An ontology identification method and device for a decision knowledge graph

Abstract

The application discloses a kind of ontology identification method and device for decision knowledge graph, the method includes: to the text material of decision related field, ontology knowledge extraction is carried out, and the ontology text set is obtained;The ontology text set includes the attribute text set and rule text set of each ontology;The attribute text set includes various attribute texts;The rule text set includes various rule texts;The pre-processing text data set is obtained by pre-processing to the ontology text set;The text data set of each ontology in the pre-processing text data set is handled by ontology type identification, and the type identification result of each ontology is obtained;The type of the ontology includes fact type ontology, rule type ontology and decision type ontology.

Description

Technical Field

[0001] This invention relates to the fields of knowledge graphs, industrial data processing, and intelligent decision-making technologies, specifically to an ontology recognition method and apparatus for decision-making knowledge graphs. Background Technology

[0002] With the rapid development of big data and artificial intelligence technologies, decision knowledge graphs are being applied more and more widely in various fields such as government administration, enterprise operations, and intelligent decision-making. Their core value lies in achieving efficient knowledge retrieval, reasoning, and reuse through structured ontology organization. As the basic building block of knowledge graphs, accurately identifying the type of ontology (factual, rule-based, decision-based) is a key prerequisite for ensuring the quality of knowledge graph construction and improving decision support capabilities.

[0003] Currently, existing ontology recognition methods mostly focus on entity extraction and classification in general domains, lacking specialized designs for decision-related domains. Textual materials in decision-related domains encompass various types, including dictionaries, textbooks, monographs, and standards. The text content includes both objective factual descriptions and complex rule definitions and decision logic explanations, resulting in overlapping attribute and rule features of the ontology. Traditional methods struggle to effectively distinguish the core differences between different ontology types. Furthermore, existing technologies often suffer from issues such as non-standard text processing, incomplete feature extraction, and simplistic classification logic in ontology recognition: insufficient precision in text disambiguation and alignment stages easily leads to deviations in subsequent feature extraction; feature construction often relies on single-dimensional textual information, failing to fully integrate the deep relationships between attribute and rule data; and classification models often employ simple single-feature judgment methods, which are ill-suited to the complex classification requirements of ontology types in decision-making domains. Ultimately, this results in low ontology type recognition accuracy and insufficient generalization ability, failing to meet the professional and high-precision requirements for constructing decision knowledge graphs.

[0004] Furthermore, existing methods do not consider the impact of data dimensionality redundancy and noise interference during feature processing, and lack effective dimensionality reduction and clustering optimization strategies. This results in high computational complexity and low processing efficiency, making them unsuitable for ontology recognition scenarios involving large-scale decision-making text data. These problems severely restrict the construction efficiency and application effectiveness of decision knowledge graphs, necessitating a targeted, accurate, and efficient ontology recognition method to improve decision accuracy. Summary of the Invention

[0005] This invention primarily addresses how to improve the construction efficiency and application effect of decision knowledge graphs, and achieve targeted, high-precision, and efficient ontology recognition to improve decision accuracy. This invention discloses an ontology recognition method and apparatus for decision knowledge graphs.

[0006] In a first aspect, this invention discloses an ontology recognition method for decision knowledge graphs, comprising: S1, Ontology knowledge is extracted from textual data in decision-related fields to obtain an ontology text set; the ontology text set includes an attribute text set and a rule text set for each ontology; the attribute text set includes various types of attribute text; the rule text set includes various types of rule text. S2, preprocess the ontology text set to obtain a preprocessed text data set; S3, perform ontology type identification processing on the text data set of each ontology in the preprocessed text data set to obtain the type identification result of each ontology; the types of the ontology include factual ontology, rule-based ontology, and decision-based ontology.

[0007] The preprocessing of the ontology text set to obtain a preprocessed text data set includes: S21, perform text disambiguation processing on the ontology text set to obtain a first text set; S22, perform text alignment processing on the first text set to obtain the second text set; S23, perform text vector transformation on the second text set to obtain a preprocessed text data set; the preprocessed text data set includes the text data set of each ontology; the text data set includes the attribute text data set and the rule text data set.

[0008] The step of performing ontology type identification processing on the text data set of each ontology in the preprocessed text data set to obtain the type identification result of each ontology includes: S31, perform a first classification process on the attribute text data set in the text data set of each ontology in the preprocessed text data set to obtain the corresponding first classification value; S32, perform a second classification process on the rule text data set in the text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value; S33, perform joint classification processing on the first classification value and the second classification value to obtain the type identification result of the ontology.

[0009] The first classification process, which involves performing a first classification on the attribute text data set of each ontology in the preprocessed text data set to obtain a corresponding first classification value, includes: S311, For the text data sequence of each attribute in the attribute text data set of each ontology in the preprocessed text data set, calculate the corresponding statistical feature vector respectively; The statistical feature vector includes the mean of the text data sequence. ,variance skewness and kurtosis The calculation formula is as follows: , , , in, Let i be the value of the i-th element in the text data sequence. The length of the text data sequence; S312, construct the global feature matrix by using the statistical feature vector of each attribute as a row vector. ,in The number of attributes; S313, regarding the global feature matrix Perform dimensionality reduction projection to obtain the dimensionality reduction matrix: S314, using the row vectors of the reduced dimension matrix as basic elements, perform clustering processing on the basic elements of the reduced dimension matrix to obtain category information; the category information includes several basic elements belonging to the category; S315, take the average value of all basic elements contained in the category information containing the most basic elements as the first classification value corresponding to the ontology.

[0010] The global feature matrix Dimensionality reduction projection is performed to obtain the dimension reduction matrix, including: S3131, regarding the global feature matrix The covariance matrix is ​​calculated. Its calculation expression is: ; S3132, for Perform feature decomposition and select the top features. The eigenvectors corresponding to the largest eigenvalues ​​constitute the projection matrix. ; S3133, regarding the global feature matrix The dimensionality reduction matrix is ​​calculated. Its calculation expression is: .

[0011] The step of performing a second classification process on the rule text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value includes: S321, for each rule's text data sequence in the rule text data set of each ontology in the preprocessed text data set, perform piecewise linear fitting to obtain the corresponding fitting matrix: S322, Perform metric fusion calculation on each fitted matrix to obtain a set of metric values; S323, perform clustering on the set of metric values ​​of all fitted matrices to obtain category information; each category information includes several elements, wherein the elements are the set of metric values; S324, determine the category information with the largest number of elements, which is the classification category information; S325, calculate the mean of all the measurement value sets contained in the classification category information, which is the second classification value corresponding to the ontology.

[0012] The joint classification processing of the first classification value and the second classification value to obtain the type identification result of the ontology includes: S331, Based on the first classification value and the second classification value, a classification feature vector is constructed; S332, Perform feature encoding processing on the classification feature vector to obtain the encoded vector; S333, perform feature distance calculation on the encoded vector to obtain a distance value; S334, based on a preset ontology type recognition result discrimination model, determine the range of values ​​to which the distance value belongs, and determine the type information corresponding to the range of values, which is the ontology type recognition result.

[0013] A second aspect of the present invention discloses an ontology recognition device for decision knowledge graphs, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the ontology recognition method for decision-oriented knowledge graphs.

[0014] In a third aspect, the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked by a computer, are used to execute the ontology recognition method for decision-oriented knowledge graphs.

[0015] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the ontology recognition method for decision-oriented knowledge graphs.

[0016] The beneficial effects of this invention are as follows: This invention systematically extracts ontology knowledge from textual data in decision-related fields, clearly distinguishing between attribute text sets and rule text sets of the ontology. This provides structured and accurate data source support for subsequent type identification, effectively avoiding the identification bias problem caused by mixed textual information in traditional methods, and significantly improving the basic data quality of ontology identification. This invention employs a preprocessing workflow that combines text disambiguation, text alignment, and text vector transformation. By leveraging specialized NLP tools and models, it achieves standardized processing of text data, effectively eliminating ambiguous information in the text, aligning the differences in expression between texts from different sources, and transforming unstructured text into computable vector data. This lays a solid foundation for subsequent feature extraction and classification, significantly reduces noise interference during feature extraction, and enhances the overall method's anti-interference capability.

[0017] This invention designs a first classification process for attribute text datasets, which involves constructing statistical feature vectors, reducing the dimensionality of the global feature matrix, and performing cluster analysis. By calculating key statistical features such as the mean, variance, skewness, and kurtosis of the text sequence, and combining a dimensionality reduction strategy based on covariance matrix eigenvalue decomposition with an iteratively optimized clustering algorithm, the inherent feature patterns of attribute text are accurately discovered. The resulting first classification value can objectively reflect the essential attribute features of the ontology, providing a reliable attribute dimension basis for type identification.

[0018] This invention employs a second classification process for rule text datasets, involving piecewise linear fitting, metric fusion calculation, and cluster analysis. By dynamically fitting the rule text to extract its dynamic feature matrix, it integrates multiple matrix norms, maximum eigenvalues, and rank values ​​to construct a comprehensive set of metrics. Then, through cluster analysis, it obtains a second classification value that can characterize the core features of the rule text. This fully captures the complex logical features of the rule text and overcomes the shortcomings of traditional methods in characterizing rule information. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention. Detailed Implementation

[0020] To better understand the content of this invention, an embodiment is provided here.

[0021] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention.

[0022] In a first aspect, this invention discloses an ontology recognition method for decision knowledge graphs, comprising: S1, Ontology knowledge is extracted from textual data in decision-related fields to obtain an ontology text set; the ontology text set includes an attribute text set and a rule text set for each ontology; the attribute text set includes various types of attribute text; the rule text set includes various types of rule text. S2, preprocess the ontology text set to obtain a preprocessed text data set; S3, perform ontology type identification processing on the text data set of each ontology in the preprocessed text data set to obtain the type identification result of each ontology; the types of the ontology include factual ontology, rule-based ontology, and decision-based ontology.

[0023] The textual materials in the decision-related field include existing dictionaries, classic textbooks, monographs, standards and norms in the decision-related field.

[0024] The attribute text refers to the text of various attribute information of an ontology. For example, the attribute text of a geographic name ontology includes geographic coordinates, value range, spatial precision, etc., while the rule text includes naming method, projection method, relationship information with adjacent ontologies, relationship calculation rules, etc.

[0025] Factual ontologies address the "what" question, including general spatiotemporal ontologies that express domain ontologies, objectively existing entity element ontologies, and event ontologies that actually occur. Rule-based ontologies address the "why" question, referring to ontologies containing axioms, theorems, laws, and computational rules governing the objective existence, evolution, interaction, and influence of various factual ontologies within the domain. Decision-based ontologies address the "how" question, referring to ontologies that provide decision-making support for solving various application tasks within the domain.

[0026] The ontology knowledge extraction can be implemented using TextBlob or PyTorch-NLP, based on a preset set of ontology keywords; the ontology is an entity in the knowledge graph domain. The preprocessing of the ontology text set to obtain a preprocessed text data set includes: S21, perform text disambiguation processing on the ontology text set to obtain a first text set; S22, perform text alignment processing on the first text set to obtain the second text set; S23, perform text vector transformation on the second text set to obtain a preprocessed text data set; the preprocessed text data set includes the text data set of each ontology; the text data set includes attribute text data set and rule text data set; the text data set includes several text data sequences. The text vector conversion process can be implemented using the word2vec command or the text vector conversion command in Python.

[0027] The text disambiguation process is implemented using the disambiguation function in the Expert.ai NL API library in Python.

[0028] The text alignment process can be implemented using a seq2seq model or word vector alignment in NLP.

[0029] The step of performing ontology type identification processing on the text data set of each ontology in the preprocessed text data set to obtain the type identification result of each ontology includes: S31, perform a first classification process on the attribute text data set in the text data set of each ontology in the preprocessed text data set to obtain the corresponding first classification value; S32, perform a second classification process on the rule text data set in the text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value; S33, perform joint classification processing on the first classification value and the second classification value to obtain the type identification result of the ontology.

[0030] The first classification process, which involves performing a first classification on the attribute text data set of each ontology in the preprocessed text data set to obtain a corresponding first classification value, includes: S311, For the text data sequence of each attribute in the attribute text data set of each ontology in the preprocessed text data set, calculate the corresponding statistical feature vector respectively; The statistical feature vector includes the mean of the text data sequence. ,variance skewness and kurtosis The calculation formula is as follows: , , , in, Let i be the value of the i-th element in the text data sequence. The length of the text data sequence.

[0031] S312, construct the global feature matrix by using the statistical feature vector of each attribute as a row vector. ,in The number of attributes; S313, regarding the global feature matrix Perform dimensionality reduction projection to obtain the dimensionality reduction matrix: The text data sequence mean calculation method used in this invention can accurately extract the core distribution center of the attribute text data sequence, objectively reflect the overall value level of the sequence, provide a basic reference standard for subsequent feature analysis, effectively avoid feature misjudgment caused by the deviation of a single data point, and ensure the accuracy of the characterization of the basic features of attribute text.

[0032] The text data sequence variance calculation method used in this invention can quantitatively describe the degree of deviation of each element in the attribute text data sequence from the core distribution center, clearly present the dispersion state of the data, make up for the deficiency that relying solely on central tendency cannot fully reflect the data characteristics, make the feature characterization of attribute text more three-dimensional and comprehensive, and provide key discrete feature support for subsequent feature matrix construction.

[0033] The text data sequence skewness calculation method used in this invention can accurately capture the distribution skewness characteristics of attribute text data sequences, reveal the asymmetric distribution pattern of data on both sides of the core distribution center, break through the limitations of traditional feature descriptions that only focus on the degree of concentration and dispersion, further enrich the statistical feature dimensions of attribute text, and make the feature characterization more consistent with the actual data distribution.

[0034] The text data sequence kurtosis calculation method used in this invention can effectively measure the steepness or flatness of the distribution of attribute text data sequences, reflect the concentration of data values, and, combined with the characteristics of central tendency, dispersion, and distribution asymmetry, construct a complete statistical feature system. This comprehensively outlines the inherent feature patterns of attribute text data sequences and provides a comprehensive and in-depth feature basis for the subsequent construction of feature matrices.

[0035] The global feature matrix covariance matrix calculation method used in this invention can accurately measure the degree of linear correlation between various statistical features of attribute text, clearly reveal the inherent correlation rules between different feature dimensions, provide key correlation basis for subsequent dimensionality reduction processing, avoid the loss of key information due to ignoring feature correlation during dimensionality reduction, and ensure that the data after dimensionality reduction can still retain the core correlation information of the original features.

[0036] The global feature matrix Dimensionality reduction projection is performed to obtain the dimension reduction matrix, including: S3131, Calculate the covariance matrix ; S3132, for Perform feature decomposition and select the top features. The eigenvectors corresponding to the largest eigenvalues ​​constitute the projection matrix. ; S3133, calculate the dimensionality-reduced matrix. .

[0037] S314, using the row vectors of the reduced dimension matrix as basic elements, perform clustering processing on the basic elements of the reduced dimension matrix to obtain category information; The step of clustering the basic elements of the reduced-dimensional matrix using the row vectors as basic elements to obtain category information includes: S3141, Initialization Cluster centers ; S3142, for each sample Calculate its Euclidean distance to all centers. And assign it to the category corresponding to the nearest center; S3143, update the center of each category to the mean of its constituent samples; S3144, Repeat steps S3142 to S3143 until the center change is less than the threshold. The category information is obtained; the category information includes the samples contained in each category.

[0038] S315, take the average of all basic elements contained in the category information with the largest number of samples as the first classification value corresponding to the ontology.

[0039] The dimension reduction matrix calculation method adopted in this invention constructs a projection matrix by selecting core feature vectors through feature decomposition, and projects the high-dimensional global feature matrix into a low-dimensional space. While significantly reducing data dimensionality redundancy and subsequent computational complexity, it retains the key information of the original features to the maximum extent, achieving a balance between data simplification and feature effectiveness, improving the overall processing efficiency of the method, and avoiding the risk of overfitting caused by high-dimensional data.

[0040] The Euclidean distance calculation method between samples and cluster centers used in this invention can accurately measure the similarity between each dimension-reduced attribute feature sample and each cluster center, providing an objective and quantitative basis for sample category assignment, ensuring the accuracy of sample classification during the clustering process, and allowing samples with similar attribute features to be accurately classified into one category, laying a reliable foundation for the subsequent determination of the first classification value.

[0041] The step of performing a second classification process on the rule text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value includes: S321, Perform piecewise linear fitting on the text data sequence for each rule to obtain the corresponding fitting matrix: S322, Perform metric fusion calculation on each fitted matrix to obtain a set of metric values; S323, perform clustering on the set of metric values ​​of all fitted matrices to obtain category information; each category information includes several elements, wherein the elements are the set of metric values; S324, determine the category information with the largest number of elements, which is the classification category information; S325, calculate the mean of all the measurement value sets contained in the classification category information, which is the second classification value corresponding to the ontology.

[0042] The step of performing piecewise linear fitting on the text data sequence for each rule to obtain the corresponding fitting matrix includes: Divide the text data sequence into Segment, minimum length of each segment ; For each data segment Fitting a linear model Calculate the sum of squared residuals ; like This segment is further divided into two segments until all segments satisfy the condition. .

[0043] Extracting the fitting parameters for each segment and segment length Construct a rule-based dynamic feature matrix , This represents the final number of segments.

[0044] The step of performing metric fusion calculation on each fitted matrix to obtain a set of metric values ​​includes: For each fitted matrix, the corresponding set of matrix norm values ​​is calculated; the set of matrix norm values ​​includes Norms, Frobenius norms, row norms, column norms; For each fitted matrix, the corresponding maximum eigenvalue and rank are calculated; Using each fitted matrix The norm, Frobenius norm, row norm, column norm, largest eigenvalue, and rank are used to construct the corresponding set of metrics.

[0045] The calculated mean of all the measurement values ​​included in the classification information is used as the second classification value corresponding to the ontology. This mean is calculated as the mean of each element in the set of all measurement values ​​included in the classification information, and is used as the second classification value. The mean of all elements includes... Norm mean, Frobenius norm mean, row sum norm mean, column sum norm mean, maximum eigenvalue mean, and rank mean; The piecewise linear fitting residual sum of squares method used in this invention can quantitatively evaluate the degree of fit between each segment of the regular text data sequence and the linear fitting model, providing a clear criterion for judging the rationality of the segmentation. By dynamically splitting segments that do not meet the fitting requirements, it ensures that the piecewise fitting of the regular text data sequence can accurately capture its inherent change pattern, and that the constructed regular dynamic feature matrix can truly reflect the core features of the regular text.

[0046] The joint classification processing of the first classification value and the second classification value to obtain the type identification result of the ontology includes: S331, Based on the first classification value and the second classification value, a classification feature vector is constructed; S332, Perform feature encoding processing on the classification feature vector to obtain the encoded vector; S333, perform feature distance calculation on the encoded vector to obtain a distance value; S334, based on a preset ontology type recognition result discrimination model, determine the range of values ​​to which the distance value belongs, and determine the type information corresponding to the range of values, which is the ontology type recognition result.

[0047] The ontology type identification result discrimination model includes several value ranges and corresponding type information. The type information includes factual ontology, rule-based ontology, and decision-based ontology.

[0048] The classification feature vector is represented as follows: ,in The vector corresponding to the first category value. This is the vector corresponding to the second classification value; The step of performing feature encoding processing on the classification feature vector to obtain the encoded vector includes: Calculate the first weight vector ,in The learnable parameter matrix; The second weight vector is calculated. ; The fused output encoded vector is expressed as follows: ,in This is an element-wise multiplication.

[0049] The expression for calculating the feature distance is: , in, Here is the distance value, where and This is the mean and covariance matrix of the historical data.

[0050] The Mahalanobis distance calculation method for encoded vectors used in this invention, by combining the overall distribution characteristics (mean and covariance matrix) of historical data, can accurately measure the degree of deviation between the encoded vector and the distribution center of historical data. It effectively avoids the influence of differences in the dimensions and correlations of different feature dimensions on the distance calculation, so that the distance value can objectively and accurately reflect the feature attribution of the encoded vector. This provides a high-precision quantitative basis for the discrimination of ontology type recognition results and significantly improves the accuracy and reliability of type recognition.

[0051] This invention constructs a classification feature vector that integrates a first classification value and a second classification value. It employs a weighted feature encoding method and a feature distance calculation method based on Mahalanobis distance, combined with a pre-set type recognition result discrimination model to achieve joint classification processing. This fully utilizes the deep correlation between attribute features and rule features, breaks through the limitations of traditional single feature classification, and significantly improves the accuracy and generalization ability of ontology type recognition. It can accurately distinguish between factual, rule-based, and decision-based ontology, meeting the professional needs of decision knowledge graph construction.

[0052] This invention introduces scientific dimensionality reduction and clustering optimization strategies in the feature processing process, which effectively reduces data dimensionality redundancy, lowers model computational complexity, and improves the processing efficiency of ontology recognition. It can efficiently process ontology recognition tasks of large-scale decision-related text data, providing strong support for the rapid construction of decision knowledge graphs, and has a wide range of application scenarios and practical value.

[0053] The ontology recognition method of this invention does not rely on complex manually labeled data. It can achieve high-precision ontology type recognition through automated text processing, feature extraction and classification processes, reducing the dependence on manual intervention and reducing labor and time costs. At the same time, it has strong flexibility and adaptability, and can adapt to the characteristics of text data in different decision-related fields. It can be widely used in the construction of various decision knowledge graphs and has significant practical value and promotion value.

[0054] In all embodiments of the present invention, the variables involved in all computational expressions or mathematical functions have been dimensionlessized before computation.

[0055] In all embodiments of the present invention, the values ​​of the independent variables in the input of all computational expressions or mathematical functions meet the reasonable requirements of the input range of the computational expressions or mathematical functions, and can ensure that the computational expressions or mathematical functions can be calculated smoothly without violating physical laws or mathematical rules.

[0056] A second aspect of the present invention discloses an ontology recognition device for decision knowledge graphs, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the ontology recognition method for decision-oriented knowledge graphs.

[0057] In a third aspect, the present invention discloses a computer-readable storage medium storing computer instructions, which, when invoked by a computer, are used to execute the ontology recognition method for decision-oriented knowledge graphs.

[0058] In a fourth aspect of this invention, an information data processing terminal is disclosed, which is used to implement the ontology recognition method for decision-oriented knowledge graphs.

[0059] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. An ontology recognition method for decision-making knowledge graphs, characterized in that, include: S1, Ontology knowledge is extracted from textual data in decision-related fields to obtain an ontology text set; the ontology text set includes an attribute text set and a rule text set for each ontology; the attribute text set includes various types of attribute text; the rule text set includes various types of rule text. S2, preprocess the ontology text set to obtain a preprocessed text data set; S3, perform ontology type identification processing on the text data set of each ontology in the preprocessed text data set to obtain the type identification result of each ontology, including: S31, perform a first classification process on the attribute text data set of each ontology in the preprocessed text data set to obtain the corresponding first classification value, including: S311, For the text data sequence of each attribute in the attribute text data set of each ontology in the preprocessed text data set, calculate the corresponding statistical feature vector respectively; The statistical feature vector includes the mean of the text data sequence. ,variance skewness and kurtosis The calculation formula is as follows: , , , in, Let i be the value of the i-th element in the text data sequence. The length of the text data sequence; S312, construct the global feature matrix by using the statistical feature vector of each attribute as a row vector. ,in The number of attributes; S313, regarding the global feature matrix Perform dimensionality reduction projection to obtain the dimensionality reduction matrix: S314, using the row vectors of the reduced dimension matrix as basic elements, perform clustering processing on the basic elements of the reduced dimension matrix to obtain category information; the category information includes several basic elements belonging to the category; S315, take the average value of all basic elements contained in the category information containing the most basic elements as the first classification value corresponding to the ontology; S32, perform a second classification process on the rule text data set in the text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value; S33, perform joint classification processing on the first classification value and the second classification value to obtain the type identification result of the ontology; The types of ontology include factual ontology, rule-based ontology, and decision-based ontology.

2. The ontology recognition method for decision-oriented knowledge graphs as described in claim 1, characterized in that, The preprocessing of the ontology text set to obtain a preprocessed text data set includes: S21, perform text disambiguation processing on the ontology text set to obtain a first text set; S22, perform text alignment processing on the first text set to obtain the second text set; S23, perform text vector transformation on the second text set to obtain a preprocessed text data set; the preprocessed text data set includes the text data set of each ontology; the text data set includes the attribute text data set and the rule text data set.

3. The ontology recognition method for decision-oriented knowledge graphs as described in claim 1, characterized in that, The global feature matrix Dimensionality reduction projection is performed to obtain the dimension reduction matrix, including: S3131, regarding the global feature matrix The covariance matrix is ​​calculated. Its calculation expression is: ; S3132, for Perform feature decomposition and select the top features. The eigenvectors corresponding to the largest eigenvalues ​​form the projection matrix. ; S3133, regarding the global feature matrix The dimensionality reduction matrix is ​​calculated. Its calculation expression is: .

4. The ontology recognition method for decision-oriented knowledge graphs as described in claim 1, characterized in that, The step of performing a second classification process on the rule text data set of each ontology in the preprocessed text data set to obtain the corresponding second classification value includes: S321, for each rule's text data sequence in the rule text data set of each ontology in the preprocessed text data set, perform piecewise linear fitting to obtain the corresponding fitting matrix: S322, Perform metric fusion calculation on each fitted matrix to obtain a set of metric values; S323, perform clustering on the set of metric values ​​of all fitted matrices to obtain category information; each category information includes several elements, wherein the elements are the set of metric values; S324, determine the category information with the largest number of elements, which is the classification category information; S325, calculate the mean of all the measurement value sets contained in the classification category information, which is the second classification value corresponding to the ontology.

5. The ontology recognition method for decision-oriented knowledge graphs as described in claim 1, characterized in that, The joint classification process of the first classification value and the second classification value to obtain the type identification result of the ontology includes: S331, Based on the first classification value and the second classification value, a classification feature vector is constructed; S332, Perform feature encoding processing on the classification feature vector to obtain the encoded vector; S333, perform feature distance calculation on the encoded vector to obtain a distance value; S334, based on a preset ontology type recognition result discrimination model, determine the range of values ​​to which the distance value belongs, and determine the type information corresponding to the range of values, which is the ontology type recognition result.

6. An ontology recognition device for decision-making knowledge graphs, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the ontology recognition method for decision-oriented knowledge graphs as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when invoked by a computer, are used to execute the ontology recognition method for decision-oriented knowledge graphs as described in any one of claims 1 to 5.

8. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the ontology recognition method for decision-oriented knowledge graphs as described in any one of claims 1 to 5.

Images