An orthopedics knowledge graph construction method and system

Abstract

This invention relates to the field of knowledge graph construction technology, and discloses a method and system for constructing an orthopedic knowledge graph, comprising: firstly, receiving an orthopedic entity co-occurrence frequency matrix and a biomechanical topological distance matrix, generating an initial semantic matrix through normalization, and constructing a biomechanical-semantic fusion tensor by combining the reciprocal of the topological distance. Next, calculating the hierarchical overlap covariance and overlap ratio factor, and performing orthogonal projection on the fusion tensor to separate redundant autocorrelation features. Subsequently, generating a Riemann curvature compensation phase angle to correct spatial distortion by calculating the semantic drift exponent and the global curvature ratio. Finally, extracting the winding discretization error of the bidirectional compensation phase angle, and using the structural integrity suppression scalar to correct the orthogonal projection tensor to generate the final edge weight matrix.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, specifically to a method and system for constructing an orthopedic knowledge graph. Background Technology

[0002] In the field of medical artificial intelligence and clinical decision support systems, orthopedic knowledge graphs are the core underlying support for intelligent assisted diagnosis and treatment, surgical planning, and accurate retrieval of medical literature. Currently, common orthopedic knowledge graph construction techniques mainly rely on natural language processing, analyzing the co-occurrence frequency of text entities in electronic medical records or extracting semantic features using deep learning models to establish association weights between entities. However, existing general semantic modeling methods reveal significant limitations when applied to orthopedics, a field with strong anatomical constraints. First, existing technologies often focus only on statistical associations within medical record texts, lacking consideration for the objective mechanical transmission logic of the human skeleton. This results in a graph that may be highly related at the semantic level but completely detached from physical reality in its biomechanical spatial topology, failing to reflect the actual support and transmission relationships between bones. Second, the anatomical structure of the human skeleton has complex hierarchical nesting attributes; for example, "lower limb" and "femur" have a high degree of overlap in both physical and semantic spaces. Traditional algorithms struggle to effectively remove redundant collinear features when processing such hierarchical entities with inclusion relationships, easily leading to singularity collapse in the feature space and consequently causing confusion in the graph's logic. Furthermore, when attempting to orthogonally decouple these overlapping features to separate independent structures, changes in the spatial manifold structure can cause a shift in the originally closely related semantic centers, resulting in semantic drift and disrupting the original distribution of clinical knowledge. In generating global edges in the graph, the failure to address phase periodicity in the topological network often leads to discontinuous abrupt changes or breaks in edge weights due to phase entanglement, affecting the smoothness and logical integrity of the graph weights. Existing technical solutions still lack systematic mechanisms for balancing semantic accuracy with physical structural logic, handling deep hierarchical overlaps, and correcting spatial geometric distortions.

[0003] In summary, there is a need for an orthopedic knowledge graph construction scheme that can deeply integrate biomechanical topological constraints, effectively handle hierarchical redundancy, and achieve smooth correction of spatial features. Summary of the Invention

[0004] This invention provides a method and system for constructing an orthopedic knowledge graph, which helps to solve the problems mentioned in the background art.

[0005] This invention provides the following technical solution: a method for constructing an orthopedic knowledge graph, comprising: Receive the co-occurrence frequency matrix of orthopedic entities, calculate the total frequency of each entity, perform normalization processing on it, and generate an initial normalized semantic matrix; Receive the biomechanical topological distance matrix, calculate the reciprocal of the topological distance, multiply the reciprocal term by the initial normalized semantic matrix, and generate the biomechanical-semantic fusion tensor; Calculate the inner product of node vectors in the biomechanics-semantic fusion tensor, calculate the hierarchical overlap covariance, and generate the overlap ratio factor by combining the node's own covariance and topological distance. Extract the biomechanical-semantic fusion tensor element of the node itself, multiply the element by the overlap ratio factor, subtract the product from the biomechanical-semantic fusion tensor, and generate the orthogonalized projection tensor; By comparing the biomechanics-semantic fusion tensor with the orthogonalized projection tensor, the sum of squares of the differences in eigenvectors is calculated, and the square root operation is performed on the sum of squares to generate the semantic drift index. Calculate the global curvature ratio of the single-node drift index, calculate the normalized reference term of the orthogonalized projection tensor, and generate the curvature compensation phase angle by combining the global curvature ratio with the inverse cosine mapping angle. Collect the bidirectional compensation phase angles between nodes, calculate the difference between the bidirectional compensation phase angles, perform sine function mapping on the difference, and generate phase winding discrete error; Calculate the square of the phase winding discrete error, construct a structural integrity suppression scalar using a fractional structure with a constant denominator, multiply the orthogonalized projection tensor by this scalar to generate the final edge weight matrix.

[0006] Optionally, the receiving orthopedic entity co-occurrence frequency matrix, calculating the total frequency of each entity, performing normalization processing on it, and generating an initial normalized semantic matrix includes: The system initially acquires two sets of basic objective data matrices during runtime: The first group is an entity co-occurrence frequency matrix statistically analyzed from the orthopedic clinical electronic medical record text set. The rows and columns represent the pre-identified orthopedic entity nodes, and the matrix elements are the objective number of times the entities co-occur within the text window. The second group is the biomechanical topological distance matrix, where the matrix elements are the number of hops of the shortest mechanical conduction path between two anatomical skeletal nodes. Both sets of matrices have dimensions corresponding to the total number of entities; Calculate the sum of squares of the co-occurrence frequencies of entities between the first node and all traversed nodes, and the sum of squares of the co-occurrence frequencies of entities between the second node and all traversed nodes, respectively. Multiply the sums of two squares, and then perform a square root operation on the product to generate a normalized denominator term; The co-occurrence frequency of entities between the first node and the second node is used as the numerator; Divide the numerator by the normalized denominator to generate the initial normalized semantic similarity between the first node and the second node. Traverse all nodes and combine them to generate an initial normalized semantic matrix.

[0007] Optionally, the step of receiving the biomechanical topological distance matrix, calculating the reciprocal of the topological distance, multiplying the reciprocal term by the initial normalized semantic matrix, and generating a biomechanical-semantic fusion tensor includes: Extract the biomechanical topological distance between the first node and the second node; Calculate the reciprocal of the biomechanical topological distance; Adding the reciprocal to a constant one generates the distance attenuation gain term. Multiply the initial normalized semantic similarity between the first node and the second node by the distance decay gain term to generate the biomechanical-semantic fusion tensor element between the first node and the second node; Iterate through all nodes to generate a biomechanical-semantic fusion tensor.

[0008] Optionally, the step of calculating the inner product of node vectors in the biomechanical-semantic fusion tensor, calculating the hierarchical overlap covariance, and combining the node's own covariance with the topological distance to generate an overlap ratio factor includes: Extract the biomechanical-semantic fusion tensor elements between the first node and the traversed nodes, and the biomechanical-semantic fusion tensor elements between the second node and the traversed node; Multiply the two together and sum the product of all traversed nodes to generate the hierarchical overlap covariance between the first and second nodes. Extract the covariance of the first node and the covariance of the second node; Add the covariance of the first node itself, the covariance of the second node itself, and the biomechanical topological distance between the first and second nodes to generate a proportional denominator term; Divide the hierarchical overlap covariance by the proportional denominator to generate the overlap ratio factor between the first node and the second node.

[0009] Optionally, the step of extracting the biomechanical-semantic fusion tensor element of the node itself, multiplying this element by the overlap ratio factor, and subtracting the product from the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor includes: Extract the biomechanical-semantic fusion tensor elements between the second node and itself; Multiply this element by the overlap ratio factor to generate redundant feature components; Redundant feature components are subtracted from the biomechanical-semantic fusion tensor between the first node and the second node to generate the orthogonalized projection tensor between the first node and the second node; Iterate through all nodes and combine them to generate an orthogonalized projection tensor.

[0010] Optionally, the step of comparing the biomechanical-semantic fusion tensor with the orthogonalized projection tensor, calculating the sum of squared differences in the eigenvectors, and performing a square root operation on the sum of squared differences to generate a semantic drift index includes: Calculate the difference between the biomechanical-semantic fusion tensor element between the first node and the traversed nodes, and the orthogonalized projection tensor element at the corresponding position; Calculate the square of the difference; Sum the squared differences of all traversed nodes to generate a sum of squares. Perform a square root operation on the sum of squares to generate the semantic drift index of the first node.

[0011] Optionally, the step of calculating the global curvature ratio of the single-node drift index, calculating the normalized reference term of the orthogonalized projection tensor, and generating a curvature compensation phase angle by combining the global curvature ratio with the inverse cosine mapping angle includes: Sum the semantic drift indices of all nodes to generate the global drift sum; Divide the semantic drift index of the first node by the global drift sum to generate the global curvature ratio of the first node. Calculate the sum of squares of the orthogonalized projection tensor elements between the first node and all traversed nodes; Perform a square root operation on the sum of squares to generate a normalized baseline term; Divide the orthogonalized projection tensor elements between the first and second nodes by the normalized reference term; Perform an inverse cosine function operation on the division result to generate a reference angle; Multiply the reference angle by the global curvature ratio to generate the curvature compensation phase angle from the first node to the second node.

[0012] Optionally, the bidirectional compensation phase angle between the collection nodes is calculated, the difference between the bidirectional compensation phase angles is calculated, and a sine function mapping is performed on the difference to generate a phase winding discrete error, including: Extract the curvature compensation phase angle from the first node to the second node, and the curvature compensation phase angle from the second node to the first node; Calculate the difference between the two curvature-compensated phase angles; Perform a sine function calculation on the difference to generate the phase winding discrete error between the first node and the second node.

[0013] Optionally, the calculation of the square of the phase winding discrete error involves constructing a structural integrity suppression scalar using a fractional structure with a constant denominator, and multiplying the orthogonalized projection tensor by this scalar to generate the final edge weight matrix, including: Calculate the square of the phase winding discretization error between the first node and the second node; Add a constant 1 to the squared result to generate a suppressed denominator term; Divide the constant by the suppression denominator to generate the structural integrity suppression scalar between the first node and the second node; Multiply the orthogonalized projection tensor elements between the first and second nodes with the structural integrity suppression scalar to generate the final edge weight matrix elements between the first and second nodes. The final edge weight matrix output directly serves as the foundation for the underlying network topology of the orthopedic knowledge graph.

[0014] Optionally, a system for implementing the orthopedic knowledge graph construction method includes: Data receiving and fusion analysis module: used to receive the co-occurrence frequency matrix of orthopedic entities and the biomechanical topological distance matrix, calculate the normalized initial semantic matrix, and generate the biomechanical-semantic fusion tensor by combining the inverse term of the topological distance; The core module for overlap elimination and orthogonal decoupling is used to calculate the hierarchical overlap covariance and overlap ratio factor of the node vectors in the biomechanical-semantic fusion tensor, and to perform semantic-topological orthogonalization projection on the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor that separates redundancy. Spatial distortion and phase compensation module: used to calculate the sum of squared feature differences before and after orthogonal projection to obtain the semantic drift index, extract the global curvature ratio, and generate the Riemann curvature compensation phase angle through inverse cosine mapping; Error smoothing and topology weight generation module: used to perform sine mapping on the bidirectional compensated phase angle difference to extract the phase winding discrete error, thereby constructing a structural integrity suppression scalar and correcting the orthogonalized projection tensor to output the final edge weight matrix.

[0015] The present invention has the following beneficial effects: 1. By directly introducing the objective biomechanical physical topology of the human body into the semantic feature space of traditional medical electronic medical record text, this technical solution provides a novel perspective of physical reality constraints for the construction of orthopedic knowledge graphs. In the specific environment of orthopedic assisted diagnosis or surgical planning in medical artificial intelligence, traditional graph construction methods often have a fatal blind spot: they can only mechanically rely on natural language processing technology to count which orthopedic terms frequently appear together in medical records, but they cannot understand the actual anatomical and biomechanical structure of the human skeleton. This makes the system highly susceptible to being misled by doctors' subjective writing habits, and when faced with medical terms such as "lower limb" and "femur" which naturally have a hierarchical inclusion relationship, traditional algorithms will produce severe feature overlap due to their high binding in the text, leading to singular collapse of the spatial topology. If these overlapping features are forcibly separated within the traditional framework, the manifold structure of the original data will be destroyed, causing severe semantic drift (i.e., deformation) of originally closely related concepts, and even leading to discontinuous logical breaks and abrupt changes in the connections in the final network structure. To thoroughly overcome these challenges, this solution departs from simplistic textual statistics. Instead, it deeply integrates the actual number of hops in the mechanical transmission path as a physical gain factor with pure semantic features. Then, using overlapping covariance and orthogonal projection techniques, it cleanly removes redundant features caused by hierarchical inclusion, much like removing ghosting. For the unavoidable spatial distortion and periodic errors after this process, the solution creatively maps them to Riemann curvature compensation phase angles and utilizes the symmetry of the sine function to extract a smooth error suppression scalar for correction. In this specific orthopedic medical application environment, the most significant and unique benefit of this solution lies in its successful endowment of the underlying computer system with genuine "physical common sense." It ensures that the final output edge weight matrix is ​​no longer just a bunch of probability numbers from textual statistics, but strictly follows the objective laws of human biomechanics transmission and the tightness of real spatial distances, effectively avoiding false logical connections that violate common sense in real anatomy. At the same time, it perfectly repairs spatial deformation and bidirectional logical conflicts without losslessly stripping away conceptual redundancy, eliminating falsehoods and retaining truth, providing a highly realistic, smooth and continuous, and extremely rigorous and reliable underlying intelligent architecture support for orthopedic specialty systems that require absolutely rigorous scientific evidence.

[0016] 2. By obtaining the initial co-occurrence frequency of orthopedic entities in clinical medical records, the interaction distribution features of different nodes in the global text space are extracted. The sum of squares of the co-occurrence frequencies of the target node and all other traversed nodes is calculated, and the square root of the extracted sum of squares is used as the normalization benchmark term of the global feature scale. This transforms the absolute co-occurrence frequency into a relative pure semantic similarity state. This approach can completely eliminate the huge scale difference in the absolute occurrence frequency of different orthopedic terms, effectively preventing the high-frequency word hegemony phenomenon caused by the excessive frequency of common general medical terms. It also avoids high-frequency words from masking key but low-frequency core specialist terms in the feature space, forcing all entity features to be compressed to a unified and fair scale for measurement. This not only provides a unified benchmark data foundation for subsequent high-dimensional matrix fusion, but also ensures that subsequent algorithms focus entirely on the pure co-occurrence direction and association purity when measuring entity correlation, without being affected by the absolute number of words caused by the length of the medical record text or the doctor's personal writing habits. This significantly improves the accuracy and objectivity of the underlying semantic matrix in capturing real medical relevance.

[0017] 3. By extracting the shortest mechanical transmission path hop count between objectively existing human skeletal structures, the reciprocal structure of the biomechanical topological distance is calculated. Combined with a constant bias, a physical gain amplifier with attenuation properties is constructed. This gain term is then directly applied to the previously calculated normalized semantic similarity to generate a biomechanical-semantic fusion tensor with physical distance attenuation properties. This approach cleverly maps the objective skeletal anatomy laws of the three-dimensional real world to a virtual text semantic feature space. The algebraic property of the reciprocal of distance perfectly matches the objective physical law that spatial forces monotonically decrease with increasing physical distance. As the physical distance between skeletal entities increases, the added numerical gain will strictly and progressively decrease to zero. This fundamentally blocks the cross-site false association problem that often occurs in traditional natural language processing. The resulting fusion feature not only includes the logical relevance of medical text but also possesses the tightness of the real human biomechanical transmission structure, giving the digital feature space a real physical common sense.

[0018] 4. By calculating the sum of the inner products of different node vectors in the fusion tensor under the global traversal space, the hierarchical overlap covariance representing the containment relationship between entities is accurately extracted. In the calculation of the overlap ratio, the covariance of each node itself and the objective biomechanical topological distance are simultaneously included in the denominator structure to generate an overlap ratio factor for quantifying the degree of redundancy. This approach can accurately quantify the degree of topological overlap generated by entities with hierarchical containment relationships in the feature space. Furthermore, by utilizing the objective property that the objective physical topological distance is always greater than or equal to the minimum physical interval jump number, an absolutely positive bounded structure is constructed at the algebraic level. This completely eliminates the risk of division-by-zero anomalies and system crashes that may be triggered when features completely overlap. This allows the system to adaptively evaluate the percentage of redundancy caused by the overlap of semantic coverage of any two orthopedic terms under a continuous and safe measurement framework. This provides a proportional benchmark with absolute mathematical rigor for subsequent fine stripping and orthogonal decoupling of high-dimensional space features, ensuring the analytical integrity of the feature quantification process.

[0019] 5. By separating the autocorrelation pure fusion features of the baseline node, the baseline feature is multiplied by the overlap ratio factor obtained in the previous step to obtain redundant feature components. Then, in the high-dimensional space, the redundant projection component is directly subtracted from the original interactive fusion tensor to generate an orthogonalized projection tensor stripped of the influence of hierarchical redundancy. This step, without destroying the core features of the original node, precisely removes the collinear redundant components caused by the nesting of concept levels, like a scalpel. This allows the highly cohesive entities in the feature space to be completely algebraically decoupled, effectively preventing the local singularity collapse caused by the overlap of node concepts during the construction of the graph network. It ensures that the retained feature tensor represents the most essential and unique structural relationship of each orthopedic entity, allowing the digital feature space to return to the ideal state of orthogonality and independence. This greatly improves the structural clarity and logical independence of the subsequent graph network topology connections, eliminates the graph edge confusion caused by the mixed use of hierarchical concepts in medical terminology, and fundamentally purifies the association quality of the underlying knowledge graph.

[0020] 6. By comparing the numerical differences between the biomechanical-semantic fusion tensor and the orthogonalized projection tensor after removing redundancy, the sum of squares of these feature vector differences is calculated, and the square root operation is performed on the globally accumulated sum of squares to generate a semantic drift index that quantifies the degree of feature space distortion. This approach can extremely sensitively and accurately capture the geometric distortion caused by the pre-orthogonalization forced projection operation on the original feature manifold structure. Using the classic Euclidean space metric, the original abstract high-dimensional space basis vector change is transformed into a specific absolute deformation distance. This ensures that the system is no longer blind to the side effects of feature stripping, but clearly grasps the specific physical scale of the deviation of the original semantic center of each orthopedic entity from its original position after redundant resection. This precise quantification of spatial deformation provides the most solid and objective error measurement basis for the subsequent design of adaptive directional torsional repair schemes, avoiding the algorithmic defect in traditional methods where forced dimensionality reduction or decoupling leads to the permanent loss of the original correlation, and ensuring the traceability of feature operations.

[0021] 7. By summing the drift indices of all entities, the total global deformation is obtained. The curvature ratio of the deformation of a single node to the global deformation is calculated. Simultaneously, based on the normalized reference term of the orthogonalized projection tensor, an anticosine geometric mapping is performed to calculate the reference angle. Finally, the curvature compensation phase angle in Riemann space is generated by combining the global ratio. This ingeniously transforms the one-dimensional linear deformation distance into a rotating compass angle in manifold space, so that the correction of semantic drift is no longer a simple and crude numerical addition and subtraction, but a directional complex domain phase torsion. At the same time, the global proportion is used for adaptive adjustment to ensure that feature nodes with more severe deformation can be distributed with a larger compensation torsion weight, realizing elastic distribution within the global error system. This algebraic mapping mechanism with curvature-aware characteristics perfectly compensates for the nonlinear spatial distortion caused by forced projection, and provides a geometrically analytical torsion control parameter for subsequent accurate repair of the edge direction of the map. It ensures that the relative position topology of medical semantic features in complex high-dimensional space can still maintain a high degree of consistency.

[0022] 8. By collecting the forward and reverse compensation phase angles in the bidirectional interactive topology network, the difference between these two phase angles, which carry spatial torsion information, is calculated. The analytic state characteristics of the sine function are then used to perform a wave-symmetric mapping on this difference, generating a phase-wrap discrete error between nodes. This approach completely solves the problem of abrupt changes in periodic boundaries that are easily caused by independent phase compensation in bidirectional graph connections. It innovatively abandons the mathematically discontinuous modulo operation that easily destroys the system's differentiability. Instead, it utilizes the natural periodic symmetry and infinitely smooth differentiable properties of the sine mapping to accurately identify and quantify knots and directional conflicts in bidirectional connections within the graph network. When the bidirectional torsion angles logically match, the system error naturally converges. When periodic misalignment occurs, a smooth error warning value is output. This not only ensures the absolute continuity of the error extraction mechanism but also achieves efficient investigation and numerical stripping of deep-level network structural contradictions without destroying the overall analytical structure of the algorithm, providing high-quality error scalar support without abrupt changes for the final smooth correction.

[0023] 9. By squaring the extracted phase winding discrete error and adding a boundary constant to construct a suppression denominator, a structural integrity suppression scalar limited to the extreme value range is generated using a pure algebraic fraction structure. Finally, the purified orthogonal projection tensor is multiplied and combined with the suppression scalar to output the final edge weight matrix. This approach utilizes the smoothing penalty operator of the Cauchy distribution to complete the final pruning of the knowledge graph's underlying topology. When there are conflicting errors in the bidirectional logic between nodes, the fraction operator can spontaneously increase the denominator to produce a smooth numerical decay, forcibly and gently weakening those false network edges with logical contradictions and physical conflicts. When the logic is completely self-consistent, the full association weight is retained. This design not only perfectly completes the efficient sorting of the entire error correction and compensation link, but also adaptively outputs a final edge network structure without redundant interference and highly consistent with the distance of objective physical entities without any artificial hard threshold truncation. This thoroughly improves the continuity and reality consistency of the underlying knowledge association construction of the orthopedic auxiliary diagnosis and treatment system. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the basic process of the present invention.

[0025] Figure 2 This is a flowchart of the initial normalization and physical attenuation fusion algorithm of the present invention.

[0026] Figure 3 This is a flowchart of the hierarchical overlap extraction and orthogonal decoupling algorithm of the present invention.

[0027] Figure 4 This is a flowchart of the semantic space distortion and compensation algorithm of the present invention.

[0028] Figure 5 This is a flowchart of the winding error smoothing and edge weight generation process of the present invention.

[0029] Figure 6 This is a data flow architecture diagram of the system module for constructing the map in this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1, refer to Figure 1 A method for constructing an orthopedic knowledge graph, comprising: The system receives the co-occurrence frequency matrix of orthopedic entities, calculates the total frequency of each entity, performs normalization processing on it, and generates an initial normalized semantic matrix. The orthopedic entities refer to medical terms related to orthopedics (such as bone locations, disease names, etc.) extracted from electronic medical record text. Receive the biomechanical topological distance matrix, calculate the reciprocal of the topological distance, multiply the reciprocal term by the initial normalized semantic matrix, and generate the biomechanical-semantic fusion tensor; Calculate the inner product of node vectors in the biomechanics-semantic fusion tensor, calculate the hierarchical overlap covariance, and generate the overlap ratio factor by combining the node's own covariance and topological distance. Extract the biomechanical-semantic fusion tensor element of the node itself, multiply the element by the overlap ratio factor, subtract the product from the biomechanical-semantic fusion tensor, and generate the orthogonalized projection tensor; By comparing the biomechanics-semantic fusion tensor with the orthogonalized projection tensor, the sum of squares of the differences in eigenvectors is calculated, and the square root operation is performed on the sum of squares to generate the semantic drift index. Calculate the global curvature ratio of the single-node drift index, calculate the normalized reference term of the orthogonalized projection tensor, and generate the curvature compensation phase angle by combining the global curvature ratio with the inverse cosine mapping angle. Collect the bidirectional compensation phase angles between nodes, calculate the difference between the bidirectional compensation phase angles, perform sine function mapping on the difference, and generate phase winding discrete error; Calculate the square of the phase winding discrete error, construct a structural integrity suppression scalar using a fractional structure with a constant denominator, multiply the orthogonalized projection tensor by this scalar to generate the final edge weight matrix.

[0032] The received orthopedic entity co-occurrence frequency matrix is ​​used to calculate the total frequency of each entity, perform normalization processing on it, and generate an initial normalized semantic matrix, including: The system initially acquires two sets of basic objective data matrices during runtime: The first group is an entity co-occurrence frequency matrix statistically analyzed from the orthopedic clinical electronic medical record text set. The rows and columns represent the pre-identified orthopedic entity nodes, and the matrix elements are the objective number of times the entities co-occur within the text window. The second group is the biomechanical topological distance matrix, where the matrix elements are the number of hops of the shortest mechanical conduction path between two anatomical skeletal nodes. Both sets of matrices have dimensions corresponding to the total number of entities; Calculate the sum of squares of the co-occurrence frequencies of entities between the first node and all traversed nodes, and the sum of squares of the co-occurrence frequencies of entities between the second node and all traversed nodes, respectively. Multiply the sums of two squares, and then perform a square root operation on the product to generate a normalized denominator term; The co-occurrence frequency of entities between the first node and the second node is used as the numerator; Divide the numerator by the normalized denominator to generate the initial normalized semantic similarity between the first node and the second node. Traverse all nodes and combine them to generate an initial normalized semantic matrix.

[0033] The received biomechanical topological distance matrix is ​​used to calculate the reciprocal of the topological distance. This reciprocal term is then multiplied by the initial normalized semantic matrix to generate a biomechanical-semantic fusion tensor, including: Extract the biomechanical topological distance between the first node and the second node; Calculate the reciprocal of the biomechanical topological distance; Adding the reciprocal to a constant one generates the distance attenuation gain term. Multiply the initial normalized semantic similarity between the first node and the second node by the distance decay gain term to generate the biomechanical-semantic fusion tensor element between the first node and the second node; Iterate through all nodes to generate a biomechanical-semantic fusion tensor.

[0034] The method calculates the inner product of node vectors in the biomechanics-semantic fusion tensor, calculates the hierarchical overlap covariance, and combines the node's own covariance with the topological distance to generate an overlap ratio factor, including: Extract the biomechanical-semantic fusion tensor elements between the first node and the traversed nodes, and the biomechanical-semantic fusion tensor elements between the second node and the traversed node; Multiply the two together and sum the product of all traversed nodes to generate the hierarchical overlap covariance between the first and second nodes. Extract the covariance of the first node and the covariance of the second node; Add the covariance of the first node itself, the covariance of the second node itself, and the biomechanical topological distance between the first and second nodes to generate a proportional denominator term; Divide the hierarchical overlap covariance by the proportional denominator to generate the overlap ratio factor between the first node and the second node.

[0035] The extraction of the node's own biomechanical-semantic fusion tensor element, multiplying this element by the overlap ratio factor, and subtracting the product from the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor includes: Extract the biomechanical-semantic fusion tensor elements between the second node and itself; Multiply this element by the overlap ratio factor to generate redundant feature components; Redundant feature components are subtracted from the biomechanical-semantic fusion tensor between the first node and the second node to generate the orthogonalized projection tensor between the first node and the second node; Iterate through all nodes and combine them to generate an orthogonalized projection tensor.

[0036] The comparison of the biomechanics-semantic fusion tensor and the orthogonalized projection tensor calculates the sum of squares of the differences in eigenvectors, performs a square root operation on the sum of squares, and generates a semantic drift index, including: Calculate the difference between the biomechanical-semantic fusion tensor element between the first node and the traversed nodes, and the orthogonalized projection tensor element at the corresponding position; Calculate the square of the difference; Sum the squared differences of all traversed nodes to generate a sum of squares. Perform a square root operation on the sum of squares to generate the semantic drift index of the first node.

[0037] The calculation of the global curvature ratio of the single-node drift index, the calculation of the normalized reference term of the orthogonalized projection tensor, and the generation of the curvature compensation phase angle by combining the inverse cosine mapping angle with the global curvature ratio include: Sum the semantic drift indices of all nodes to generate the global drift sum; Divide the semantic drift index of the first node by the global drift sum to generate the global curvature ratio of the first node. Calculate the sum of squares of the orthogonalized projection tensor elements between the first node and all traversed nodes; Perform a square root operation on the sum of squares to generate a normalized baseline term; Divide the orthogonalized projection tensor elements between the first and second nodes by the normalized reference term; Perform an inverse cosine function operation on the division result to generate a reference angle; Multiply the reference angle by the global curvature ratio to generate the curvature compensation phase angle from the first node to the second node.

[0038] The bidirectional compensation phase angle between the collection nodes is calculated, the difference between the bidirectional compensation phase angles is calculated, and a sine function mapping is performed on the difference to generate a phase winding discrete error, including: Extract the curvature compensation phase angle from the first node to the second node, and the curvature compensation phase angle from the second node to the first node; Calculate the difference between the two curvature-compensated phase angles; Perform a sine function calculation on the difference to generate the phase winding discrete error between the first node and the second node.

[0039] The calculation of the square of the phase winding discrete error involves constructing a structural integrity suppression scalar using a fractional structure with a constant denominator, and multiplying the orthogonalized projection tensor by this scalar to generate the final edge weight matrix, including: Calculate the square of the phase winding discretization error between the first node and the second node; Add a constant 1 to the squared result to generate a suppressed denominator term; Divide the constant by the suppression denominator to generate the structural integrity suppression scalar between the first node and the second node; Multiply the orthogonalized projection tensor elements between the first and second nodes with the structural integrity suppression scalar to generate the final edge weight matrix elements between the first and second nodes. The final edge weight matrix output directly serves as the foundation for the underlying network topology of the orthopedic knowledge graph.

[0040] Example 2, based on Example 1, combined with... Figures 2 to 5 A detailed algorithmic process and principle explanation for each specific step is provided: A method for constructing an orthopedic knowledge graph, including: Reference Figure 2 The initial normalization and physical attenuation fusion algorithm of the present invention is described in detail. The process of receiving the orthopedic entity co-occurrence frequency matrix, calculating the total frequency of each entity, performing normalization processing on it, and generating an initial normalized semantic matrix includes: The original co-occurrence frequency matrix suffers from significant differences in the total frequency of different entities, and its direct use can lead to high-frequency words masking low-frequency key proper nouns. This step transforms absolute frequencies into relative cosine similarity states, providing a scale-uniform benchmark for subsequent fusion.

[0041] During the initial system runtime, two sets of basic objective data matrices are obtained: The first group is the entity co-occurrence frequency matrix statistically analyzed from the orthopedic clinical electronic medical record text set, denoted as... Its rows and columns represent the pre-identified orthopedic entity nodes, and the matrix elements are the objective number of times the entities co-occur within the text window; The second group is a biomechanical topological distance matrix constructed based on standard human skeletal anatomy, denoted as... Its matrix elements are the number of hops (i.e., the number of joints or bones in between) of the shortest mechanical transmission path between two anatomical skeletal nodes. The dimensions of both sets of matrices above are 1. ,in The total number of entities; To eliminate the scaling effect of absolute word frequency, pure co-occurrence direction features of entities are extracted. Next, the initial normalized semantic matrix is ​​calculated using the following formula: In the formula, For nodes With nodes The initial normalized semantic similarity between them; For nodes With nodes The frequency of co-occurrence of entities; For nodes With traversing nodes The frequency of co-occurrence of entities; For nodes With traversing nodes The frequency of co-occurrence of entities; This represents the total dimension of the matrix.

[0042] By obtaining the initial co-occurrence frequency of orthopedic entities in clinical medical records, the interaction distribution features of different nodes in the global text space are extracted. The sum of squares of the co-occurrence frequencies of the target node and all other traversed nodes is calculated, and the square root of the extracted sum of squares is used as the normalization benchmark term of the global feature scale. This transforms the absolute co-occurrence frequency into a relative pure semantic similarity state. This approach can completely eliminate the huge scale difference in the absolute occurrence frequency of different orthopedic terms, effectively preventing the high-frequency word hegemony phenomenon caused by the excessive frequency of common general medical terms. It also avoids high-frequency words from masking key but low-frequency specialty core terms in the feature space, forcing all entity features to be compressed to a unified and fair scale for measurement. This not only provides a unified benchmark data foundation for subsequent high-dimensional matrix fusion, but also ensures that subsequent algorithms focus entirely on the pure co-occurrence direction and association purity when measuring entity correlation, without being affected by the absolute number of words caused by the length of the medical record text or the doctor's personal writing habits. This significantly improves the accuracy and objectivity of the underlying semantic matrix in capturing real medical relevance.

[0043] The received biomechanical topological distance matrix is ​​used to calculate the reciprocal of the topological distance. This reciprocal term is then multiplied by the initial normalized semantic matrix to generate a biomechanical-semantic fusion tensor, including: Simple textual semantics cannot reflect the mechanical transmission relationship in orthopedics. The purpose of this step is to use mechanical topological distance to spatially and physically attenuate and modulate semantic similarity. The closer the skeletal entities are, the stronger the mechanical gain their semantic weights become.

[0044] To directly embed the tightness of mechanical transmission into the semantic space and avoid using unstable exponential empirical functions, the fusion tensor is calculated using an inverse proportional mapping law. The fusion tensor is calculated using the following formula: In the formula, For nodes With nodes Biomechanical-semantic fusion tensor elements between them; For nodes With nodes The initial normalized semantic similarity between them; For nodes With nodes Biomechanical topological distance.

[0045] By extracting the shortest mechanical transmission path hop count between objectively existing human skeletal structures, the reciprocal structure of the biomechanical topological distance is calculated. Combined with a constant bias, a physical gain amplifier with attenuation properties is constructed. This gain term is then directly applied to the previously calculated normalized semantic similarity to generate a biomechanical-semantic fusion tensor with physical distance attenuation properties. This approach cleverly maps the objective skeletal anatomy laws of the three-dimensional real world to a virtual text semantic feature space. The algebraic property of the reciprocal of distance perfectly matches the objective physical law that spatial forces monotonically decrease with increasing physical distance. As the physical distance between skeletal entities increases, the added numerical gain will strictly and progressively decrease to zero. This fundamentally blocks the cross-site false association problem that often occurs in traditional natural language processing. The resulting fusion feature not only includes the logical relevance of medical text but also possesses the tightness of the real human biomechanical transmission structure, giving the digital feature space a real physical common sense.

[0046] Reference Figure 3 The algorithm for hierarchical overlap extraction and orthogonal decoupling of this invention is described in detail. The step of calculating the inner product of node vectors in the biomechanical-semantic fusion tensor, calculating the hierarchical overlap covariance, and generating an overlap ratio factor by combining the node's own covariance and topological distance includes: The purpose of this step is to extract the inner product of the row vectors in the tensor matrix, which is used to characterize the degree of topological overlap caused by the containment relationship between entities.

[0047] For quantization nodes With nodes To determine the degree of overlap in the fusion space, we need to calculate the inner product of the two vectors as the overlap covariance, which is calculated using the following formula: In the formula, For nodes With nodes Hierarchical overlap covariance; For nodes With nodes Biomechanical-semantic fusion tensor elements between them; For nodes With nodes Biomechanical-semantic fusion tensor elements between them; To obtain a standardized overlap ratio factor and prevent division-to-zero anomalies caused by complete overlap of two nodes, the mechanical distance is used as a structural regularization term to calculate the overlap factor using the following formula: In the formula, For nodes With nodes The overlap ratio factor between them; For nodes Its own covariance; For nodes Its own covariance; For nodes With nodes Biomechanical topological distance.

[0048] By calculating the sum of the inner products of different node vectors in the fusion tensor under the global traversal space, the hierarchical overlap covariance representing the containment relationship between entities is accurately extracted. Furthermore, when calculating the overlap ratio, the covariance of each node and the objective biomechanical topological distance are innovatively incorporated into the denominator, generating an overlap ratio factor to quantify the degree of redundancy. This approach can accurately quantify the degree of topological overlap generated by entities with hierarchical containment relationships in the feature space. Moreover, by utilizing the objective property that the objective physical topological distance is always greater than or equal to the minimum physical hop count, an absolutely positively bounded structure is constructed at the algebraic level, completely eliminating the risk of division-by-zero anomalies and system crashes that may be triggered when features completely overlap. This allows the system to adaptively evaluate the percentage of redundancy caused by the semantic overlap of any two orthopedic terms within a continuous and secure metric framework. This provides a mathematically rigorous proportional benchmark for subsequent fine-grained stripping and orthogonal decoupling of high-dimensional features, ensuring the analytical integrity of the feature quantification process.

[0049] The extraction of the node's own biomechanical-semantic fusion tensor element, multiplying this element by the overlap ratio factor, and subtracting the product from the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor includes: The purpose of this step is to use the overlap ratio factor to remove collinear components from overlapping feature vectors through an operation similar to Gram-Schmidt orthogonalization, thereby separating independent structural features.

[0050] To completely remove redundant features caused by hierarchical inclusion relationships, the projected components in the overlapping dimension are subtracted from the biomechanical-semantic fusion tensor. The orthogonalized projected tensor is calculated using the following formula: In the formula, For nodes With nodes Orthogonalized projection tensor elements between them; For nodes With nodes Biomechanical-semantic fusion tensor elements between them; For nodes With nodes Biomechanical-semantic fusion tensor elements between themselves; For nodes With nodes The overlap ratio factor between them.

[0051] By separating the autocorrelation pure fusion features of the baseline node, the baseline feature is multiplied by the overlap ratio factor obtained in the previous step to obtain redundant feature components. Then, in the high-dimensional space, the redundant projection component is directly subtracted from the original interactive fusion tensor to generate an orthogonalized projection tensor stripped of the influence of hierarchical redundancy. This process, without destroying the core features of the original node, precisely removes the collinear redundant components caused by the nesting of concept levels, like a scalpel. This allows the highly cohesive entities in the feature space to be completely algebraically decoupled, effectively preventing the local singularity collapse caused by the overlap of node concepts during the construction of the graph network. It ensures that the retained feature tensor represents the most essential and unique structural relationship of each orthopedic entity, allowing the digital feature space to return to the ideal state of orthogonality and independence. This greatly improves the structural clarity and logical independence of the subsequent graph network topology connections, eliminates the graph edge confusion caused by the mixed use of hierarchical concepts in medical terminology, and fundamentally purifies the association quality of the underlying knowledge graph.

[0052] Reference Figure 4 The semantic space distortion and compensation algorithm of the present invention is described in detail. The step of comparing the biomechanical-semantic fusion tensor and the orthogonalized projection tensor, calculating the sum of squares of the differences in eigenvectors, and performing a square root operation on the sum of squares to generate a semantic drift index includes: Because orthogonalization forcibly alters the basis vectors of the feature space, it changes the relative distances between nodes. This step quantifies the absolute geometric scale of this change.

[0053] To accurately capture the degree of spatial distortion, we will next compare the Euclidean distance difference between the tensor vectors before and after orthogonalization, and calculate the semantic drift exponent using the following formula: In the formula, For nodes The semantic drift index; For nodes With nodes Biomechanical-semantic fusion tensor elements between them; For nodes With nodes The orthogonalized projection tensor elements between them.

[0054] By comparing the numerical differences between the biomechanical-semantic fusion tensor and the orthogonalized projection tensor after removing redundancy, the sum of squares of these feature vector differences is calculated, and the square root operation is performed on the globally accumulated sum of squares to generate a semantic drift index that quantifies the degree of feature space distortion. This approach can extremely sensitively and accurately capture the geometric distortion caused by the pre-orthogonalization forced projection operation on the original feature manifold structure. Using the classic Euclidean space metric, the original abstract high-dimensional space basis vector change is transformed into a specific absolute deformation distance. This means that the system is no longer blind to the side effects of feature stripping, but clearly grasps the specific physical scale of the deviation of the original semantic center of each orthopedic entity from its original position after redundant resection. This precise quantification of spatial deformation provides the most solid and objective error measurement basis for the subsequent design of adaptive directional torsional repair schemes, avoiding the algorithmic defect in traditional methods where forced dimensionality reduction or decoupling leads to the permanent loss of the original correlation, and ensuring the traceability of feature operations.

[0055] The calculation of the global curvature ratio of the single-node drift index, the calculation of the normalized reference term of the orthogonalized projection tensor, and the generation of the curvature compensation phase angle by combining the inverse cosine mapping angle with the global curvature ratio include: The purpose of this step is to convert the one-dimensional drift exponent into a rotational phase angle on the Riemannian manifold, which is used to subsequently correct the weight direction of the graph edges in the complex domain.

[0056] To ensure that the degree of node compensation is adaptively controlled by the system's global distortion ratio, the global proportion of node drift is first calculated using the following formula: In the formula, For nodes The global curvature ratio; For nodes The semantic drift index; For nodes The semantic drift index; To determine the reference angle using the purely algebraic properties of orthogonal features and apply compensation in conjunction with the curvature ratio, the phase angle is then generated via inverse cosine mapping, and the curvature compensation phase angle is calculated using the following formula: In the formula, For nodes To the node Curvature compensation phase angle in the direction; For nodes With nodes Orthogonalized projection tensor elements between them; For nodes With nodes Orthogonalized projection tensor elements between them; For nodes The global curvature ratio.

[0057] By summing the drift indices of all entities to obtain the total global deformation, the curvature ratio of the deformation of a single node to the global deformation is calculated. Simultaneously, based on the normalized reference term of the orthogonalized projection tensor, an inverse cosine geometric mapping is performed to calculate the reference angle. Finally, the curvature compensation phase angle in Riemann space is generated by combining the global ratio. This ingeniously transforms the one-dimensional linear deformation distance into a rotating compass angle in manifold space, so that the correction of semantic drift is no longer a simple and crude numerical addition and subtraction, but a directional complex domain phase torsion. At the same time, the global proportion is used for adaptive adjustment to ensure that feature nodes with more severe deformation can be distributed with a larger compensation torsion weight, realizing elastic distribution within the global error system. This algebraic mapping mechanism with curvature-aware characteristics perfectly compensates for the nonlinear spatial distortion caused by forced projection, and provides a geometrically analytical torsion control parameter for subsequent accurate repair of the edge direction of the map, ensuring that the relative position topology of medical semantic features in complex high-dimensional space can still maintain a high degree of consistency.

[0058] Reference Figure 5 This document provides a detailed explanation of the winding error smoothing and edge weight generation process of the present invention. The process includes collecting the bidirectional compensation phase angle between nodes, calculating the difference between the bidirectional compensation phase angles, performing a sine function mapping on the difference, and generating a phase winding discrete error, comprising: The edges in the graph are bidirectional; if a node arrive Compensation phase and arrive The compensation phase crosses the periodic boundary of the trigonometric function, which can cause abrupt truncation. This step aims to utilize the periodic symmetry of the sine function to accurately extract this winding error.

[0059] To avoid compromising the integrability of the system by using discontinuous remainder operations (such as modulo operations), the analytical form of the sine function is used to calculate the partial differential error of the two-way phase. The partial differential error of the two-way phase is calculated using the following formula: In the formula, For nodes With nodes Phase winding discrepancy error between them; For nodes To the node Curvature compensation phase angle in the direction; For nodes To the node Curvature compensation phase angle in the direction.

[0060] By collecting the forward and reverse compensation phase angles in the bidirectional interactive topology network, the difference between these two phase angles, which carry spatial torsion information, is calculated. The analytic state characteristics of the sine function are then used to perform a wave-symmetric mapping on this difference, generating a phase-wrap discrete error between nodes. This approach completely solves the problem of abrupt changes in periodic boundaries that are easily caused by independent phase compensation in bidirectional graph connections. It innovatively abandons the mathematically discontinuous modulo operation that easily destroys the system's differentiability, instead utilizing the natural periodic symmetry and infinitely smooth differentiable properties of the sine mapping to accurately identify and quantify knots and directional conflicts in bidirectional connections within the graph network. When the bidirectional torsion angles logically match, the system error naturally converges; when periodic misalignment occurs, a smooth error warning value is output. This not only ensures the absolute continuity of the error extraction mechanism but also achieves efficient screening and numerical stripping of deep-level network structural contradictions without disrupting the overall analytical structure of the algorithm, providing high-quality error scalar support without abrupt changes for the final smooth correction.

[0061] The calculation of the square of the phase winding discrete error involves constructing a structural integrity suppression scalar using a fractional structure with a constant denominator, and multiplying the orthogonalized projection tensor by this scalar to generate the final edge weight matrix, including: The extracted winding error is converted into a penalty scalar to suppress edges with logical conflicts, and then the final knowledge graph topological weights without mutations are calculated to complete the closed-loop output.

[0062] The structural integrity inhibition scalar is calculated using the following formula: In the formula, For nodes With nodes Structural integrity inhibition scalar between; For nodes With nodes Phase winding discrepancy error between them; Finally, to generate the final atlas data structure that satisfies continuity and contains deep biomechanical associations, a modified scalar is applied to the orthogonal space to generate the final edge weight matrix using the following formula: In the formula, For nodes in a knowledge graph With nodes The final edge weight matrix elements between them; For nodes With nodes Orthogonalized projection tensor elements between them; For nodes With nodes Structural integrity inhibition scalar between; Output matrix It serves directly as the underlying network topology for constructing the orthopedic knowledge graph.

[0063] By squaring the extracted phase winding discrete error and adding a boundary constant to construct a suppression denominator, a structural integrity suppression scalar limited to the extreme value range is generated using a pure algebraic fraction structure. Finally, the purified orthogonal projection tensor is multiplied and combined with the suppression scalar to output the final edge weight matrix. This approach utilizes the smoothing penalty operator of the Cauchy distribution to complete the final pruning of the knowledge graph's underlying topology. When there are conflicting errors in the bidirectional logic between nodes, the fraction operator can spontaneously increase the denominator to produce a smooth numerical decay, forcibly and gently weakening those false network edges with logical contradictions and physical conflicts. When the logic is completely self-consistent, the full association weight is retained. This design not only perfectly completes the efficient sorting of the entire error correction and compensation link, but also adaptively outputs a final edge network structure without redundant interference and highly consistent with the distance of objective physical entities without any artificial hard threshold truncation. This thoroughly improves the continuity and reality consistency of the underlying knowledge association construction of the orthopedic auxiliary diagnosis and treatment system.

[0064] Reference Figure 6 Example 3: A system for implementing the orthopedic knowledge graph construction method, comprising: Data receiving and fusion analysis module: used to receive the co-occurrence frequency matrix of orthopedic entities and the biomechanical topological distance matrix, calculate the normalized initial semantic matrix, and generate the biomechanical-semantic fusion tensor by combining the inverse term of the topological distance; The core module for overlap elimination and orthogonal decoupling is used to calculate the hierarchical overlap covariance and overlap ratio factor of the node vectors in the biomechanical-semantic fusion tensor, and to perform semantic-topological orthogonalization projection on the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor that separates redundancy. Spatial distortion and phase compensation module: used to calculate the sum of squared feature differences before and after orthogonal projection to obtain the semantic drift index, extract the global curvature ratio, and generate the Riemann curvature compensation phase angle through inverse cosine mapping; Error smoothing and topology weight generation module: used to perform sine mapping on the bidirectional compensated phase angle difference to extract the phase winding discrete error, thereby constructing a structural integrity suppression scalar and correcting the orthogonalized projection tensor to output the final edge weight matrix.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing an orthopedic knowledge graph, characterized in that, include: Receive the co-occurrence frequency matrix of orthopedic entities, calculate the total frequency of each entity, perform normalization processing on it, and generate an initial normalized semantic matrix; Receive the biomechanical topological distance matrix, calculate the reciprocal of the topological distance, multiply the reciprocal term by the initial normalized semantic matrix, and generate the biomechanical-semantic fusion tensor; Calculate the inner product of node vectors in the biomechanics-semantic fusion tensor, calculate the hierarchical overlap covariance, and generate the overlap ratio factor by combining the node's own covariance and topological distance. Extract the biomechanical-semantic fusion tensor element of the node itself, multiply the element by the overlap ratio factor, subtract the product from the biomechanical-semantic fusion tensor, and generate the orthogonalized projection tensor; By comparing the biomechanics-semantic fusion tensor with the orthogonalized projection tensor, the sum of squares of the differences in eigenvectors is calculated, and the square root operation is performed on the sum of squares to generate the semantic drift index. Calculate the global curvature ratio of the single-node drift index, calculate the normalized reference term of the orthogonalized projection tensor, and generate the curvature compensation phase angle by combining the global curvature ratio with the inverse cosine mapping angle. Collect the bidirectional compensation phase angles between nodes, calculate the difference between the bidirectional compensation phase angles, perform sine function mapping on the difference, and generate phase winding discrete error; Calculate the square of the phase winding discrete error, construct a structural integrity suppression scalar using a fractional structure with a constant denominator, multiply the orthogonalized projection tensor by this scalar to generate the final edge weight matrix.

2. The method for constructing an orthopedic knowledge graph according to claim 1, characterized in that, The received orthopedic entity co-occurrence frequency matrix is ​​used to calculate the total frequency of each entity, perform normalization processing on it, and generate an initial normalized semantic matrix, including: The system initially acquires two sets of basic objective data matrices during runtime: The first group is an entity co-occurrence frequency matrix statistically analyzed from the orthopedic clinical electronic medical record text set. The rows and columns represent the pre-identified orthopedic entity nodes, and the matrix elements are the objective number of times the entities co-occur within the text window. The second group is the biomechanical topological distance matrix, where the matrix elements are the number of hops of the shortest mechanical conduction path between two anatomical skeletal nodes. Both sets of matrices have dimensions corresponding to the total number of entities; Calculate the sum of squares of the co-occurrence frequencies of entities between the first node and all traversed nodes, and the sum of squares of the co-occurrence frequencies of entities between the second node and all traversed nodes, respectively. Multiply the sums of two squares, and then perform a square root operation on the product to generate a normalized denominator term; The co-occurrence frequency of entities between the first node and the second node is used as the numerator; Divide the numerator by the normalized denominator to generate the initial normalized semantic similarity between the first node and the second node. Traverse all nodes and combine them to generate an initial normalized semantic matrix.

3. The method for constructing an orthopedic knowledge graph according to claim 2, characterized in that, The received biomechanical topological distance matrix is ​​used to calculate the reciprocal of the topological distance. This reciprocal term is then multiplied by the initial normalized semantic matrix to generate a biomechanical-semantic fusion tensor, including: Extract the biomechanical topological distance between the first node and the second node; Calculate the reciprocal of the biomechanical topological distance; Adding the reciprocal to a constant one generates the distance attenuation gain term. Multiply the initial normalized semantic similarity between the first node and the second node by the distance decay gain term to generate the biomechanical-semantic fusion tensor element between the first node and the second node; Iterate through all nodes to generate a biomechanical-semantic fusion tensor.

4. The method for constructing an orthopedic knowledge graph according to claim 3, characterized in that, The method calculates the inner product of node vectors in the biomechanics-semantic fusion tensor, calculates the hierarchical overlap covariance, and combines the node's own covariance with the topological distance to generate an overlap ratio factor, including: Extract the biomechanical-semantic fusion tensor elements between the first node and the traversed nodes, and the biomechanical-semantic fusion tensor elements between the second node and the traversed node; Multiply the two together and sum the product of all traversed nodes to generate the hierarchical overlap covariance between the first and second nodes. Extract the covariance of the first node and the covariance of the second node; Add the covariance of the first node itself, the covariance of the second node itself, and the biomechanical topological distance between the first and second nodes to generate a proportional denominator term; Divide the hierarchical overlap covariance by the proportional denominator to generate the overlap ratio factor between the first node and the second node.

5. The method for constructing an orthopedic knowledge graph according to claim 4, characterized in that, The extraction of the node's own biomechanical-semantic fusion tensor element, multiplying this element by the overlap ratio factor, and subtracting the product from the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor includes: Extract the biomechanical-semantic fusion tensor elements between the second node and itself; Multiply this element by the overlap ratio factor to generate redundant feature components; Redundant feature components are subtracted from the biomechanical-semantic fusion tensor between the first node and the second node to generate the orthogonalized projection tensor between the first node and the second node; Iterate through all nodes and combine them to generate an orthogonalized projection tensor.

6. The method for constructing an orthopedic knowledge graph according to claim 5, characterized in that, The comparison of the biomechanics-semantic fusion tensor and the orthogonalized projection tensor calculates the sum of squares of the differences in eigenvectors, performs a square root operation on the sum of squares, and generates a semantic drift index, including: Calculate the difference between the biomechanical-semantic fusion tensor element between the first node and the traversed nodes, and the orthogonalized projection tensor element at the corresponding position; Calculate the square of the difference; Sum the squared differences of all traversed nodes to generate a sum of squares. Perform a square root operation on the sum of squares to generate the semantic drift index of the first node.

7. The method for constructing an orthopedic knowledge graph according to claim 6, characterized in that, The calculation of the global curvature ratio of the single-node drift index, the calculation of the normalized reference term of the orthogonalized projection tensor, and the generation of the curvature compensation phase angle by combining the inverse cosine mapping angle with the global curvature ratio include: Sum the semantic drift indices of all nodes to generate the global drift sum; Divide the semantic drift index of the first node by the global drift sum to generate the global curvature ratio of the first node. Calculate the sum of squares of the orthogonalized projection tensor elements between the first node and all traversed nodes; Perform a square root operation on the sum of squares to generate a normalized baseline term; Divide the orthogonalized projection tensor elements between the first and second nodes by the normalized reference term; Perform an inverse cosine function operation on the division result to generate a reference angle; Multiply the reference angle by the global curvature ratio to generate the curvature compensation phase angle from the first node to the second node.

8. The method for constructing an orthopedic knowledge graph according to claim 7, characterized in that, The bidirectional compensation phase angle between the collection nodes is calculated, the difference between the bidirectional compensation phase angles is calculated, and a sine function mapping is performed on the difference to generate a phase winding discrete error, including: Extract the curvature compensation phase angle from the first node to the second node, and the curvature compensation phase angle from the second node to the first node; Calculate the difference between the two curvature-compensated phase angles; Perform a sine function calculation on the difference to generate the phase winding discrete error between the first node and the second node.

9. The method for constructing an orthopedic knowledge graph according to claim 8, characterized in that, The calculation of the square of the phase winding discrete error involves constructing a structural integrity suppression scalar using a fractional structure with a constant denominator, and multiplying the orthogonalized projection tensor by this scalar to generate the final edge weight matrix, including: Calculate the square of the phase winding discretization error between the first node and the second node; Add a constant 1 to the squared result to generate a suppressed denominator term; Divide the constant by the suppression denominator to generate the structural integrity suppression scalar between the first node and the second node; Multiply the orthogonalized projection tensor elements between the first and second nodes with the structural integrity suppression scalar to generate the final edge weight matrix elements between the first and second nodes. The final edge weight matrix output directly serves as the foundation for the underlying network topology of the orthopedic knowledge graph.

10. A system employing the orthopedic knowledge graph construction method of claim 1, characterized in that, include: Data receiving and fusion analysis module: used to receive the co-occurrence frequency matrix of orthopedic entities and the biomechanical topological distance matrix, calculate the normalized initial semantic matrix, and generate the biomechanical-semantic fusion tensor by combining the inverse term of the topological distance; The core module for overlap elimination and orthogonal decoupling is used to calculate the hierarchical overlap covariance and overlap ratio factor of the node vectors in the biomechanical-semantic fusion tensor, and to perform semantic-topological orthogonalization projection on the biomechanical-semantic fusion tensor to generate an orthogonalized projection tensor that separates redundancy. Spatial distortion and phase compensation module: used to calculate the sum of squared feature differences before and after orthogonal projection to obtain the semantic drift index, extract the global curvature ratio, and generate the Riemann curvature compensation phase angle through inverse cosine mapping; Error smoothing and topology weight generation module: used to perform sine mapping on the bidirectional compensated phase angle difference to extract the phase winding discrete error, thereby constructing a structural integrity suppression scalar and correcting the orthogonalized projection tensor to output the final edge weight matrix.

Images