A knowledge reasoning method based on online graph processing technology for knowledge graph

Through the knowledge reasoning method of online graph processing technology, combined with the upper volume of information dimensions and topological dimensions, and combined with the shared network, the multi-dimensional analysis requirements in the knowledge graph are solved, and the deep multi-dimensional and multi-level knowledge mining of the knowledge graph is realized, which meets the Users' complex query and analysis needs fill in the shortcomings of the existing technology.

CN117972111BActive Publication Date: 2024-11-19SICHUAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410209061.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-26
Publication Date
2024-11-19
Estimated Expiration
2044-02-26

AI Technical Summary

Technical Problem

The existing Graph OLAP technology is difficult to meet the multidimensional analysis needs in information networks, especially in the analysis of inter-entities, which lacks the combination of topological structure hierarchical relationships, making it difficult to carry out effective multidimensional and multi-level analysis, especially in the scenario of knowledge graphs. Down.

Method used

Using knowledge reasoning method based on online graph processing technology, through the upper volume of information dimensions and topological dimensions, combined with shared networks, multi-dimensional and multi-level analysis of knowledge graphs is realized, and triple data is stored and processed by MySQL database to perform nodes and Merge edges, calculate the probability values ​​and topological structure of entities, and mine deep information and knowledge.

Benefits of technology

It realizes in-depth query and analysis of the knowledge graph, meets the complex needs of users, provides management of multi-source heterogeneous data and in-depth multi-dimensional and multi-level knowledge mining, fills the shortcomings of insufficient analysis capabilities of existing technology, and is a project equipment Design provides deeper insights.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117972111B_ABST
    Figure CN117972111B_ABST
Patent Text Reader

Abstract

The present invention discloses a knowledge reasoning method based on online graph processing technology for knowledge graphs. By combining data fusion and storage methods, and utilizing information dimension roll-up, topology dimension roll-up, and shared networks, the present invention solves the problem that a large amount of multimodal data such as text, operation data, and calculation data are generated in the process of project R&D and design, and lacks unified storage and automatic association, resulting in a lack of knowledge mining and exploration discovery, as well as the defect of insufficient analysis capability of existing OLGP technology. The present invention realizes the management of multi-source heterogeneous data and the deep multi-dimensional and multi-level presentation and mining and utilization of knowledge, meets the computing needs of users for complex query and analysis of large-scale multimodal data, provides deeper insights and support for the design, safety, and maintenance of project equipment, and gives new vitality and application value to online graph processing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to a knowledge reasoning method oriented to a knowledge graph based on online graph processing technology. Background Art

[0002] With the rapid development of information technology, a large amount of data with massive entities and associations has emerged. Graph data models have powerful expressive capabilities and can directly and effectively express data with associated semantics. Data built using graph models are called information networks, which can model data in a large number of fields in the real world. For example, in the field of social networks, users and the relationships between users can be constructed as information networks to mine potential associations between users and predict user behavior.

[0003] Traditional online analytical processing (OLAP) technology based on relational data is an effective decision support tool that can provide views of data from different dimensions and granularities, aggregate and analyze multidimensional data to support effective decision making. Due to the diverse types of entities in information networks and the intricate relationships between entities, the traditional OLAP model cannot effectively meet the multidimensional analysis needs of information networks. To this end, the concept of Graph OLAP is proposed, which introduces online analytical processing technology into information network analysis to support multidimensional analysis of graph data, transform the original information network in different dimensions and granularities, and mine hidden relationships between different entities. The overall research trend of multidimensional analysis of information networks is to develop from isomorphic information networks to heterogeneous information networks, and from the analysis of entities to the analysis of relationships.

[0004] The existing work related to Graph OLAP mainly divides the dimensions of nodes and edges into information dimension and topology dimension and designs corresponding OLAP operations, systematically defines the relevant concepts of multidimensional analysis of graph data, and proposes a binary star model based on this, realizing the aggregation algorithm of information dimension and topology dimension. On the basis of information dimension and topology dimension, the entity dimension is introduced. The path-dimensional graph cube model is proposed, which guides the aggregation of network structure with meta-path, constructs entity hypercube and carries out calculation based on it. The shortcomings of the existing online graph processing methods are that the analysis of relationships and attributes between entities is relatively simple, which is difficult to meet the complex query requirements issued by users, and the analysis of relationships between entities does not combine the hierarchical relationship between topological structures, so it is difficult to apply it to the scenario of project composition network, and to carry out effective multi-dimensional and multi-level analysis and processing of the knowledge graph formed by documents.

[0005] The design of digital project equipment requires a large amount of existing design knowledge and experience, which are generally contained in rich documentation. How to conduct multi-dimensional analysis and processing of digital documentation and integrate it into the R&D and design process of digital project equipment is an important research issue.

[0006] The scenario of the knowledge reasoning algorithm for the knowledge graph in the present invention is the project equipment knowledge graph stored in the form of distributed data formed after the data in the project file is extracted, fused, and stored in several steps. It is a heterogeneous information network containing five types of nodes and different attributes. The knowledge graph includes five types of nodes from high to low conceptual levels of equipment, components, sub-components, parts, and materials. The edges between the nodes are all directed edges, representing the composition relationship between the nodes, such as a part is composed of a certain material. There are no connecting edges between nodes of the same type, only edges between nodes of adjacent levels, and the nodes at the higher level point to the nodes at the lower level. Among them, the nodes of the part type include attribute size values, such as 167.46mm x117.22mm x35.33mm, and the nodes of the material type include attribute chemical composition content, such as 0.04~0.045. Summary of the invention

[0007] The purpose of the present invention is to provide a knowledge reasoning method based on online graph processing technology for knowledge graphs in order to solve the above problems. The present invention realizes the knowledge fusion and storage modeling of the triple data of the project after knowledge extraction, and then applies the OLGP online graph processing technology to the project composition structure data set to realize the analysis and calculation processing of the knowledge graph at different dimensions and levels, thereby mining deep information and knowledge, providing valuable data mining and insights, and meeting users' deep query and analysis needs for project equipment knowledge graphs.

[0008] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0009] A knowledge reasoning method for knowledge graph based on online graph processing technology includes the following steps:

[0010] Step 1: The text information to be processed is converted into triples by word segmentation, stop word removal, and part-of-speech tagging, and then data fusion is performed on each triple, and the data is corrected;

[0011] Step 2, create a table in the MySQL database, insert data statements, and store the processed data in the MySQL database in the form of relational data;

[0012] Step 3: In the relational table of the MySQL database, the project is used as the dimension and the equipment composition structure diagram of the project is used as the measurement value; the equipment knowledge graphs of different projects are subjected to information dimension roll-up calculation. The graph merging process includes node merging and edge merging, that is, all nodes and edges in the equipment composition structure graphs of different projects are merged into one graph to obtain an information dimension aggregation graph. The entity and relationship information of the graph are changed while the topological structure remains unchanged; in the merged tree structure, the probability value attribute of each entity is calculated, which reflects the frequency of the equipment appearing in different projects;

[0013] Step 4: Roll up the nodes of the part type in the knowledge graph in the MySQL database according to their topological dimension attribute-size value to form a number of part nodes in a high topological dimension; at the same time, roll up the nodes of the material type associated with the part nodes of different size ranges according to their topological dimension attribute-chemical composition content to form a number of high-dimensional material nodes in a high topological dimension;

[0014] Step 5: For the nodes of the four conceptual hierarchical types of equipment, components, sub-components, and parts in the knowledge graph of the MySQL database, four different shared networks are calculated respectively. Each shared network describes the relationship between the shared lower-level entities among all entities at the same level, and a data structure graph based on the network structure, path, and relationship between entities is obtained;

[0015] Step 6, based on the shared networks of different levels calculated in step 5, combined with the topological dimension attributes of the nodes in the network, all nodes of the same type in the shared network whose topological dimension attribute values ​​are in the same range are merged into high-dimensional nodes, and the original shared network is topologically rolled up to obtain a new shared network composed of high-dimensional nodes with a new topological structure.

[0016] A further solution is that in step 1, data error correction includes the following two categories:

[0017] Rule-based error correction method: Scan and verify data based on a predefined set of rules. Each rule represents a constraint on data correctness. When data violates these constraints, the data is considered an error and traced back to the source, supplemented by manual modification.

[0018] Error correction method based on graph pattern matching: Matching is performed according to predefined graph patterns. The predefined graph patterns represent the normal structure in the knowledge base. The parts that do not match these predefined patterns are searched in the knowledge base. If a structure that does not match the predefined pattern is detected, it is regarded as an error, and data tracing is performed, supplemented by manual modification.

[0019] A further solution is that in step 2, the relational data includes: entity table, entity attribute table, relationship table, merge record table and high-dimensional table; the entity table and entity attribute table jointly define the key entities of the project and their related attributes; the dynamic design of the entity attribute table allows engineers to add or adjust attributes for entities at any time without making any structural modifications; the relationship table is used to mine and store complex relationships and their attributes between project components; the merge record table records which projects are merged to serve OLGP reasoning; the high-dimensional table records the information dimension, and the topology dimension rolls up related information and attributes to serve OLGP reasoning.

[0020] A further solution is that in step 2, the entity table includes 3 columns, wherein the first column is the primary key, storing the unique ID value of each device, the second column is the type of the entity, and the third column stores the name of the entity; the entity attribute table includes 5 columns, wherein the first column is the ID value of the primary key attribute, the second column is the entity ID value to which the attribute belongs, the third column is the attribute name, the fourth column is the attribute value, and the fifth column is the project to which the attribute belongs; the relationship table includes 6 columns, wherein the first column is the unique ID value of the relationship, the second and third columns are the ID values ​​of the head and tail entities of the relationship respectively, the fourth column is the relationship type, and the fifth column is the relationship name. The fifth column is the weight, and the sixth column is the project to which the relationship belongs; the merged record table includes 3 columns, the first column is the unique ID value of the record generated during each information dimension merge, the second column is the several projects merged, and the third column is the name corresponding to the merge; the high-dimensional table includes 4 columns, the first column is the unique ID value corresponding to the node in the graph after the information dimension is rolled up, the second column is the ID value of the node before the corresponding merge, the third column is the ID value of the record generated during the information dimension merge, and the fourth column stores the main results generated by the information dimension rollup - the probability value of each project equipment appearing in different projects.

[0021] A further solution is that in step 3, the specific implementation process is as follows: by taking out the data in the relational table of the MySQL database and performing calculations to complete the knowledge reasoning based on the information dimension roll-up and facing the knowledge graph, the results are stored in the relational table, the high-dimensional table, and the merged record table. In the relational table of the MySQL database, all the device nodes included in the project map to be merged are queried respectively, and the ID values ​​of all the device nodes queried in the relational table are stored in the data set of the corresponding project respectively, and the two sets are merged into a list and all the node ID values ​​in the list are counted. The number of occurrences divided by the number of projects is the probability value of the device node corresponding to the node ID value appearing in different projects, and the calculated probability values ​​are stored one by one in the corresponding table items in the high-dimensional table. The probability value obtained by reasoning meets the user's analysis needs for the utilization of different devices. The information in all the relationship tables of the two projects is taken out and merged into a set, and all the information in the set is inserted into a specific table item in the database to complete the information dimension roll-up, so as to obtain a view of the project composition structure from a macro-global perspective. This process includes both node merging and edge merging, and the topological structure of the graph does not change, allowing users to analyze the overall equipment composition structure of the project more comprehensively and comprehensively, and realize knowledge reasoning from different angles and dimensions for the knowledge graph. Each time the information dimension is rolled up, a row of data queried from the entity table needs to be inserted into the merge record table to record the projects involved in the roll-up and the name-related information corresponding to the merge. Through this process, the information of nodes and relationships in graphs of different dimensions is combined, and a more systematic and comprehensive analysis is achieved that is superior to the existing OLGP method.

[0022] A further solution is that in step 4, the specific implementation process is as follows: by performing calculations on the data in the entity attribute table and the relationship table to complete the knowledge reasoning based on the topological dimension roll-up oriented knowledge graph, the results are stored in the relationship table. First, the chemical composition attribute values ​​of all material nodes are queried from the entity attribute table in the MySQL database and added separately, and the mapping relationship between the results and the ID values ​​of each material node is stored in the dictionary, and then the total chemical composition content values ​​of each material node are sorted from high to low; according to the above dictionary and the attribute value division rules of different high-dimensional material nodes, the ID values ​​of the low-dimensional nodes corresponding to different high-dimensional nodes before rolling up are stored in different lists in sequence; for the nodes of the part type, all their size values ​​are also queried from the entity attribute table in the MySQL database and the volume is calculated accordingly, and the mapping relationship between the results and the ID value is stored in the dictionary, and the node ID values ​​within different size value ranges are stored in different lists after the results are sorted. This is the process of merging different nodes into different high-dimensional nodes according to the range of their attribute values ​​in the topological dimension roll-up. In this process, the topological structure of the graph changes greatly, and the object of analysis is transformed from a specialized node to a generalized node. After the nodes are merged, the data corresponding to the high-dimensional nodes are inserted into the entity table; finally, the ID value of the part node before the merger corresponding to the high-dimensional part node that is associated with the high-dimensional material node with a total chemical content in different intervals is queried from the relationship table, and the result is matched with the aforementioned list of different low-dimensional material node ID values, and the calculated composition association relationship between the high-dimensional nodes at different levels after topological dimension rollup is inserted into the relationship table. Through this operation, the deep association relationship between material nodes with different total chemical content and part nodes with different size values ​​is found, the topological association between nodes at different levels is opened up, and the hidden knowledge between different types of nodes is obtained. This method takes into account the topological structure of the knowledge graph itself, path information, and attribute information of the entity that the OLGP method did not involve in the past, thereby mining deeper knowledge in the graph.

[0023] A further solution is that in step 5, the specific implementation process is as follows: the composition relationship and hidden sharing relationship between nodes in the knowledge graph are used to carry out computational reasoning, and these knowledge are stored in the relationship table. For any node, all nodes participating in its composition structure are queried from the relationship table in the MySQL database, that is, all tail entity IDs whose head entity ID is the node are queried in the relationship table, and then all nodes of the same level that have a sharing relationship with the node are further queried in the relationship table according to the query result, that is, the tail entity in the query relationship table includes all head entity ID values ​​of the ID value in the aforementioned query result, and all nodes with a sharing relationship with the node are determined through two queries. The weight value of the relationship is the absolute value of the difference between the attribute values ​​of the two nodes divided by the normalization parameter, and finally the calculated result is inserted into the relationship table. Repeat the above process for all nodes at any level, and a shared network that reflects the sharing relationship between nodes at any level is obtained, which realizes the analysis and reasoning of deep sharing connections between nodes of the same type, and meets the user's query and analysis needs for complex composition relationships between nodes in the graph. The advantage of this method over the traditional OLGP method is that it combines the path information, structural information and shared relationships between entities at different levels of the knowledge graph itself, and infers the hidden associations between nodes from a new perspective and dimension.

[0024] A further solution is that in step 6, the specific implementation process is as follows: on the basis of the high-dimensional nodes obtained by topological dimension merging, the sharing relationship between nodes is further combined to obtain a deep-level association relationship between high-dimensional nodes and high-dimensional nodes. The ID values ​​of several original low-dimensional nodes before merging corresponding to each part-level high-dimensional node are stored as attribute values ​​in the corresponding table item of the entity attribute table, and the ID values ​​are stored in the list in step 4. This operation retrieves the high-dimensional node information obtained by rolling up the topological dimension. The head entity and tail entity of all edges with a relationship type of shared material in the relationship table are queried, that is, the head and tail low-dimensional node IDs of all edges in the shared network stored in the relationship table in step 5 are retrieved, and this operation retrieves the sharing relationship between nodes. According to the head and tail low-dimensional node IDs of each edge in the obtained shared network, the corresponding high-dimensional node ID value is queried from the entity attribute table, so as to obtain each pair of part-level high-dimensional nodes with a shared association relationship. This step combines the high-dimensional entity information and the shared network obtained by topological dimension merging. Finally, the calculated high-dimensional node ID pair with implicit shared association relationship is stored in the relationship table for user query and analysis. Through this method, we can obtain deeper associations and hidden knowledge between nodes combined with attribute value information. Compared with the existing OLGP method, this method makes more comprehensive and thorough use of the topological structure information, path information, relationship information and attributes between entities in the knowledge graph, combines the topological dimension roll-up and the hidden shared relationships in the network, and more deeply mines the deep knowledge in the graph from multiple angles and levels, providing users with in-depth data insights.

[0025] The beneficial effects of the present invention are:

[0026] (1) Graph OLAP, an online graph processing technology for knowledge graphs, was applied to the actual scenario of a certain research institute's project equipment composition structure network. Combined with data fusion and storage methods, information dimension roll-up, topology dimension roll-up, and shared network were used to solve the problem of a large amount of multimodal data such as text, operation data, and calculation data generated during the project R&D and design process. The lack of unified storage and automatic association led to the lack of knowledge mining and exploration, as well as the defect of insufficient analysis capabilities of existing OLGP technology. The management of multi-source heterogeneous data and the deep multi-dimensional and multi-level presentation and mining of knowledge were realized, meeting users' computing needs for complex query and analysis of large-scale multimodal data, providing deeper insights and support for project equipment design, safety, and maintenance, and giving online graph processing technology new vitality and application value.

[0027] (2) Make full use of the attributes of entities in the knowledge graph, the information of the relationship between entities, the topological structure of heterogeneous information networks, and path-related information to carry out calculations, combine the component relationships in the device knowledge graph to build a shared network, and use the topological structure information of the network based on the topological dimension attributes to carry out mining and exploration of hidden associations between deep nodes, filling the gaps in previous research methods.

[0028] (3) While expanding the operations and methods of Graph OLAP analysis and processing, it also takes into account the security and privacy of the data. It combines several algorithms based on information dimension, topology dimension, shared network, and topology dimension roll-up based on shared network in this scenario, and comprehensively and comprehensively mines and analyzes the potential associations and knowledge in the device knowledge graph from different angles and dimensions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a specific flow chart of the present invention.

[0031] Figure 2 It is a schematic diagram of the storage structure of the knowledge graph of the present invention in the MySQL relational database.

[0032] Figure 3 It is a schematic diagram of knowledge reasoning based on the information dimension and oriented to the knowledge graph.

[0033] Figure 4 It is a schematic diagram of knowledge reasoning based on the topological dimension and oriented to the knowledge graph.

[0034] Figure 5 It is a schematic diagram of knowledge reasoning based on a shared network oriented knowledge graph.

[0035] Figure 6 It is a schematic diagram of knowledge reasoning oriented to the knowledge graph with comprehensive topological dimension roll-up and shared relationships. DETAILED DESCRIPTION

[0036] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0037] The design of digital project equipment requires a lot of existing design knowledge and experience, which are generally contained in rich document materials. How to carry out multi-dimensional analysis and processing of digital document materials and integrate them into the R&D and design process of digital project equipment is an important research issue. For the extracted text file content in the R&D and design process, the graph structure knowledge base of the information network is used for modeling, storage, operation and analysis. The topological dimension and information dimension of heterogeneous information networks are introduced to perform dynamic graph roll-up and graph drill-down operations. A knowledge reasoning and analysis method for real-world domain knowledge graphs that supports cross-level concept associations is proposed. Among them, roll-up refers to aggregating data upward to a coarser level along a certain dimension to obtain summary data from detailed data, while drill-down is the inverse operation of roll-up. The information dimension roll-up organizes graph snapshots according to different angles and controls which dimensions these snapshots are observed from. In the topological dimension roll-up, the merge is performed inside the subgraph, which narrows the focus to a set of generalized objects, summarizes the original network and hides specific details.

[0038] The data in the document is extracted, fused and stored in several steps to form a heterogeneous information network stored in the form of relational data. It is a data set for the knowledge reasoning algorithm based on the knowledge graph. The knowledge graph includes five types of nodes from high to low concept levels: equipment, components, sub-components, parts, and materials. The edges between the nodes are all directed edges, representing the composition relationship between the nodes, such as a part is composed of a certain material. There are no edges between nodes of the same type, only edges between nodes of adjacent levels, and the high-level nodes point to the low-level nodes. Among them, the nodes of the part type include attribute size values, such as 167.46mm x 117.22mm x35.33mm, and the nodes of the material type include attribute chemical composition content, such as 0.04~0.045.

[0039] The present invention is based on the knowledge reasoning method for knowledge graph based on online graph processing technology, and extracts the large-scale irregular data included in the documents in the research institute, merges and stores it in the form of relational data into the MySQL database. The knowledge graph data is carried out multi-dimensional, multi-level, and deep calculation and analysis. Combined with the complex and diverse query needs of users, taking into account the attributes of entities, the relationship between entities, the topological structure information of the information network itself, and the path information, an efficient knowledge reasoning method based on knowledge graph is proposed, including information dimension roll-up, topological dimension roll-up, shared network construction, and topological dimension roll-up based on shared network. The method of the present invention is implemented by Python programming. Among them, the information dimension roll-up rolls up the graph according to the information dimension attribute of the graph-the project to which it belongs, and the tree-like device graphs corresponding to different projects are merged to calculate the overall device network and the probability value of each device appearing. After analysis, the importance of different components and the user's interest level are obtained. The topological dimension roll-up is based on the attribute value of the node, combined with the topological structure of the network for analysis to dig out the deep connection between nodes at different levels and different attribute values. The construction of the shared network uses the information of the composition relationship between nodes in the network to calculate the shared relationship network of nodes at the same level. The topological dimension rollup based on the shared network combines the topology and path structure of the network, the relationship information between entities, and fully considers the characteristics of the network and different types of nodes. The topological dimension rollup is further expanded for the shared network, deeply excavating the implicit knowledge of the network and the hidden associations between entities, showing rich and in-depth data. The results generated after multi-dimensional and multi-level analysis and processing are finally presented to users in the form of knowledge graphs to meet their diverse and complex query and analysis needs.

[0040] The knowledge reasoning method based on online graph processing technology for knowledge graph of the present invention includes the following steps:

[0041] (1) After knowledge extraction, the project-related triple data needs to be fused and error correction is performed on the project-related data.

[0042] This step includes: collecting text information in the field of the project, converting the text information to be processed into triples by word segmentation, removing stop words, and part-of-speech tagging, then fusing data for each triple, and correcting the data;

[0043] The knowledge extraction format is in the form of triples, such as {steam generator, component, shell}, {shell, component composition, gasket}, {gasket, diameter, 100 mm}. The data error correction is divided into the following two categories:

[0044] Rule-based error correction method: Verify and correct non-conformities in the data according to a predefined set of rules. The core is to use a clear, predefined set of rules to scan and verify the data. When the data violates these rules, the method regards the data as errors and takes relevant measures. Specifically, this method first defines several rules and constructs a rule set: for example, the subject of all triples with components as components should be a specific device or part, and the object is another specific component. For the triple {steam generator, component, shell}, the subject is the steam generator (equipment) and the object is the shell (component). Therefore, this triple satisfies the above rules. When traversing the entire data set and matching each triple with the predefined rules, if a rule-violating triple similar to {steam generator, component, 100mm} appears, the system marks the above triple as an error group and performs data traceability, supplemented by manual modification.

[0045] Error correction method based on graph pattern matching: Matching is performed according to predefined graph patterns. The predefined graph patterns represent the normal structure in the knowledge base. The key to this method is to define a series of graph patterns that represent the normal structure in the knowledge base. The system searches the knowledge base for parts that do not match these predefined patterns. Once a structure that does not match the predefined pattern is detected, it is considered an error. Specifically, this method first predefines a graph pattern, such as the hierarchical structure between entities of a project, from top to bottom: equipment-component-part. The system then starts matching. If {steam generator, component, gasket} appears, it violates the expected graph hierarchy because the parts should not directly form the equipment. Therefore, the system marks this mismatched structure as an error group, and also performs data traceability, supplemented by manual modification.

[0046] (2) Create a table in the MySQL database, insert data statements, and store the processed data in the MySQL database in the form of relational data; realize the reasonable storage and modeling of the project equipment knowledge graph to serve the execution of subsequent algorithms. The knowledge graph data stored in the database in this step is logically a heterogeneous information network, including nodes of five conceptual hierarchical types: equipment, components, sub-components, parts, and materials. The relationship between nodes is a directed edge, representing the composition relationship between nodes. The part node includes the attribute size value, and the material node includes the attribute chemical composition content.

[0047] Relational data includes: Entity table (Entities) and entity attribute table (EntityAttributes) together define the key entities of the project and their related attributes. The dynamic design of the entity attribute table allows engineers to add or adjust attributes for entities at any time without making any structural modifications. The relationship table (Relations) is used to deeply explore and store the complex relationships and attributes between project components. The merge record table (MergeRecords) mainly records which projects are merged to serve OLGP reasoning. The high-dimensional table (EntityAppearance) records the relevant information and attributes of the information dimension rollup to serve OLGP reasoning.

[0048] The entity table includes the columns EntityID, EntityType, and EntityName. The first column is the primary key, which stores the unique ID value of each device. The second column is the type of the entity, which is set to the five conceptual hierarchical types of equipment, components, subcomponents, parts, and materials. The third column stores the name of the entity, such as flange, steam generator, etc. The entity attribute table includes the columns AttributeID, EntityID, AttributeName, AttributeValue, and NuclearStation. The first column is the ID value of the primary key attribute, the second column is the entity ID value to which the attribute belongs, the third column is the attribute name, such as the size value, the fourth column is the attribute value, such as 167.46mm x 117.22mm x 35.33mm, and the fifth column is the project to which the attribute belongs. The relationship table includes the columns RelationID, StartEntityID, EndEntityID, RelationType, Weight, and NuclearStation. The first column is the unique ID value (primary key) of the relationship. The second and third columns are the ID values ​​of the head and tail entities of the relationship, respectively. The fourth column is the relationship type. For example, the relationship type between the part entity and the material entity is material composition. The fifth column is the weight, and the sixth column is the project to which the relationship belongs. The merged record table includes the columns MergeID, MergedStations, and MergedName. The first column is the unique ID value (primary key) of the record generated during each information dimension merge process. The second column is the merged projects, and the third column is the name corresponding to the merge. The high-dimensional table includes the columns EntityAppearance, EntityID, MergeID, and AppearanceProbability. The first column is the unique ID value (primary key) corresponding to the node in the graph after the information dimension is rolled up. The second column is the ID value of the node before the corresponding merge. The third column is the ID value of the record generated during the information dimension merge process. The fourth column stores the main result generated by the information dimension rollup - the probability value of each project equipment appearing in different projects. See the storage structure diagram of the knowledge graph in the MySQL relational database for details. Figure 2 .

[0049] (3) Use the information dimension attributes included in the graph to carry out multi-dimensional analysis and calculation on the project composition structure knowledge graph to meet the user's query and analysis needs for the global composition structure of the project and the importance of the nodes.

[0050] It is an important requirement to analyze the composition structure of different projects and the overall composition structure and calculate which parts of them appear more frequently in different projects. First, according to the composition structure tree knowledge graph of different projects stored in the relational table in step 2, the project is used as the dimension and the project equipment composition structure diagram is used as the metric value. The equipment knowledge graphs of different projects are subjected to the information dimension roll-up calculation method. The graph merging process includes the merging of nodes and the merging of edges. This process is equivalent to merging all nodes and edges in the equipment composition structure graphs of different projects into one graph to obtain an information dimension aggregation graph. The entity and relationship information of the graph change while the topological structure does not change. In the merged tree structure, the probability value attribute is calculated for each entity. This attribute reflects the frequency of the equipment appearing in different projects, indirectly reflects the user's interest in it and the importance of the equipment node, and provides important reference value for the user's query needs. The higher-dimensional information dimension aggregation graph calculated after the information dimension roll-up shows the comprehensive equipment composition structure diagram of the project, providing users with a more macro perspective to observe and analyze the composition structure of the project. See the schematic diagram of the knowledge reasoning method based on information dimension attributes for knowledge graphs. Figure 3 .

[0051] The specific implementation process is: by taking out the data from the relational table in the MySQL database and performing calculations to complete the knowledge reasoning based on the information dimension roll-up oriented knowledge graph, the results are stored in the relational table, high-dimensional table, and merged record table. In step 2, the relational table stored in the MySQL database is queried for all the device nodes included in the project graph to be merged, assuming that they are project A and project B respectively. The ID values ​​of all device nodes appearing in project A and project B obtained from the query in the relational table are stored in sets A and B respectively. The two sets are merged into a list and all the node ID values ​​in the list are counted. The number of occurrences divided by the number of projects is the probability value of the device node corresponding to the node ID value appearing in different projects. The calculated probability values ​​are stored one by one in the corresponding table items in the high-dimensional table. The probability value obtained by reasoning meets the user's analysis needs for the utilization of different devices. At the same time, the information in all the relationship tables of Project A and Project B is taken out and merged into a set, and all the information in the set is inserted into a specific table item in the database to complete the information dimension roll-up, so as to obtain a view of the project composition structure from a macro-global perspective. This process includes both node merging and edge merging, and the topological structure of the graph does not change, allowing users to analyze the overall equipment composition structure of the project more comprehensively and comprehensively, and realize knowledge reasoning from different angles and dimensions for the knowledge graph. Each time the information dimension is rolled up, a row of data queried from the entity table must be inserted into the merge record table to record the projects involved in the roll-up and the name-related information corresponding to the merge. Through this process, the information of nodes and relationships in graphs of different dimensions is combined, and a more systematic and comprehensive analysis is achieved that is superior to the existing OLGP method.

[0052] (4) By utilizing the topological dimension attributes included in the knowledge graph, combined with the attributes of the entity and the topological structure of the network itself, knowledge reasoning based on the topological dimension roll-up and oriented to the knowledge graph is carried out in response to users' query needs for hidden deep-level relationships between different types of entities at different levels.

[0053] Users are usually interested in the attributes of different types of equipment. Entities are set as five conceptual hierarchical types: equipment, components, sub-components, parts, and materials. High-level entities are composed of low-level entities. Parts contain size attributes, and materials contain chemical composition attributes. In response to users' query and analysis needs for different node attributes, combined with the network's topological structure information and path information, a knowledge reasoning algorithm based on topological dimension rollup and oriented to knowledge graphs is proposed. The nodes of the part type stored in the database in step 2 are rolled up according to their topological dimension attribute-size value. Specifically, all part nodes are divided into several types according to the range of size values, and several low-dimensional nodes included in each range are merged to form a part node in a high topological dimension. At the same time, the nodes of the material type associated with the part nodes of different size ranges are rolled up according to their topological dimension attribute-chemical composition content. Specifically, all material nodes are divided into several types according to the range of total chemical composition content, and several low-dimensional material nodes included in each range are merged to form a high-dimensional material node in a high topological dimension. In this process, the size value of the part node and the total chemical content value of the material node are used as dimensions, the range of the size value and the total chemical content value is used as the dimension level, and the topological structure of the equipment knowledge graph after the topological dimension rollup is used as the measurement. For the topological dimension rollup, the merge is performed inside the network. This process transforms the focus of analysis from specialized objects to a set of generalized object sets, and the topological structure in the topological dimension aggregation graph formed changes significantly. At this time, according to the topological dimension aggregation graph composed of high-dimensional nodes formed after the topological dimension rollup, combined with the topological structure and path information of the original equipment graph, the deep-level associations and hidden knowledge between parts of different sizes and materials with different total chemical content values ​​can be obtained, the topological dimension associations between nodes at different levels are opened up, and the hidden associations and potential knowledge between nodes of different types at different levels are obtained. By analogy, the composition relationship between other different types of equipment at adjacent levels can be obtained in the same way. For example, it can be calculated that the common attribute of the part nodes associated with the material nodes with a total chemical content range of (0.15-0.20) is that the size is large, while small-sized parts usually have a composition relationship with materials with a total chemical content range of (0.20-0.24). Figure 4 .

[0054] The specific implementation process is: by expanding the calculation for the data in the entity attribute table and the relationship table to complete the knowledge reasoning based on the topological dimension roll-up oriented knowledge graph, the result is saved in the relationship table. Take the calculation of the association relationship between the node of the material type and the node of the part type as an example. First, query the chemical composition attribute values ​​of all material nodes from the entity attribute table of the database in step 2 and add them separately, store the mapping relationship between the results and the ID values ​​of each material node in the dictionary, and then sort the total chemical composition content values ​​of each material node from high to low. According to the above dictionary and the attribute value division rules of different high-dimensional material nodes, the ID values ​​of the low-dimensional nodes corresponding to different high-dimensional nodes before rolling up are stored in different lists in turn. For the nodes of the part type, all their size values ​​are also queried from the entity attribute table in the database of step 2 and the volume is calculated accordingly. The mapping relationship between the result and the ID value is stored in the dictionary, and the node ID values ​​within different size value ranges are stored in different lists after sorting the results. This is the process of merging different nodes into different high-dimensional nodes according to the range of their attribute values ​​in the topological dimension rollup. In this process, the topological structure of the graph changes greatly, and the object of analysis is transformed from specialized nodes to generalized nodes. After the node is merged, the data corresponding to the high-dimensional node is inserted into the entity table. Finally, the ID value of the part node before merging corresponding to the high-dimensional part node that has an association relationship with the high-dimensional material node with a total chemical content in different intervals is queried from the relationship table, and the result is matched with the aforementioned list of different low-dimensional material node ID values, and the calculated composition association relationship between the high-dimensional nodes of different levels after the topological dimension rollup is inserted into the relationship table. Through this operation, the deep association relationship between the material nodes with different total chemical content and the part nodes with different size values ​​is found, the topological association between nodes at different levels is opened up, and the hidden knowledge between different types of nodes is obtained. This method takes into account the topological structure of the knowledge graph itself, the path information, and the attribute information of the entity that the OLGP method did not involve in the past, thereby mining deeper knowledge in the graph.

[0055] (5) In order to make more full and thorough use of the information related to the structure, path, and relationship between entities contained in the information network itself, and to meet the user's query and analysis needs for the complex component relationships in the project equipment information network, a knowledge reasoning algorithm based on a shared network oriented knowledge graph is proposed.

[0056] The shared network is a homogeneous information network. All nodes in the network are of the same type. The edges between nodes are undirected edges, which describe the relationship between two nodes of the same type that they share the same lower-level adjacent node. In other words, if two nodes of the same type have a composition relationship with the same lower-level node, then there will be an edge between the two nodes of the same type that reflects the shared relationship. The attributes of the nodes are consistent with the original knowledge graph. The calculation of the edge weight is related to the similarity of the head and tail nodes of the edge. The higher the similarity between the two nodes, the greater the weight of the edge, and the closer the attributes are, the higher the similarity between the two nodes. For the nodes of the four conceptual hierarchical types of equipment, components, sub-components, and parts in the knowledge graph stored in the database in step 2, four different shared networks will be calculated separately through the method of this step. Each shared network describes the relationship between shared low-level entities between all entities at the same level, providing users with valuable deep data insights based on the structure, path, and relationship between entities of the network. See the schematic diagram of the knowledge reasoning method based on the shared network for the knowledge graph. Figure 5 .

[0057] The specific implementation process is: use the composition relationship and hidden sharing relationship between nodes in the knowledge graph to carry out computational reasoning, and these knowledge are stored in the relationship table. For each node, query all nodes participating in its composition structure from the relationship table stored in the database in step 2, that is, query all tail entity IDs whose head entity ID is the node in the relationship table, and then further query all nodes of the same level that have a sharing relationship with the node in the relationship table according to the query results, that is, query the tail entity in the relationship table including all head entity ID values ​​of the ID value in the above query results. Through two queries, all nodes that have a sharing relationship with the node are determined, and the weight value of the relationship is the absolute value of the difference between the attribute values ​​of the two nodes divided by the normalization parameter. Finally, the calculated result is inserted into the relationship table. Repeat the above process for all nodes at any level, and a shared network that reflects the sharing relationship between nodes at any level is obtained, which realizes the analysis and reasoning of deep sharing connections between nodes of the same type, and meets the user's query and analysis needs for complex composition relationships between nodes in the graph. The advantage of this method over the traditional OLGP method is that it combines the path information, structural information and shared relationships between entities at different levels of the knowledge graph itself, and infers the hidden associations between nodes from a new perspective and dimension.

[0058] (6) Integrate the knowledge reasoning algorithm based on topological dimension roll-up for knowledge graph and the knowledge reasoning algorithm based on shared network for knowledge graph, utilize the topological dimension attributes, topological structure and path information included in the network as well as the attributes of entities and the relationship information between entities, perform topological dimension roll-up on the information network based on the shared network, and mine and calculate the hidden associations and knowledge in the project equipment knowledge graph.

[0059] This step includes: on the basis of the knowledge reasoning algorithm based on the knowledge graph oriented to the shared network in step 5, combined with the relationship and attribute information between entities at a deeper level of the information network, the implicit association relationship and knowledge between the device nodes with the shared relationship are obtained, and a topological dimension roll-up method based on the shared network is proposed. On the basis of the shared networks of different levels calculated in step 5, combined with the topological dimension attributes of the nodes in the network, all nodes with topological dimension attribute values ​​in the same range of the same type of nodes in the shared network are merged into high-dimensional nodes, and the original shared network is topologically rolled up to obtain a new shared network composed of high-dimensional nodes including a new topological structure. The weight of the edge in the rolled-up network is the weighted average of the weights of the corresponding edges in the original network. The difference between the topological dimension roll-up in this method and the topological dimension roll-up in step 4 is that the topological dimension roll-up in step 4 rolls up the nodes of adjacent levels at the same time to calculate the association relationship between nodes of different types, while the topological dimension roll-up in this method only performs topological dimension roll-up on the nodes of the same level in the shared network to mine the hidden information and knowledge between nodes of the same type. By combining the shared network with the topological dimension roll-up algorithm, the implicit shared association relationship between high-dimensional nodes that combines node attribute knowledge is calculated. For example, through the results of the topological dimension roll-up of the part-level shared network, it can be observed and analyzed that parts with weaker mechanical properties actually have a shared relationship with parts with stronger mechanical properties. Each node in the rolled-up network is a high-dimensional node at the part level, and the relationship between nodes reflects the potential shared relationship between high-dimensional nodes at the part level. This calculation result, which combines the relationship between entities, the attributes of entities, and the topological structure and path information of the information network itself, provides users with extremely in-depth and intuitive analysis results from a new perspective. See the schematic diagram of the knowledge reasoning method for the knowledge graph based on the comprehensive topological dimension roll-up and shared relationship. Figure 6 .

[0060] The specific implementation process is as follows: on the basis of the high-dimensional nodes obtained by topological dimension merging, the sharing relationship between nodes is further combined to obtain the deep-level association relationship between high-dimensional nodes and high-dimensional nodes. Taking the shared network at the part level as an example, first, the ID values ​​of the original low-dimensional nodes before merging corresponding to each part-level high-dimensional node (stored in the list in step 4) are stored as attribute values ​​in the corresponding table item of the entity attribute table. This operation retrieves the high-dimensional node information obtained by rolling up the topological dimension. The head entity and tail entity of all edges with the relationship type of shared material in the relationship table are queried, that is, the head and tail low-dimensional node IDs of all edges in the shared network stored in the relationship table in step 5 are retrieved. This operation retrieves the sharing relationship between nodes. Next, according to the head and tail low-dimensional node IDs of each edge in the shared network (as attribute values), the corresponding high-dimensional node ID value is queried from the entity attribute table, and the high-dimensional entity information and shared network obtained by topological dimension merging are combined to obtain each pair of part-level high-dimensional nodes with a sharing association relationship. Finally, the calculated high-dimensional node ID pairs with implicit sharing association relationships are stored in the relationship table for user query and analysis. Through this method, we can obtain deeper associations and hidden knowledge between nodes combined with attribute value information. Compared with the existing OLGP method, this method makes more comprehensive and thorough use of the topological structure information, path information, relationship information and attributes between entities in the knowledge graph, combines the topological dimension roll-up and the hidden shared relationships in the network, and more deeply mines the deep knowledge in the graph from multiple angles and levels, providing users with in-depth data insights.

[0061] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further explain various possible combinations. In addition, the various different embodiments of the present invention can also be combined arbitrarily, as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A knowledge reasoning method based on online graph processing technology for knowledge graph, characterized in that: The following steps are involved: Step 1: The text information to be processed is converted into triples by word segmentation, stop word removal, and part-of-speech tagging, and then data fusion is performed on each triple, and the data is corrected; Step 2, create a table in the MySQL database, insert data statements, and store the processed data in the MySQL database in the form of relational data; Step 3: In the relational table of the MySQL database, the project is used as the dimension and the equipment composition structure diagram of the project is used as the metric value; the equipment knowledge graphs of different projects are subjected to information dimension roll-up calculation. The graph merging process includes the merging of nodes and edges. In the merged tree structure, the probability value attribute of each entity is calculated; Step 4: Roll up the nodes of the part type in the knowledge graph in the MySQL database according to their topological dimension attribute-size value to form a number of part nodes in a high topological dimension; at the same time, roll up the nodes of the material type associated with the part nodes of different size ranges according to their topological dimension attribute-chemical composition content to form a number of high-dimensional material nodes in a high topological dimension; Step 5, for the nodes of the four conceptual hierarchical types of equipment, components, sub-components, and parts in the knowledge graph in the MySQL database, four different shared networks are calculated respectively, each shared network describes the relationship between shared low-level entities between all entities at the same level, and obtains a data structure map based on the structure, path, and relationship between entities of the network; in the step 5, the specific implementation process is as follows: the composition relationship and hidden shared relationship between nodes in the knowledge graph are used to carry out computational reasoning, and these knowledge are stored in the relationship table; for any node, all nodes participating in its composition structure are queried from the relationship table in the MySQL database, that is, all tail entity IDs whose head entity ID is the node are queried in the relationship table, and then all nodes of the same level that have a shared relationship with the node are further queried in the relationship table according to the query result, that is, all head entity ID values ​​whose tail entities in the relationship table include the ID values ​​in the aforementioned query results are queried, and all nodes that have a shared relationship with the node are determined through two queries, and the weight value of the relationship is the absolute value of the difference between the attribute values ​​of the two nodes divided by the normalization parameter, and finally the calculated result is inserted into the relationship table; repeating the above process for all nodes at any level, a shared network reflecting the shared relationship between nodes at any level is obtained; Step 6, based on the shared networks of different levels calculated in step 5, combined with the topological dimension attributes of the nodes in the network, all nodes of the same type in the shared network whose topological dimension attribute values ​​are in the same range are merged into high-dimensional nodes, and the original shared network is topologically rolled up to obtain a new shared network composed of high-dimensional nodes including a new topological structure; in said step 6, the specific implementation process is as follows: based on the high-dimensional nodes obtained by topological dimension merging, further combined with the sharing relationship between the nodes, so as to obtain a deep-level association relationship between high-dimensional nodes and high-dimensional nodes; the ID values ​​of several original low-dimensional nodes before merging corresponding to each part layer high-dimensional node are stored as attribute values ​​in the corresponding table item of the entity attribute table, and the ID value of each part layer high-dimensional node is stored in the corresponding table item of the entity attribute table. The D value is stored in the list in step 4. This operation retrieves the high-dimensional node information obtained by rolling up the topological dimension, and queries the head entity and tail entity of all edges with the relationship type of shared material in the relationship table, that is, the head and tail low-dimensional node IDs of all edges in the shared network stored in the relationship table in step 5 are retrieved. This operation retrieves the shared relationship between the nodes; according to the head and tail low-dimensional node IDs of each edge in the shared network, the corresponding high-dimensional node ID value is queried from the entity attribute table to obtain each pair of part-layer high-dimensional nodes with a shared association relationship. This step combines the high-dimensional entity information and the shared network obtained by merging the topological dimensions; finally, the calculated high-dimensional node ID pairs with implicit shared association relationships are stored in the relationship table for user query and analysis.

2. A knowledge reasoning method based on online graph processing technology for knowledge graph as claimed in claim 1, characterized in that: In step 1, data error correction includes the following two categories: Rule-based error correction method: Scan and verify data based on a predefined set of rules. Each rule represents a constraint on data correctness. When data violates these constraints, the data is considered an error and traced back to the source, supplemented by manual modification. Error correction method based on graph pattern matching: Matching is performed according to predefined graph patterns. The predefined graph patterns represent the normal structure in the knowledge base. The parts that do not match these predefined patterns are searched in the knowledge base. If a structure that does not match the predefined pattern is detected, it is regarded as an error, and data tracing is performed, supplemented by manual modification.

3. The knowledge reasoning method based on online graph processing technology and oriented to knowledge graph as claimed in claim 1, characterized in that: In step 2, the relational data includes: entity table, entity attribute table, relationship table, merge record table and high-dimensional table; the entity table and entity attribute table jointly define the key entities of the project and their related attributes; the relationship table is used to mine and store the complex relationships and attributes between project components; the merge record table records which projects are merged to serve OLGP reasoning; the high-dimensional table records the information dimension, and the topology dimension rolls up the related information and attributes to serve OLGP reasoning.

4. A knowledge reasoning method based on online graph processing technology and oriented to knowledge graph as claimed in claim 3, characterized in that: In step 2, the entity table includes 3 columns, wherein the first column is the primary key, storing the unique ID value of each device, the second column is the type of the entity, and the third column stores the name of the entity; the entity attribute table includes 5 columns, wherein the first column is the ID value of the primary key attribute, the second column is the entity ID value to which the attribute belongs, the third column is the attribute name, the fourth column is the attribute value, and the fifth column is the project to which the attribute belongs; the relationship table includes 6 columns, wherein the first column is the unique ID value of the relationship, the second column and the third column are the ID values ​​of the head and tail entities of the relationship respectively, the fourth column is the relationship type, the fifth column is the weight, and the sixth column is the project to which the relationship belongs; the merged record table includes 3 columns, wherein the first column is the unique ID value of the record generated during each information dimension merge, the second column is the merged projects, and the third column is the name corresponding to the merge; the high-dimensional table includes 4 columns, wherein the first column is the unique ID value corresponding to the node in the graph after the information dimension is rolled up, the second column is the ID value of the corresponding node before the merge, the third column is the ID value of the record generated during the information dimension merge, and the fourth column stores the main result generated by the information dimension roll-up—the probability value of each project device appearing in different projects.

5. The knowledge reasoning method based on online graph processing technology and oriented to knowledge graph as claimed in claim 1, characterized in that: In the step 3, the specific implementation process is as follows: by taking out the data in the relational table in the MySQL database and performing calculations to complete the knowledge reasoning based on the information dimension roll-up for the knowledge graph, the results are stored in the relational table, the high-dimensional table, and the merged record table; in the relational table of the MySQL database, all the device nodes included in the project map to be merged are queried respectively, and the ID values ​​of all the device nodes queried in the relational table are stored in the data set of the corresponding project respectively, the two sets are merged into a list and all the node ID values ​​in the list are counted, and the number of occurrences divided by the number of projects is the probability value of the device node corresponding to the node ID value appearing in different projects; the calculated probability values ​​are stored one by one in the corresponding table items in the high-dimensional table, and the probability values ​​obtained by reasoning meet the user's analysis needs for the utilization of different devices; the information in all the relational tables of the two projects is taken out and merged into a set, and all the information in the set is inserted into a specific table item of the database to complete the information dimension roll-up, so as to obtain a view of observing and analyzing the project composition structure from a macro-global perspective, and each time the information dimension is rolled up, a row of data queried from the entity table needs to be inserted into the merged record table to record the projects involved in the roll-up and the name-related information corresponding to the merge.

6. The knowledge reasoning method based on online graph processing technology and oriented to knowledge graph as claimed in claim 1, characterized in that: In step 4, the specific implementation process is as follows: by performing calculations on the data in the entity attribute table and the relationship table to complete the knowledge reasoning oriented to the knowledge graph based on the topological dimension roll-up, the results are saved in the relationship table; the chemical component attribute values ​​of all material nodes are queried from the entity attribute table in the MySQL database and added respectively, and the mapping relationship between the results and the ID values ​​of each material node is stored in the dictionary, and then the total chemical component content values ​​of each material node are sorted from high to low; according to the above dictionary and the attribute value division rules of different high-dimensional material nodes, the ID values ​​of the low-dimensional nodes corresponding to different high-dimensional nodes before roll-up are stored in different lists in sequence; for the nodes of the part type, the same is done from M The entity attribute table in the ySQL database is queried to retrieve all its dimension values ​​and the volume is calculated accordingly. The mapping relationship between the result and the ID value is stored in the dictionary, and the node ID values ​​in different dimension value ranges are stored in different lists after sorting the results. After the node is merged, the data corresponding to the high-dimensional node is inserted into the entity table. The high-dimensional part node that has an association relationship with the high-dimensional material node with a total chemical content in different intervals is queried from the relationship table for the part node ID value before the merger, and the result is matched with the aforementioned list storing the ID values ​​of different low-dimensional material nodes, and the composition association relationship between the high-dimensional nodes at different levels after the topological dimension is rolled up is inserted into the relationship table.