A propositionized triple specific reasoning parameter structure and reasoning method based on science and engineering knowledge point reasoning network

Abstract

This invention discloses a propositional triplet reasoning parameter structure and reasoning method based on a reasoning network of science and engineering knowledge points, relating to the fields of knowledge graphs, graph theory reasoning, artificial intelligence, and educational informatization. The method extracts semantic units to construct propositional knowledge point nodes, assigning them abstraction, verifiability, and historical state parameters, and generating node weights by combining mastery, preference, and task parameters. Logical edges with logical type numbers and natural language interpretations are established between nodes, and a Fog aggregation structure and a Bridge bridging structure are constructed to form a parameter propagation chain from the knowledge point layer, logical edge layer, Fog layer to the Bridge layer. A two-stage path search is performed based on the lowest common superior fog to generate a personalized reasoning chain; and structural health detection and collaborative parsing with external language models are supported. This invention achieves fine-grained control and personalized adaptation of reasoning paths, improving the traceability and auditability of reasoning.

Description

Technical Field

[0001] This invention relates to the fields of knowledge graphs, graph theory reasoning, artificial intelligence, educational informatization, and STEM knowledge engineering technology, and particularly to a propositional knowledge point parameter modeling method, a logical connection parameterization method, a path reasoning method under a multi-layer fog aggregation structure, and a personalized interpretation method for STEM knowledge networks. Background Technology

[0002] Existing knowledge graph systems typically organize knowledge using a "entity-relationship-entity" triple approach. However, in STEM scenarios, many systems still suffer from issues such as inconsistent semantic granularity of nodes, unclear logical meaning of inference edges, lack of quantitative expression of node credibility, difficulty in incorporating historically invalidated knowledge into a unified structure, and difficulty in incorporating individual user differences into path computation. Especially in scenarios such as teaching, research assistance, and interdisciplinary knowledge retrieval, relying solely on knowledge point nodes described in natural language is insufficient to directly support rigorous graph theory reasoning, and it is also difficult to guarantee the applicability of reasoning paths across different users, tasks, and knowledge regions. On the other hand, some systems have narrowed the search scope and improved reasoning efficiency in large-scale networks through knowledge aggregation, hierarchical partitioning, or abstracting intermediate structures. However, a unified, propagable, and updatable parameter structure is still lacking among node parameters, edge parameters, aggregated entity parameters, and personalized preferences. Particularly in STEM knowledge, the same proposition, in addition to natural language fields such as name, definition, formula, and description, also possesses attributes that directly affect the value of reasoning, such as the level of expression, strength of evidence, and historical status. Without parameterized modeling of these attributes, graph search can only rely on static edge weights or simple topological relationships, failing to achieve fine-grained control over knowledge credibility, abstraction span, historical validity, prior learning relationships, and user preferences. Furthermore, while large language models can serve as interactive interfaces to assist users in expressing needs and understanding results, without external structured knowledge networks and path constraints, they are still prone to producing illusions, omitting key conditions, or outputting unauditable conclusions in multi-step reasoning in STEM fields. Therefore, it is necessary to propose a new, rigorous propositional triplet-based specific reasoning parameter structure, establishing a unified parameter transmission mechanism at the knowledge point layer, logical edge layer, aggregation layer, and path layer to support verifiable, personalized, and updatable reasoning methods for STEM scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide a rigorous propositional triplet-based reasoning parameter structure and reasoning method for reasoning networks based on science and engineering knowledge points. This method is used to construct a multi-layer parameter system based on knowledge point triplets, fog aggregation structures, and bridging structures. This system consists of basic knowledge point parameters, personalized knowledge point parameters, logical type parameters, fog influence factors, bridging weights, and path scoring parameters, thereby achieving fine control, personalized adaptation, and structured interpretation of the reasoning path of the knowledge network.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: First, the input raw science and engineering knowledge data is parsed, and semantic units that can stand alone, be asserted, and participate in reasoning are extracted as knowledge point nodes.

[0005] Each knowledge point node includes at least a name and a basic description, and may further include at least one of the following: definition, formula, source, subject classification, applicable conditions, example description, application scenario, and historical description.

[0006] For complex narratives that contain multiple levels of expression, they are broken down into multiple single-proposition nodes, and relationships are established through logical connections to ensure that each node has a clear and stable semantic role in reasoning.

[0007] Secondly, this invention sets basic three-axis parameters for each knowledge point node, which include at least the abstraction parameter A, the verification parameter E, and the historical state parameter H.

[0008] The abstraction parameter A describes the level of representation of a knowledge point, and includes at least: A0 Phenomenon layer, representing phenomena, facts, and data patterns that can be directly observed, measured, or experimentally obtained; A1 represents the theoretical level of science and engineering, indicating laws, models, theories, mechanisms, or explanatory propositions; A2 Engineering layer represents equipment, processes, procedures, designs, operating steps, engineering criteria, or performance constraints; A3 Mathematical layer represents definitions, theorems, corollaries, proof structures, axiomatic statements, and their equivalent transformations.

[0009] The veracity parameter E describes the form and strength of evidence supporting a knowledge point, and includes at least: E0 Unverified E1 Personal Experience E2 Small sample observation E3 theoretical prediction, E4 Laboratory Replication Validation E5 multiple experimental verifications, E6 Engineering or Field Application Verification E7 Standardization Knowledge E8 consensus-level knowledge, E9 Theoretical Limit E10 Mathematical Theorem Level.

[0010] The historical state parameter H is used to describe the effectiveness and evolutionary position of knowledge points in the contemporary knowledge system, and includes at least: H0 remains valid in modern times. H1 was replaced by the superordinate theory. H2 has been falsified or contradicts contemporary systems.

[0011] The three parameters mentioned above are orthogonal to each other and together constitute the basic parameter space of the knowledge point system, and can be dynamically updated during the introduction of new evidence, theoretical updates, or knowledge base maintenance.

[0012] Furthermore, this invention sets up a parameter mapping module to map the abstraction parameter A, the verification parameter E, and the historical state parameter H into basic weights of knowledge points.

[0013] The mapping method can be a preset table lookup, a rule function, a piecewise function, an interval mapping, a vector mapping, or a combination thereof.

[0014] Preferably, the system generates a basic weight node_base_weight for each knowledge point based on a preset function, which represents the basic reasoning priority of that knowledge point without considering specific user preferences.

[0015] Furthermore, this invention sets personalized parameters for each knowledge point.

[0016] The personalized parameters include at least the mastery factor and the preference factor. It may also include at least one of the following: task objective parameters, learning stage parameters, application scenario parameters, and domain priority parameters.

[0017] The mastery level parameter is used to characterize a user's current mastery level of a knowledge point, thus playing a role in automatically selecting the learning starting point, determining intermediate transition nodes, and controlling the path length. The preference level parameter is used to characterize the user's preferences for specific domains, specific levels of abstraction, specific verifiability, specific historical states, specific knowledge styles, specific interpretation methods, etc.

[0018] The system combines the basic weights of knowledge points with personalized parameters to generate the final weight of each knowledge point, node_final_weight. Preferably, the final weight can be generated using multiplication, weighted multiplication, a combination of multiplication and addition, or a preset fusion function.

[0019] Furthermore, this invention establishes logical connections between knowledge points. Each logical connection includes at least a starting knowledge point, an ending knowledge point, a logical type number, and a natural language explanation field.

[0020] The explanation is used to describe the reasons, conditions, and semantic boundaries for establishing the logical connection.

[0021] The logical type is a finite set, including at least one or more of the following: 01 Equivalent 02 Derivation 03 Includes or belongs to 04 Prerequisites or necessary foundations 05 Analogy 06 Assumptions or assumptions 07 References or inspiration 08. Negation, falsification, or contradiction 09 Historical equivalence or superior substitution 10. Summarize or conclude 11. Application or practice.

[0022] Different logical types have preset directionality, semantic constraints, and default parameters.

[0023] The system configures a default parameter `relation_type_default_weight` for each logical type and allows further setting of the logical type preference parameter `logic_preference_factor(type)`.

[0024] The logical edge may also include at least one of the following: source reliability parameter, explanation completeness parameter, condition matching parameter, local region parameter, and task mode parameter.

[0025] The system uses the above parameters to determine the final edge weight of the logical edge, edge_final_weight.

[0026] Furthermore, this invention introduces the fog polymerization structure Fog.

[0027] Fog is an abstract collection of several knowledge points, including at least fog_ID, fog_name, fog_level, description information, a set of sub-knowledge points and the number of sub-knowledge points, and further includes the fog influence factor fog_influence_factor obtained by aggregating the parameters of its subordinate knowledge points.

[0028] The fog influence factor is generated by the abstraction parameter A, the verification parameter E, the historical state parameter H of the knowledge points inside the Fog, and their corresponding mapping results according to one or more of the following: mean, weighted mean, maximum value, quantile, frequency distribution statistics, entropy index, or preset aggregation function. It is used to characterize the stability, authority, coverage, or reasoning priority of the Fog in the current knowledge network.

[0029] Furthermore, Fog can be configured with personalized parameters, including Fog-level mastery parameters, Fog-level preference parameters, and Fog-level local preference parameters for different logic types.

[0030] The Fog-level local preference parameters are used to weight and correct logical connections between knowledge points within a Fog or across Fogs.

[0031] Furthermore, this invention generates a bridge structure (Bridge) by statistically analyzing the logical connections between knowledge points of different Fogs.

[0032] Each Bridge must include at least the starting Fog, the ending Fog, the number of logical edges across the Fog, the distribution of logical types, and the bridge weights.

[0033] The system generates the final bridge weight (bridge_final_weight) based on the edge_final_weight of the cross-Fog logical edges, the distribution of logical types, the number of edges, the fog influence factor, and the user's local preference parameters.

[0034] Thus, this invention forms a parameter propagation chain from the knowledge point layer, logic edge layer, Fog layer to the Bridge layer.

[0035] Furthermore, the present invention performs reasoning based on the above parameter structure.

[0036] When the system receives the starting knowledge point and the target knowledge point, or only the target knowledge point, the system first filters the candidate starting point set based on the user's level of mastery, the difficulty of the target, the necessary prerequisite relationships, the final weight of the knowledge point, and the task objective. In the case where only the target knowledge point is given, the system automatically determines one or more nodes that are suitable as the starting point of the learning chain, so as to avoid starting reasoning from content that the user has not yet mastered and cannot understand, and also to avoid starting reasoning from content that the user has already fully mastered and that would lead to an excessively long chain.

[0037] Subsequently, the system determines the lowest common upper-level fog LCAF of the Fog to which the starting knowledge point and the target knowledge point belong, and performs Fog-level path search first, and then knowledge point-level path search within the area defined by the LCAF. In the Fog layer, the system constructs macroscopic paths based on bridge_final_weight, fog_influence_factor, Fog transition cost, and user local preferences; in the knowledge point layer, the system constructs microscopic logical chains based on node_final_weight, edge_final_weight, logical type constraints, explanation completeness, condition matching degree, and historical state constraints.

[0038] The system ultimately combines the abstract path of the Fog layer with the specific logical chain of the knowledge point layer to output personalized reasoning results.

[0039] Furthermore, the present invention may also include a structural health check module.

[0040] The module is used to detect directed cycles in the 04 preceding or necessary basic subgraphs, conflict clusters formed by 08 negation, falsification or contradictory relationships, cycles in the 09 historical substitution chain, the proportion of isolated points, and the distribution of strong and weak logical edges. Based on the inspection results, it performs explanation review, logical edge weight reduction, local parameter recalculation, Bridge local recalculation or manual review.

[0041] Furthermore, the present invention also includes a user interaction and explanation module that collaborates with an external language model.

[0042] The interpretation module includes at least one of a template interpretation module, a rule interpretation module, a retrieval enhancement module, and an external language model module.

[0043] The external language model module does not directly replace the structured reasoning process in this invention, but rather serves as a natural language input / output interface and a parameter-assisted generation interface connected to the system of this invention.

[0044] During the input phase, users can describe their learning needs, knowledge background, comprehension difficulties, target content, application scenarios, or preferred directions to an external language model using natural language. The external language model module performs semantic parsing on user input, extracting or inferring at least one or more of the following information: Target knowledge points, candidate starting knowledge points, user mastery level parameters, user preference level parameters, task objective parameters, learning stage parameters, and explanation style parameters; The information is then converted into a structured input that can be processed by this invention and sent to the parameterized inference module.

[0045] Subsequently, based on the structured input, the present invention performs parametric reasoning on the knowledge point layer, Fog layer, and Bridge layer to generate structured logic chains, path scoring results, and related parameter information.

[0046] During the output phase, the external language model module receives the structured logic chain, knowledge point description, logical type sequence, Fog path and corresponding parameters generated by the present invention, and generates natural language explanations, analogies, case extensions, example deductions, difficulty adjustment content or interactive question-and-answer responses that conform to the user's knowledge level and expression preferences under the constraints of the structured logic chain and parameter constraints.

[0047] Since the input and output of the external language model module are both constrained by the structured logic chain and parameters generated by this invention, it can improve human-computer interaction capabilities and reduce the illusion risk and uncertainty of results in multi-step reasoning scenarios in science and engineering while maintaining the traceability, verifiability and auditability of the logic chain. Beneficial effects

[0048] Compared with the prior art, the present invention has at least the following beneficial effects: First, the present invention elevates knowledge points from simple natural language entries to rigorous propositional nodes with abstraction parameter A, verification parameter E, and historical state parameter H, thereby enabling knowledge points to participate in unified parameter calculation and graph theory reasoning.

[0049] Second, by combining basic knowledge point parameters with user-personalized parameters, this invention enables personalized control over the path starting point, path length, abstraction span, and content preferences, making the reasoning chain generated by the system more adaptable to the comprehension abilities and task requirements of different users.

[0050] Third, by unifying the design of logical types, natural language interpretation fields, logical edge parameters, and logical type preference parameters, this invention transforms logical edges from simple semantic annotations into computable, adjustable, and auditable reasoning units.

[0051] Fourth, by constructing fog influence factors and Bridge weights, this invention enables the parameters of the knowledge point layer to propagate to the aggregation layer and cross-regional layer, thereby enhancing the fine-grained basis for path selection while maintaining reasoning efficiency.

[0052] Fifth, this invention can simultaneously support global logical preference control, local area logical preference control, and knowledge point layer personalized control, making it suitable for various scenarios such as teaching recommendation, scientific research assistance, interdisciplinary analysis, knowledge visualization, and structured retrieval.

[0053] Sixth, by placing the external interpretation module after the structured reasoning result, this invention reduces the risk of illusion in multi-step reasoning in science and engineering using pure language models, and improves the consistency and auditability of the result interpretation. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the overall process of parametric knowledge network construction, parametric reasoning, and interpretation in this invention. Detailed Implementation

[0055] like Figure 1 As shown, the present invention includes a parameterized knowledge network construction stage and a parameterized reasoning and interpretation stage.

[0056] Figure 1Even when modules marked "optional" are not enabled, the system can still perform parametric inference based on preset parameters, user-inputted start or end points, or existing structured request data. The invention will be further described below with reference to specific embodiments, but the invention is not limited to these embodiments.

[0057] Example 1: Knowledge Point Construction.

[0058] The system receives raw STEM knowledge data from textbooks, papers, manually entered forms, database records, or other sources, extracts content that can be independently stated, asserted, and included in the reasoning chain, and generates propositional knowledge point nodes.

[0059] Each knowledge point node should include at least a name and a basic description, and may further include a definition, formula, source, subject classification, applicable conditions, examples, and application scenarios.

[0060] For complex narratives that simultaneously include observational facts, theoretical explanations, and engineering practices, the system breaks them down into multiple knowledge point nodes and connects them through corresponding logical connections.

[0061] For example, the original content "under certain conditions, it is observed that current and voltage are proportional, and Ohm's law is derived from this, which can be further used for resistive circuit design" can be broken down into phenomenon layer nodes, theoretical layer nodes, and engineering layer nodes, and each node is assigned a corresponding abstraction parameter.

[0062] Example 2: Mapping of basic parameters for knowledge points.

[0063] The system assigns an abstraction parameter A, a verification parameter E, and a historical state parameter H to each knowledge point, and generates the basic weight node_base_weight of the knowledge point through preset mapping rules.

[0064] In one implementation, the system uses a combination of table lookup and rule functions to map different levels of abstraction, different levels of evidence, and different historical states to different basic parameter values.

[0065] Knowledge points with a historical state of H2 can be significantly downweighted in the general learning mode, but remain accessible in the error tracking mode or the history of science mode.

[0066] Example 3: Personalized parameter generation.

[0067] The system maintains parameters for different users' knowledge mastery and preference levels.

[0068] If the target knowledge point is "divergence theorem", and the user has mastered "vector field" and "surface integral" but not "flux definition", the system can automatically set "flux definition" as the preferred candidate starting point, without backtracking to earlier elementary vector concepts or starting directly from more difficult theorem-level nodes.

[0069] At this point, the system generates the final weight of the knowledge point, node_final_weight, based on the basic weight of the knowledge point and the user parameters.

[0070] In one implementation, node_final_weight = node_base_weight × mastery_factor × preference_factor × task_factor.

[0071] Example 4: Construction of logical edges.

[0072] The system establishes logical edges between knowledge points, and each logical edge includes at least a start point, an end point, a logical type number, and an explanation.

[0073] Taking "the concept of limit is a necessary foundation for the definition of integral" as an example, the system establishes a logical boundary of "the concept of limit - 04 - the definition of integral"; Taking "the formula for uniformly accelerated linear motion can be derived from Newton's second law" as an example, the system establishes the logical boundary "Newton's second law - 02 - formula for uniformly accelerated linear motion"; Taking "experimental observation results support the formation of a certain law" as an example, the system establishes a logical boundary of "experimental results - 10 - theoretical law".

[0074] The system further generates the final edge weight (edge_final_weight) of the logical edge based on the default parameters of the logical type, the completeness of the explanation, and the reliability of the source.

[0075] Example 5: Fog and Bridge construction.

[0076] The system automatically categorizes knowledge points into different levels of Fog based on subject classification hierarchy and custom aggregation rules.

[0077] For each Fog, the system calculates the distribution of A, E, and H of its internal knowledge points and generates the Fog_influence_factor accordingly.

[0078] Subsequently, the system counts the number, type distribution, and final edge weights of cross-regional logical edges between different Fogs, and generates the Bridge and its bridge_final_weight.

[0079] Example 6: Parametric Reasoning.

[0080] After receiving the starting point and the target point, the system first determines the lowest common upper-level fog LCAF. Within the area defined by the LCAF, the system constructs the Fog layer path based on the Bridge weight and the fog influence factor, and then constructs the knowledge point layer path based on the final weight of the knowledge point and the final edge weight of the logical edge, and combines the two into the final inference chain.

[0081] If multiple paths with similar total weights exist, the system can further disambiguate based on the proportion of strong logical edges, the number of Fog transitions, path length, or consistency with user preferences.

[0082] Example 7: Structural health check.

[0083] The system periodically checks whether there are cycles in the subgraph (04), whether the relationships form conflict clusters (08), and whether the substitution chains form unreasonable loops (09). Based on the detection results, it triggers explanation review, parameter correction, or local recalculation.

[0084] Example 8: Input-output interpretation in collaboration with external language models.

[0085] In this embodiment, in addition to accepting structured parameters directly input by the user, the system also allows the user to express their needs in natural language through an external language model.

[0086] Users can describe to the external language model the target content they want to learn, their existing foundation, difficulties in understanding, preferred style, desired depth, or application scenario.

[0087] After parsing the user input, the external language model generates structured request data compatible with the present invention. The structured request data includes at least one or more of the following: target knowledge points, candidate starting points, mastery level parameters, preference level parameters, task target parameters, and explanation style parameters.

[0088] After receiving the structured request data, the system calls the parameterized reasoning module of the present invention to perform path calculations in the knowledge point layer, Fog layer and Bridge layer to generate a structured logic chain.

[0089] After obtaining the final logical chain, the system sends the knowledge point sequence, logical type sequence, Fog path, Bridge information and corresponding parameters to the external language model module again.

[0090] The external language model module uses the structured logic chain as external constraints to perform natural language interpretation, example deduction, analogy explanation, emphasis, difficulty adjustment, or interactive question and answer on the results.

[0091] For example, when a user simply states, "I want to understand why we should learn divergence before Maxwell's equations, and I don't really like pure mathematical proofs," the external language model can first convert this need into target knowledge points, mastery level parameters, and preference parameters, and then the present invention can generate the corresponding logical chain. Subsequently, the external language model, based on this logical chain and combining knowledge point descriptions and logical relationships, provides step-by-step explanations to the user, without replacing the structured reasoning of this invention.

[0092] In this embodiment, the external language model is responsible for parsing user needs and generating parameters on the input side, and for expressing structured results in natural language and expanding interactive features on the output side. Meanwhile, the construction of the logical chain, path constraints, weight calculation, and inference selection in the middle are still completed by this invention.

[0093] Therefore, without sacrificing structural rigor, the system's natural interactivity and personalized adaptability can be improved.

[0094] In addition, the method of the present invention can be run in a computer device, which includes, but is not limited to, a processor, a memory, an input / output interface, a communication interface, and a display device. The processor executes program instructions stored in the memory to implement all or part of the steps of the method described in the various embodiments of the present invention. The computer device can be a personal computer, a workstation, a server, a mobile terminal, an edge computing device, a cluster node, or other electronic devices with data processing capabilities.

[0095] This invention can also be deployed on local servers, private clouds, public clouds, hybrid clouds, or other network computing environments to support distributed database storage, cross-terminal access, multi-user collaborative input, parameter synchronization updates, and inference service calls. Each knowledge point node, logical connection, Fog structure, Bridge structure, parameter mapping rules, user-personalized parameters, and inference results can be stored centrally or distributed; they can be processed by a single computing node or by multiple computing nodes collaboratively.

[0096] This invention can be implemented using dedicated hardware, such as ASICs, FPGAs, DSPs, edge computing chips, or other dedicated logic devices; it can also be implemented using software executed by general-purpose hardware, such as program code running on CPUs, GPUs, or other general-purpose processors; or it can be implemented using a combination of hardware and software. The functional modules can be physically integrated into a single unit or deployed in a distributed manner and interact through networks, buses, or inter-process communication. Merging, splitting, reorganizing, or changing the deployment location of the modules does not affect the substantive content of this invention.

[0097] When this invention is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium includes, but is not limited to, read-only memory, random access memory, disk, optical disk, solid-state drive, USB flash drive, portable hard drive, flash memory, or other media capable of storing program code. When the program code is executed by a processor, the processor performs all or part of the steps of the methods described in the various embodiments of this invention, including but not limited to: parsing science and engineering knowledge data, constructing propositional knowledge points, labeling abstraction parameter A, labeling verification parameter E, labeling historical state parameter H, generating basic weights for knowledge points, generating personalized parameters, constructing logical connections, writing the explanation field, calculating fog impact factors, aggregating bridge weights, determining candidate starting points, calculating LCAF, searching paths at the Fog layer, searching paths at the knowledge point layer, generating personalized inference chains, checking structural health, outputting results, and collaboratively calling the explanation module.

[0098] In the flowcharts or block diagrams shown in the accompanying drawings, blocks may represent a functional module, program segment, algorithm step, or part of code, which contains executable instructions for performing a specified logical function. In some alternative embodiments, the functions marked in the blocks may not be executed in the order shown in the drawings; two or more steps may also be executed in parallel, interleaved, or in different orders under different conditions. The accompanying drawings are schematic drawings intended to illustrate the core ideas and key relationships of the invention and should not be construed as limiting the scope of protection of the invention.

[0099] In this invention, the specific values ​​for the basic knowledge point parameters, user-personalized parameters, logical type parameters, fog influence factors, Bridge weights, and path scoring parameters can be obtained through table lookup mapping, rule functions, piecewise functions, normalization, regularization, heuristic functions, weighted averages, nonlinear transformations, or combinations thereof. The numerical range, aggregation method, update frequency, caching strategy, and triggering conditions of these parameters can all be adjusted according to specific application scenarios without altering the fundamental idea of ​​this invention: controlling the reasoning path of a science and engineering knowledge network through a multi-layered parameter structure.

[0100] In this invention, the external explanation module can be a template explanation module, a rule explanation module, a retrieval enhancement module, a language model module, or a combination thereof. In one embodiment, the external language model module can receive user natural language requests on the input side and generate structured request data such as target knowledge points, mastery level parameters, preference level parameters, task parameters, and explanation style parameters. On the output side, it can receive the structured logical chain, knowledge point description, logical type sequence, Fog path, and related parameter information generated by this invention, and generate natural language explanations, case expansions, example deductions, difficulty adjustment content, or interactive question-and-answer responses under the constraints of the structured results. The access method, calling order, interaction rounds, and constraint strength of the external explanation module can all be adjusted without affecting the essence of this invention, which generates the core logical chain by the structured reasoning module.

[0101] To avoid ambiguity, the terminology used in this document is explained as follows: 1) "Including," "containing," and their variations are non-exclusive terms, indicating that the presence of the stated element does not exclude the existence of other unlisted elements. 2) "And / or" indicates that the listed objects can exist individually or simultaneously. 3) "First," "Second," "Third," etc., are only used to distinguish similar objects and do not indicate a specific order or importance. 4) "Connection," "Coupling," and "Association," in a non-conflicting context, can indicate a direct relationship or an indirect relationship established through intermediate nodes, Fog structures, Bridge structures, cache modules, or other data structures. 5) The terms "basic weight of knowledge points," "final weight of knowledge points," "final weight of logical edges," "fog influence factor," "final weight of Bridge," and "path cost" used in this document are all parameter expressions used for inference control. Their specific forms can be scalars, vectors, matrices, interval values, probability values, score values, or combinations thereof. Unless otherwise specified, a smaller weight indicates a lower inference cost and a better path, while a larger weight can, in another implementation, indicate a higher priority after an equivalent transformation. 6) "Lowest Common Ancestor (LCAF)" refers to the lowest-level common ancestor aggregation node in the Fog hierarchy that can simultaneously contain both the starting knowledge point and the target knowledge point. 7) "Prerequisite Necessary Foundation Relationship" indicates a relationship where one knowledge point is necessary for the understanding, derivation, application, or correct use of another knowledge point; "Historical Substitution Relationship" indicates that an old proposition is replaced by a new proposition in a more general or rigorous system; "Negation / Falsification / Contradiction Relationship" indicates a logical conflict, empirical conflict, or falsification relationship between two propositions. 8) "External Language Model Module" refers to a natural language processing model accessed through a network, local interface, process call, plugin mechanism, or other means. Whether it is deployed locally, in the cloud, or at the edge does not constitute a limitation of this invention. 9) Expressions such as "if," "when," "in response to," "when detected," and "when determined" can all represent conditional trigger semantics.

[0102] Unless the context otherwise requires, the embodiments, parameter forms, function forms, update strategies, caching methods, parallel disambiguation rules, path scoring rules, and interpretation output methods described herein are all preferred examples for implementing the present invention, and not exhaustive limitations thereof. Those skilled in the art can make various modifications, substitutions, combinations, or equivalent transformations to the structural forms, module divisions, parameter values, processing flows, interaction methods, and display styles in the above embodiments without departing from the spirit and substance of the present invention, and all such modifications, substitutions, combinations, or equivalent transformations should fall within the protection scope of the present invention.

[0103] In summary, any modifications, equivalent substitutions, improvements, extensions, or combinations made within the spirit and principles of this invention should be included within the scope of protection of this invention. The scope of protection of this invention shall be determined by the claims, and the specification and drawings are used to interpret the claims.

Claims

1. A specific reasoning method for propositional triples based on a reasoning network of science and engineering knowledge points, characterized in that, Includes the following steps: S1. Analyze the input STEM knowledge data, extract semantic units that can stand alone, be asserted, and participate in reasoning, and construct propositional knowledge point nodes; each knowledge point node includes at least a name and basic description, and is assigned an abstraction parameter A, a verifiability parameter E, and a historical state parameter H; S2. According to a preset parameter mapping rule, map the abstraction parameter A, verifiability parameter E, and historical state parameter H to the knowledge point base weight node_base_weight; and obtain the user's personalized parameters for the target user, and generate the final weight node_final_weight of the knowledge point based on the user's personalized parameters; S3. Establish logical connections between knowledge point nodes, each logical connection including at least a starting knowledge point, an ending knowledge point, a logical type number, and a natural language explanation field explanation; and determine the final edge weight edge_final_weight of the logical edge according to the pre-configured default parameters of the logical type, the final weight of the knowledge point, and optional additional parameters; S4. According to the subject classification hierarchy and / or custom aggregation rules, organize multiple knowledge point nodes into a Fog structure; and based on the abstraction parameter A, verifiability parameter E, and historical state of the knowledge points within the Fog... The parameter H or its corresponding mapping result determines the fog influence factor fog_influence_factor; and a bridge structure Bridge is generated by statistically analyzing the number, type, and edge weight of logical connections between knowledge points in different Fogs, and the final weight of the bridge is determined by bridge_final_weight; S5, when the starting knowledge point and the target knowledge point are received, or only the target knowledge point is received, a set of candidate starting points is determined based on the user's mastery level, target difficulty, necessary prerequisite relationships, final weight of knowledge points, and task objectives; S6, the lowest common upper-level fog LCAF of the Fog to which the starting knowledge point and the target knowledge point belong is determined, and within the area defined by the LCAF, Fog layer path search is first performed based on the final weight of the bridge bridge_final_weight and the fog influence factor fog_influence_factor, and then knowledge point layer path search is performed based on the final weight of the knowledge point node_final_weight, the final weight of the logical edge edge_final_weight, and logical type constraints, generating a personalized reasoning chain composed of abstract paths in the Fog layer and specific logical chains in the knowledge point layer; S7, the personalized reasoning chain is output.

2. The method according to claim 1, characterized in that, The knowledge point nodes also include at least one of the following: definition, formula, source, subject classification, applicable conditions, example illustration, application scenario, and historical explanation; for compound narratives that contain multiple levels of expression, they are split into multiple single proposition nodes and relationships are established through logical connections.

3. The method according to claim 1, characterized in that, The abstraction parameter A includes at least one or more of the following levels: A0 Phenomenon level, A1 Science and theory level, A2 Engineering level, A3 Mathematical level; the verifiability parameter E includes at least one or more of the following levels: E0 Unverified, E1 Personal experience, E2 Small sample observation, E3 Theoretical prediction, E4 Laboratory replication verification, E5 Multiple experimental verification, E6 Engineering or field application verification, E7 Standardized knowledge, E8 Consensus-level knowledge, E9 Theoretical limit, E10 Mathematical theorem level; the historical state parameter H includes at least one or more of the following levels: H0 Still valid in the present, H1 Replaced by higher-level theory, H2 Falsified or contradictory to the contemporary system.

4. The method according to claim 1, characterized in that, The user personalization parameters in step S2 include at least the mastery parameter and the preference parameter, as well as at least one of the following: task goal parameter, learning stage parameter, application scenario parameter, and domain priority parameter.

5. The method according to claim 1, characterized in that, The final weight of the knowledge point in step S2, node_final_weight, is determined by the basic weight of the knowledge point, node_base_weight, and the user's personalized parameters.

6. The method according to claim 1, characterized in that, The logical type number mentioned in step S3 is taken from a preset finite logical type set, which includes at least one or more of the following: 01 equivalence, 02 derivation, 03 inclusion or membership, 04 prerequisite or necessary basis, 05 analogy, 06 hypothesis or assumption, 07 reference or inspiration, 08 negation or falsification or contradiction, 09 historical equivalence or superior substitution, 10 induction or summary, 11 application or practice.

7. The method according to claim 1 or 6, characterized in that, Each logical connection corresponds to at least one natural language explanation field, which explains the reason, condition and semantic boundary for establishing the logical connection; multiple non-conflicting logical connection types are allowed between the same pair of knowledge points.

8. The method according to claim 1, characterized in that, The optional additional parameters in step S3 include at least one of the following: source reliability parameter, explanation completeness parameter, condition matching parameter, local region parameter, and task mode parameter; and also include a logic type preference parameter logic_preference_factor(type) set for different logic types, which is used to adjust the priority of the corresponding logic edges of different logic types in the personalized reasoning process.

9. The method according to claim 1, characterized in that, The fog influence factor fog_influence_factor mentioned in step S4 is generated by the abstraction parameter A, the confirmation parameter E, the historical state parameter H of the Fog internal knowledge points and their corresponding mapping results in one or more ways, such as mean, weighted mean, maximum value, quantile, frequency distribution statistics, entropy index or preset aggregation function.

10. The method according to claim 1, characterized in that, Each Fog also includes a Fog-level mastery parameter, a Fog-level preference parameter, and Fog-level local preference parameters for different logic types; the Fog-level local preference parameters are used to weight and correct the logical connections between knowledge points within a Fog or across Fog logic connections.

11. The method according to claim 1 or 10, characterized in that, The final bridge weight in step S4 is generated based on the final weight of the cross-Fog logical edge, the distribution of logical types, the number of edges, the fog influence factor (fog_influence_factor), and the user's local preference parameters.

12. The method according to claim 1, characterized in that, When the abstraction parameter A, the verification parameter E, or the historical state parameter H of a knowledge point are updated, the system recalculates the knowledge point base weight node_base_weight corresponding to that knowledge point, and triggers the update of the fog influence factor fog_influence_factor of its Fog and the bridge final weight bridge_final_weight associated with it.

13. The method according to claim 1, characterized in that, In step S5, when only the target knowledge point is received, the system automatically determines one or more candidate starting point sets based on the user's level of mastery, target difficulty, necessary prerequisite relationships, and the final weight of the knowledge point, in order to avoid starting reasoning from content that the user has not yet mastered and cannot understand, or starting reasoning from content that the user has fully mastered and that would lead to an excessively long chain of reasoning.

14. The method according to claim 1, characterized in that, Step S6 involves Fog layer path search and knowledge point layer path search using one or more of the following graph theory methods: Dijkstra's algorithm, A* algorithm, Floyd algorithm, Bellman-Ford algorithm, breadth-first search, depth-first search, heuristic search algorithm, or equivalent graph theory search methods; and when multiple candidate paths exist, they are disambiguated in parallel based on the proportion of strong logical edges, the number of Fog transitions, the path length, or the degree of consistency with user preferences.

15. The method according to claim 1, characterized in that, It also includes a structural health check step; the structural health check includes at least one or more of the following: performing directed cycle detection on the 04 pre- or necessary basic subgraph, statistically analyzing the conflict clusters formed by the 08 negation, falsification or contradiction relationships, performing cycle checks on the 09 historical substitution chains, statistically analyzing the proportion of isolated points, and monitoring the distribution of strong and weak logical edges; and performing explanation review, logical edge weight reduction, local parameter recalculation, Bridge local recalculation or manual review based on the check results.

16. The method according to claim 1, characterized in that, It also includes user interaction and explanation steps in collaboration with an external language model; wherein, in the input stage, the external language model module receives the user's natural language requirements and generates at least one structured request data among the target knowledge points, candidate starting knowledge points, user mastery level parameters, user preference level parameters, task target parameters, learning stage parameters, and explanation style parameters; in the output stage, it receives the structured logical chain, knowledge point description, logical type sequence, Fog path, and corresponding parameters generated by this method, and generates natural language explanations, analogies, case extensions, example deductions, difficulty adjustment content, or interactive question-and-answer responses under the constraints of the structured logical chain and parameters.

17. A propositional triplet-based concrete reasoning system based on a reasoning network of science and engineering knowledge points, characterized in that, include: The knowledge point construction module is used to parse science and engineering knowledge data and construct propositional knowledge point nodes; The parameter mapping module is used to assign abstraction parameter A, verification parameter E, and historical state parameter H to knowledge point nodes, and generate the basic weight of knowledge point node_base_weight; The personalized parameter module is used to obtain user-personalized parameters and form the final weight of knowledge points, node_final_weight; The logical connection construction module is used to establish logical connections between knowledge point nodes with logical type numbers and natural language explanation fields, and to determine the final edge weight edge_final_weight; the Fog / Bridge construction module is used to generate Fog structures, determine the fog influence factor fog_influence_factor, count cross-Fog logical connections, and generate the final weight bridge_final_weight of the bridge. The candidate starting point determination module is used to determine the set of candidate starting points based on the user's level of mastery, the difficulty of the target, the necessary prerequisite relationships, the final weight of knowledge points, and the task objective. A two-stage reasoning module is used to first perform a Fog layer path search based on the lowest common upper-level fog (LCAF), and then perform a knowledge point layer path search to generate a personalized reasoning chain; an output module is used to output the personalized reasoning chain; wherein, the system is configured to perform the method described in any one of claims 1 to 16.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 16.

Images