. To achieve efficient and reliable reverse mapping, this invention proposes a semantic path evaluation algorithm: searching the DIKWP semantic graph for all possible path chains from candidate Purpose nodes down to given data, and assigning scores based on factors such as path length, semantic relevance, and constraint matching, selecting the most reasonable explanation path chain as the output. See [link to relevant documentation]. Figure 4In this way, the system not only provides the inferred intent, but also offers a semantic link from intent to data as an explanation, improving the understandability and credibility of the reasoning results.

[0011] 4. Semantic Loop Consistency Detector: To ensure the logical consistency of the entire DIKWP link, this invention introduces a semantic loop detection mechanism. Specifically, after the forward mapping from the Purpose layer to the Data layer and the reverse mapping from the Data layer back to the Purpose layer are completed, the system will form a closed semantic loop between multiple layers. The semantic loop detector verifies this loop, judging whether the higher-level intent has sufficiently generated the expected information and knowledge, and whether the lower-level data and information can consistently support the back-reasoning of the original intent. The detector uses a series of consistency indicators, such as information completeness, consistency, and accuracy, to compare the output of each layer with the input of adjacent layers. Only when the semantic content of each layer remains highly consistent and conflict-free in the closed loop is the detector determined that the mapping is successful; otherwise, a correction mechanism will be triggered. Through this closed-loop verification, this invention can promptly detect semantic deviations or information losses in inter-layer transformations and feed them back to the system for adjustment (such as adjusting the intent vector or supplementing knowledge nodes), thereby achieving self-correction and semantic flow optimization. This mechanism ensures the reliability of system operation and avoids unreasonable reasoning results caused by layer breaks. The process is as follows: Figure 5 As shown.

[0012] 5. Executable Semantic Graph Operation Interface: This invention ultimately provides a set of semantic graph operation interfaces for developers, used to call the aforementioned mapping functions and construct a transparent and interpretable reasoning process. This interface encapsulates key functions from intent to multi-layered semantic representation and from data to intent inference, including but not limited to: "intent vectorization" function, semantic tensor transformation function, knowledge graph query and embedding function, reverse semantic path search function, and closed-loop consistency verification function. Developers can call these interfaces to input inputs from the original application scenario (e.g., user target requirements or observation data) into the system, obtain intermediate results of layer-by-layer semantic transformation, and the final reasoning output. The interface design follows the principle of human comprehensibility; for example, it supports outputting intermediate explanation nodes in the form of graph structures or logical chains, making the system's decision-making process transparent. This is particularly important for fields requiring interpretable AI (such as medical diagnosis and legal reasoning), where decision-makers can review each step of the reasoning chain provided by the AI. By providing an executable interface, the semantic mapping method of this invention can be easily integrated into existing AI systems or used to develop new applications, realizing cross-domain controllable large-scale model architectures, interpretable reasoning systems, and cognitive robots.

[0013] This invention achieves a technological breakthrough in semantic mapping between different layers of the DIKWP model through the above technical solutions, and has the following beneficial effects:

[0014] Intent-driven semantic transformation: This invention, guided by high-level intent, bridges the gap between data, intelligence, and intent, enabling AI systems to "know why." This overcomes the shortcomings of traditional knowledge graphs and large models, which are data-centric and lack goal orientation. The system can automatically filter and compress low-level information based on a given goal, effectively improving the relevance and specificity of reasoning.

[0015] The precision of semantic mathematics: The use of mathematical tools such as tensors and category theory ensures the rigor and equivalence of cross-level semantic transformations, avoiding inconsistencies and ambiguities in the semantic transformation process. This formalized semantic mapping mechanism improves the reliability of system reasoning, making each step of the transformation traceable and theoretically guaranteeing that the semantics are not distorted.

[0016] Bidirectional reasoning and explanation: It supports both deductive reasoning based on intent and abductive reasoning based on evidence, forming a complete semantic loop. This combined positive and negative mechanism enables the system not only to propose decisions but also to explain them: the forward link provides the steps to achieve the goal, while the reverse link verifies that these steps do indeed lead to the goal and can explain why these steps are reasonable. Therefore, this invention significantly improves the interpretability and credibility of AI decision-making.

[0017] Adaptive closed-loop correction: Through a semantic closed-loop detector, the system possesses the ability to automatically detect and correct inference biases. When the output of a certain layer is inconsistent with the overall goal, closed-loop feedback triggers the system to adjust the representation of the corresponding layer (e.g., recalculating the intent vector or supplementing new knowledge nodes) until the closed loop is consistent. This self-feedback mechanism improves the system's robustness to uncertainty and incompleteness, prevents the inference process from deviating from the expected path, and enhances the safety and reliability of the AI ​​system.

[0018] Universal Interface and Ease of Use: The provided semantic graph operation interface allows developers to invoke the functions of this invention without needing to understand the complex underlying mathematical details, integrating them into various applications. For example, in AI supervision scenarios, the interface of this invention can be used to audit the AI ​​decision-making process, tracing every semantic link from data to decision; in cognitive robots, the interface can empower robots to infer human intentions from perceptual data and plan actions accordingly. The modular design of the interface facilitates expansion and maintenance, demonstrating significant engineering practical value.

[0019] In summary, this invention, based on the DIKWP multi-layer cognitive model, proposes an intention-oriented mathematical semantic mapping method and system, solving the problem of the lack of intention-driven and inter-layer semantic transformation mechanisms in traditional semantic graphs. This invention has broad application prospects in fields such as controllable large-scale models, explainable AI, AI safety supervision, and cognitive robots, and can provide key support for building a new generation of intention-oriented and semantically transparent artificial intelligence systems. Its innovation lies in combining semantic mathematics with cognitive models, achieving a closed-loop connection from data to intention, injecting "goal awareness" and "semantic controllability" into artificial intelligence, representing a significant advancement in AI semantic understanding and reasoning technology. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the DIKWP five-layer semantic model and its closed-loop interaction structure in this invention;

[0021] Figure 2 This is a schematic diagram of the DIKWP multilayer semantic tensor field and cross-layer functor mapping in this invention;

[0022] Figure 3 This is a schematic diagram of the "intent vectorization - knowledge graph embedding" process in this invention;

[0023] Figure 4 This is the reverse semantic path search and scoring mechanism in this invention;

[0024] Figure 5 This is the semantic closed-loop consistency detection and correction process in this invention. Detailed Implementation

[0025] Example 1: Construction of the DIKWP Semantic Tensor Field Model. This example details how to use tensor fields to represent semantic content and the rules for tensor transformation between layers for the semantic units of each layer of the DIKWP model. First, we define the order and dimension of the semantic tensor for each layer: for example, the semantics of the data layer can be represented as a first- or second-order tensor (vector or matrix), containing the original data and its attribute dimensions; the semantic units of the information layer, due to the introduction of relations and meanings, can be represented as higher-order tensors, associating data with context; the semantic units of the knowledge layer (such as concepts and rules) can be represented as higher-order tensors or tensor networks to capture the complex relationships between concepts; the wisdom layer involves the integration of decision-making and rules, and can use tensor operators to combine multiple knowledge tensors; the intention layer can be seen as an abstraction of the desired goal, which can be represented by a high-dimensional vector representing its direction and weight distribution in the semantic space. Next, using functor theory in category theory, the semantic tensors of each layer are formalized as objects on the category, and cross-layer mappings are defined as functor mappings between categories. These mappings adhere to the principle of semantic equivalence, ensuring that if two underlying elements correspond semantically, they retain their correspondence after mapping to higher levels. This category mapping ensures the reversibility and consistency of inter-layer transformations through natural transformations. For example, if two elements in the data layer correspond to the same information unit through mapping in the information layer, the original two data elements should be retrievable during reverse mapping (unless information loss is expected and labeled). In practical implementations, these mapping functions can be established using tensor operation libraries, transforming category theory principles into executable algorithms (e.g., defining a set of linear / nonlinear transformation matrices to map lower-level tensor products to upper-level tensors). Notably, to maintain semantic equivalence, constraints are introduced when designing tensor transformations, such as maintaining the invariance of a specific metric (semantic similarity). When a tensor transformation violates semantic equivalence constraints, the system records the error for subsequent closed-loop correction. Through the above methods, a unified semantic tensor field model is constructed, laying the foundation for subsequent intent vectorization and reverse reasoning.

[0026] Example 2: "Intent Vectorization - Knowledge Graph Embedding" Algorithm Flow. This example details how to map a user-given high-level intent into a vector representation and retrieve relevant semantic content from a knowledge graph. Assume a user inputs a target description (e.g., "Optimize energy consumption production plan"). The system first performs semantic parsing on the natural language target, extracting core intent keywords and constraints (such as "energy consumption," "production plan," and "optimize"). Next, the system encodes the parsed intent elements into a high-dimensional vector using a pre-trained intent semantic vector model (e.g., a deep learning-based encoder). The various dimensions of this vector may correspond to different semantic latent factors, such as time preference, cost focus, and environmental weight, which are automatically learned during the training process. Next, the system performs "intent vector semantic retrieval" in its knowledge graph database. Specifically, each node (representing a knowledge concept or instance) and important relationships in the knowledge graph are pre-embedded and trained, possessing a corresponding vector representation. The system calculates the sum of all node vectors and intent vectors in the knowledge graph. The similarity scores (e.g., cosine similarity) are used to select the series of knowledge nodes with the highest rankings. As content related to the intent, for example, for the intent to optimize energy, the search results might include knowledge nodes such as "energy-saving production processes," "energy monitoring datasets," and "smart factory optimization rules." Subsequently, the system semantically expands these initially relevant nodes: it finds directly connected upstream and downstream nodes through graph relationships, expanding the potentially related information set. Simultaneously, combined with the reasoning rules of the intelligent layer, it filters out nodes irrelevant to the current scenario (e.g., knowledge not belonging to the manufacturing field). The final output is a "semantic subgraph" for the input intent, containing entities and relationships closely related to the intent at the data, information, and knowledge layers. The nodes and edges of this subgraph will serve as knowledge support for subsequent reasoning or decision-making. It's worth noting that if the user has further constraints (e.g., the cost must be below a certain threshold), the system will prioritize satisfying these constraints during retrieval, and then sort according to semantic similarity. Through this algorithm, the abstract target intent is successfully transformed into a concrete, multi-layered semantically interwoven set of content, essentially finding "landed" data and knowledge basis for the target.

[0027] Example 3: Reverse Semantic Reasoning Mechanism. Suppose we have a set of observed factual data and need to infer the underlying intentions.

[0028] In this embodiment, we demonstrate how to achieve this through a reverse mapping mechanism. Scenario example: A cognitive robot observes several phenomena in a factory workshop: machine A is frequently starting and stopping, energy consumption indicators are abnormally high, and the operator has changed the production formula. Based on this discrete information, the robot needs to infer the goals that the manager might be currently focusing on. First, the system converts these observation data into representations at the information and knowledge layers: extracting information such as "frequent machine start-ups and shutdowns," "high energy consumption," and "changes in production formula," and identifying related knowledge concepts (such as equipment maintenance, energy conservation and emission reduction, production optimization, etc.) in the knowledge base. Next, the system initiates a reverse reasoning algorithm. It uses these knowledge concepts as clues to search in the intent space. The intent space can be understood as a series of candidate goals at the Purpose layer, such as "reducing energy costs," "improving production efficiency," and "equipment maintenance plans." The system uses a semantic path evaluation algorithm to assess whether each candidate objective can explain the observed information. For example, for the candidate intent "reduce energy consumption costs," the system finds a path in the DIKWP graph that starts from this intent, passes through smart rules ("reducing energy consumption requires optimizing recipes and equipment operation"), knowledge nodes ("energy consumption monitoring"), and reaches data nodes (machine start-up and shutdown counts, energy consumption data). The system calculates the degree of matching between this path and the observed data (whether all key nodes on the path can be found in the observations, whether the inference chain conforms to domain common sense, etc.), and assigns an evaluation result to the intent. Figure 1 The system scored highly on the explanation. Looking at another candidate intent, "improving productivity," while formula changes and machine start-up / shutdown might also be related, "increased energy consumption" is undesirable (negatively correlated) under the goal of improving efficiency, hence the lower explanation score. After comparison, the system concluded that "reducing energy costs" was the most likely underlying driving intent and output this hypothetical intent. Simultaneously, the system provided a semantic explanation chain supporting this hypothesis (i.e., the path from intent to data mentioned above) as the basis for inference. If multiple highly-scoring hypothetical intents exist (e.g., simultaneously "reducing energy consumption" and "equipment maintenance"), the system can further combine context or interact with humans to confirm the true goal. In application, this mechanism enables AI to reason about intentions not explicitly stated by humans, possessing a certain degree of "mental reasoning" ability. For example, in medical diagnostic assistants, AI can infer the doctor's diagnostic focus based on scattered symptom data; in intelligent assistants, it can guess the user's implicit needs based on a series of user actions, providing possibilities for more proactive services.

[0029] Example 4: Semantic Loop Detection and Correction. In this example, we explain how the system uses a loop detector to ensure consistency between forward and reverse mapping results and performs self-correction. Continuing the scenarios of Examples 2 and 3, when the system obtains a semantic subgraph based on intent and infers an intent hypothesis based on data, we want to verify whether the two match to form a loop. The specific process is as follows: The system compares the semantic subgraph obtained by forward reasoning with the interpretation path obtained by reverse reasoning, comparing the key semantic nodes and relationships. On the one hand, it checks whether the hypothetical intent obtained by reverse reasoning is the same as the initial input intent (in the case of a clear input intent). If the two are inconsistent, it indicates that the system has deviated in the forward or reverse process. At this time, the loop detector records the inconsistent intent pair and triggers the adjustment process. For example, the knowledge subgraph obtained by forward retrieval may have missed some information, causing the reverse reasoning to go astray; or the reverse reasoning may have inferred a different intent, indicating that the user's true intent may not match the expression. On the other hand, even if the intent is consistent, the system also needs to check the loop consistency of the intermediate layer. For example, does a knowledge node K extracted in the forward flow appear in the reverse explanation chain? Does each reasoning rule in the reverse chain have a corresponding basis in the forward process? For each DIKWP layer, corresponding semantic conservation checkpoints are defined: the data layer checks the difference between the output data and the input data; the information layer checks whether the extracted information sufficiently supports the reasoning of the knowledge layer; the knowledge layer checks whether the knowledge used leads to the decision of the wisdom layer; and the wisdom layer's decision correctly achieves the intended goal. Any mismatch will be detected by the detector. For example, if the original decision of the wisdom layer fails to improve the energy consumption index under simulation, the closed loop is not closed. After detecting the problem, the detector will send the corresponding signal to the relevant module for correction: it may be necessary to re-retrieve supplementary knowledge, adjust the intention vector weights, or update the decision scheme of the wisdom layer. The entire closed-loop detection and correction process can be iterated multiple times until all checkpoints are passed or the preset convergence condition is reached. Through this embodiment, the system demonstrates autonomous adaptability and can ensure consistency when performing complex tasks. For example, in multi-sensor decision-making for autonomous vehicles, closed-loop detection ensures consistency in the "perception → understanding → planning → action" chain, preventing perceptual errors from causing erroneous actions. In artificial consciousness models, closed-loop mechanisms are considered crucial to prevent AI from exhibiting self-inconsistent behavior. In summary, this embodiment demonstrates how the present invention rigorously controls semantic quality in a closed-loop manner, giving the system's reasoning process an inherent correctness constraint.

[0030] Example 5: Executable Interface Integration and Application Example. This example illustrates how to encapsulate the above methods into an interface and integrate them into a specific application, as well as the resulting effects. We take an "AI Legal Assistant System" as an example: This system needs to infer the possible intent of a case's ruling based on legal knowledge and case data, and provide an explanation. Developers utilize the API interface provided by this invention to connect the legal assistant system to the semantic mapping module. First, the developer calls the `PurposeVector = IntentEncoder(user_goal_description)` interface to encode the goal described by the lawyer or user (e.g., "to secure a reduced sentence for the client") into an intent vector. Next, `semantic_subgraph = GetSemanticSubgraph(PurposeVector)` is called to obtain a knowledge subgraph related to this goal, which includes node relationships such as relevant legal provisions, past case knowledge, and case evidence. Then, the system enters the automatic reasoning stage: when new evidence data is input, `predicted_intent = InverseReasoning(evidence_data)` is called, and the system attempts to infer the possible intent of the prosecution / defense, such as "proving the suspect's subjective malice" or "proving the client's negligence was not intentional." Simultaneously, the system calls `explanation_path = GetExplanationPath(predicted_intent, evidence_data)` to obtain a reasoning chain from the assumed intent to the evidence. Next, it calls `consistency_score = CheckClosedLoop(PurposeVector, predicted_intent, semantic_subgraph, explanation_path)` to perform a closed-loop consistency check, determining whether the user's initial goal aligns with the system's predicted goal and whether the knowledge subgraph matches the explanation chain. If the score is below a threshold, the interface returns information about the inconsistency, such as "missing key evidence explaining the transaction intent," allowing developers to decide whether to request more information from the user or adjust system parameters. Finally, once closed-loop consistency is achieved, the developer calls `GenerateReport(semantic_subgraph, explanation_path)`, and the system generates a detailed report based on the semantic subgraph and reasoning chain. This report includes: an analysis of the target (which legal knowledge and supporting evidence were used), possible conclusions (e.g., recommended legal strategies), and an explanation of the reasoning process (step-by-step explanation of how the conclusion was derived from the evidence). This report can be reviewed by lawyers, increasing their trust in the AI's results.Through this interface integration, the AI ​​legal assistant achieves a deep semantic connection from user intent to case data, proactively identifying insufficient evidence and explaining the basis of reasoning, thus making complex legal reasoning transparent and verifiable. Similarly, in other applications, such as medical diagnostic AI, the interface of this invention can be used to establish a semantic loop between the doctor's hypothetical diagnosis and patient data, providing interpretable diagnostic suggestions; large-scale model security monitoring can use the interface to identify in real time whether the model's output deviates from the intended intent and provide corrective information. These all demonstrate the practical value and diverse applicability of the interface of this invention.

[0031] In summary, the present invention, through the above embodiments, elaborates in detail the implementation details of the mathematical semantic mapping method and system between intent-oriented and DIKWP multi-layers.

[0032] By leveraging key technologies such as semantic tensor fields, intent vectorization and graph embedding, reverse reasoning, and loop closure detection, this invention successfully integrates intent-driven and multi-layer semantic transformation, significantly improving the interpretability, controllability, and semantic understanding depth of artificial intelligence systems.

[0033] It achieves cross-level intelligent reasoning while ensuring semantic rigor and consistency, which is of great significance for building a new generation of trustworthy artificial intelligence.

[0034] The technical solutions described in each embodiment can be combined with each other to meet the needs of different scenarios. All technical transformations or equivalent substitutions within the scope of protection of this invention are included.

Claims

1. A mathematical semantic mapping method for intent and DIKWP (Data D—Information I—Knowledge K—Wisdom W—Intent P) multi-layers, characterized in that, include: S1: Construct a semantic tensor field covering five layers: D / I / K / W / P, perform tensor quantization representation on semantic units of each layer, and define cross-layer functor mapping and natural transformation according to category theory, and apply semantic isometry / equivalence constraints. S2: Aggregate and encode the semantics of the intent layer to obtain the intent vector p, and map p to the knowledge graph embedding space to retrieve relevant nodes and relationships, generating a semantic subgraph associated with the target; S3: Perform a bottom-up reverse semantic path search on the input data / information, enumerate paths from candidate intents to data nodes in the DIKWP semantic graph and calculate scores to determine the interpretation path; S4: Perform semantic closed-loop consistency detection based on the forward mapping result and the reverse interpretation path. If there is inconsistency, trigger correction to adjust the intent vector, supplement knowledge nodes, or correct the inter-layer transformation. S5: Output the semantic subgraph, explanation chain, and consistency score through the executable interface, and generate a report containing target analysis, conclusions, and reasoning explanations.

2. The method according to claim 1, wherein, The cross-layer tensor transformation in S1 maintains the preset semantic metrics (including semantic similarity) unchanged, and uses natural transformation to ensure the reversibility and consistency of forward and reverse mappings.

3. The method according to claim 1 or 2, wherein, S2's intent encoding uses a deep learning-based semantic encoder to obtain a high-dimensional intent vector p.

4. The method according to any one of claims 1 to 3, wherein, Knowledge graph retrieval sorts nodes based on the similarity (including cosine similarity) between p and the embedded nodes in the graph, and selects the top-k relevant nodes. When generating the semantic subgraph, semantic expansion is performed, and nodes and relationships that are irrelevant to the current scene are filtered in combination with the rules of the wisdom layer.

5. The method according to any one of claims 1 to 4, wherein, S3's path scoring comprehensively considers path length, semantic relevance, and constraint matching to select the most reasonable explanatory chain.

6. The method according to any one of claims 1 to 5, wherein, S4's closed-loop consistency check sets semantic conservation checkpoints at each DIKWP layer: the data layer checks data consistency, the information layer checks support sufficiency, the knowledge layer checks reasoning accessibility, and the wisdom layer checks decision achievement. Corrections for closed-loop inconsistencies include at least: recalculating intent vector weights, supplementing or replacing knowledge nodes, adjusting the wisdom layer decision rules, or modifying cross-layer transformation operators.

7. The method according to any one of claims 1 to 6, wherein, The executable interface encapsulates functions such as intent vectorization, semantic tensor transformation, knowledge graph query and embedding, reverse semantic path search and closed-loop consistency verification, and outputs intermediate explanation nodes and evaluation reports in the form of graph structure or logical chain.

8. A semantic mapping system for implementing the method of any one of claims 1 to 7, characterized in that, include: The intent encoding module, semantic tensor mapping module, reverse reasoning module, closed-loop consistency detection and correction module, and interface module are respectively used to generate intent vector p, construct DIKWP semantic tensor field and perform cross-layer mapping, perform semantic path search and scoring, complete closed-loop consistency detection and correction, and output semantic subgraph / interpretation chain / consistency score and report.

9. The system according to claim 8, wherein, The knowledge graph stores vector embeddings of nodes and relationships, and the module generates the semantic subgraph using similarity retrieval and semantic expansion.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the computer to perform the method of any one of claims 1 to 7.

Images