Geographic knowledge graph densification method based on classified knowledge

Abstract

The invention belongs to the technical field of artificial intelligence and geographic information, and particularly relates to a geographic knowledge graph densification method based on classified knowledge. The method comprises the following steps: constructing an element set of classified knowledge; calculating and generating semantic enhancement nodes of the geographic knowledge graph; generating a semantic enhanced relationship of the geographic knowledge graph; and fusing to generate a dense geographic knowledge graph. The knowledge graph optimization method is scientific, universal and efficient, the density and integrity of the geographic knowledge graph can be improved, and data support is provided for downstream applications such as geographic information analysis, spatial reasoning and recommendation systems driven by the knowledge graph.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and particularly relates to a method for knowledge graph completion with semantic densification. Background Art

[0002] With the wide application of knowledge graphs in various fields, the denseness of knowledge graphs has become a key factor in improving their reasoning and application capabilities. A knowledge graph represents entities and their mutual relationships in a structured manner, thereby helping machines better understand and utilize information. However, in many current fields, there is a sparsity problem in knowledge graphs, that is, the entity and relationship data in knowledge graphs are relatively lacking, resulting in low accuracy of reasoning results and low efficiency of knowledge transfer. Especially in application scenarios facing complex and multi-domain integration, the sparsity problem of knowledge graphs is particularly prominent.

[0003] In order to improve the semantic denseness of knowledge graphs, researchers have gradually realized that introducing more fine-grained classification and semantic features is an effective means. A classification index system is a framework that can refine the classification of entities and relationships in a knowledge graph according to different semantic levels. Through the classification index system, multi-level and multi-dimensional semantic information in the knowledge graph can be captured, making the relationships in the graph closer and the denseness improved.

[0004] However, most of the existing knowledge graph completion methods only rely on the existing structural information and ignore the introduction of domain-specific classification semantic features. In addition, the current mainstream methods are often limited to a single data source and cannot effectively utilize the rich information in multi-modal data (such as text, images, etc.) to further enhance the denseness of the graph. Due to the complexity of knowledge systems in different fields, a single-source index system cannot meet the multi-dimensional knowledge expression requirements.

[0005] Therefore, there is an urgent need for a general knowledge graph densification method based on a classification index system. This method can not only systematically collect classification index systems from national standards, academic papers, and multi-modal data, but also apply these index systems to the knowledge graph through a reasonable classification framework, thereby greatly improving the denseness of the graph. In view of the deficiencies of the above technologies, those skilled in the art are committed to developing a general, scientific, and automatic ecological civilization mode recommendation method. Summary of the Invention

[0006] The present invention discloses a method for densifying a knowledge graph based on a knowledge graph and a classification index framework, aiming to solve the sparsity problem existing in existing methods when dealing with complex knowledge systems. In the current knowledge graph for large-scale data processing, the problem of sparse relationships easily occurs, affecting the effects of reasoning, querying, and knowledge discovery. To improve the density of the knowledge graph, the present invention proposes a method for densifying a knowledge graph based on an authoritative classification index framework, which can significantly improve the connectivity and semantic richness between nodes in the graph.

[0007] To achieve the above object, the present invention provides a method for completing a semantic-dense knowledge graph, including the following steps: S1. Construct a set of elements of classification knowledge; S2. Calculate and generate nodes with enhanced semantics of the geographical knowledge graph; S3. Generate relationships with enhanced semantics of the geographical knowledge graph; S4. Integrate to generate a dense geographical knowledge graph.

[0008] 1. Further, the specific method of step S1 includes the following sub-steps: S11. Obtain classification knowledge elements , and form a classification knowledge set , specifically as follows:

[0009]

[0010] Among them, is the classification knowledge set; represents the i-th classification knowledge in the classification knowledge set; is the number of classification knowledge elements in the classification knowledge set; represents the semantic element corresponding to the i-th classification knowledge; represents the range of the value domain interval; " " indicates that it is uncertain whether the boundary is closed; represents the name corresponding to the value domain interval; represents the number of interval elements of the attribute; S12. Import the geographical knowledge graph, specifically as follows:

[0011] Among them, is the set of triples of the geographical knowledge graph; is the index of the triples of the geographical knowledge graph; is the g-th triple in the set of triples of the geographical knowledge graph; are respectively the geographical entity, relationship, and attribute of the geographical knowledge graph; refers to the number of all triples in the geographical knowledge graph; S13. Determine the set E of geographical entities to be complemented and the triples associated with the geographical entities to be complemented , specifically as follows:

[0012]

[0013] where E is the set of geographical entities to be complemented; CT is the set of triples associated with the geographical entities to be complemented; k is the index of the geographical entities to be complemented in GeoKG; is the geographical entity to be complemented; is the triple associated with the geographical entity to be complemented; is the number of geographical entities to be complemented; are the geographical entities, relationships, and attributes of the geographical knowledge graph respectively; represents the number of geographical entities to be complemented; S14. Screen the effective classification knowledge set, and the specific screening rules are as follows:

[0014] where is the effective classification knowledge set screened from the classification knowledge set; S15. Establish an index between CT and S through semantic content:

[0015] where is the mapping established between the triples associated with the geographical entities to be complemented and the classification knowledge set through semantic content.

[0016] 2. Further, the specific method of step S2 includes the following sub-steps: S21. Determine the of the classification knowledge in the effective classification knowledge set, input , and generate semantic enhanced nodes. The specific generation function is as follows:

[0017] where is the semantic enhanced node of the geographical knowledge graph, is the jth classification semantics of the classification knowledge with subscript i in is The minimum and maximum values of the range of the j-th classification semantics of the classification knowledge with subscript i is the attribute node value of the triple set associated with the geographical entity to be completed; Indicates an empty set and no node is generated.

[0018] 3. Further, the specific method of step S3 includes the following sub-steps: S31. Determine the in the valid classification knowledge set, input , and generate a semantic enhancement relationship, that is, the edge between the node to-be-completed geographical entity and the semantic enhancement node. The specific generation function is as follows:

[0019] Among them, is the geographical knowledge graph semantic enhancement relationship; Indicates the semantics corresponding to the classification knowledge with subscript i; is the direction of the semantic enhancement relationship, and the direction points from the to-be-completed geographical entity to the semantic enhancement node associated with in S21; Indicates an empty set and no relationship is generated.

[0020] 4. Further, the specific method of step S4 includes the following sub-steps: S41. Merge the semantic enhancement nodes generated in S21 with the in GeoKG to generate the total set of nodes of the completed geographical knowledge graph, specifically as follows:

[0021] Among them, is the total set of nodes of the completed geographical knowledge graph, is the set of semantic enhancement nodes of the geographical knowledge graph, is the set of geographical entities in the geographical knowledge graph; S42. Merge the semantic enhancement relationships generated in S31 with the in GeoKG to generate the total set of relationships of the completed geographical knowledge graph, specifically as follows:

[0022] Among them, is the total set of relationships of the completed geographical knowledge graph, is the set of semantic enhancement relationships of the geographical knowledge graph, is the set of geographical entity relationships of the geographical knowledge graph.

[0023] Beneficial effects: Compared with existing methods, a method for densifying a geographical knowledge graph based on classification knowledge provided by the present invention has the following beneficial effects: Improving the semantic density of the knowledge graph: Through the classification semantic classification and automatic expansion mechanism proposed by the present invention, it is possible to effectively fill the blank areas in the knowledge graph, especially in fields with a high degree of graph structure sparsity, achieving higher semantic coverage and relationship connection degrees, and enhancing the structural integrity of the knowledge graph.

[0024] (1) Precise classification semantic classification: By constructing a classification system corpus and using a classification algorithm to automatically classify knowledge, this method ensures the accurate classification and organization of semantic elements. At the same time, through the verification and adjustment of the classification results, the classification accuracy is improved, ensuring the rigor and usability of the knowledge graph structure.

[0025] (2) Multi-source data integration and strong knowledge expansion ability: This method can automatically obtain classification metrics and related knowledge from a variety of heterogeneous data sources, integrate data from different sources, increase the breadth and depth of the knowledge graph, and enhance its ability in multi-domain and multi-dimensional knowledge representation.

[0026] (3) Complicating the graph relationships and enhancing the reasoning ability: By generating multi-level complex relationships, this method further enriches the association methods between entities in the graph. Through the application of an inference engine and a multi-level relationship generation algorithm, new knowledge triples can be dynamically generated, enhancing the reasoning ability and scalability of the knowledge graph.

[0027] (4) Visualization and verification of the densification effect: Through the visual display of the graph structure and a density analysis tool, this method can intuitively present the optimization effect of the knowledge graph. Implementers can monitor the densification process of the graph in real time and conduct precise effect evaluation to ensure the continuous improvement of the knowledge graph. Description of the drawings

[0028] Figure 1 It is the overall flowchart of the method for semantic densification of the knowledge graph Figure 2 Comparison chart of the performance improvement of the semantic densification of the geographical knowledge graph Detailed implementation manners The present invention will be further described in detail below with reference to the drawings.

[0029] See Figures 1-2 , a method for densifying a geographical knowledge graph based on classification knowledge provided by the present invention can generally be divided into three major aspects: constructing a classification knowledge semantic enhancement element set, calculating and generating a geographical knowledge graph semantic enhancement node, and determining and generating a geographical knowledge graph semantic enhancement relationship. Specifically, taking the partial ecological civilization mode knowledge graph ECMKG as an implementation example, it includes the following steps: S1. Construct the element set of classification knowledge S11. Obtain various classification knowledge from national standards, industry standards, thematic maps, high-quality papers, official reports, and news reports through manual search. For example, in the "Notice of the State Council on Adjusting the Urban Scale Classification Standard", the urban level is classified according to the population, and it is summarized that Classification = , and the specific population urban scale classification knowledge is as follows: . Among them = 1, indicating that there is one classification knowledge; = 2, indicating that the first classification knowledge has two categories; S12. Import the ecological civilization model knowledge graph ECMKG (ECMKG is a graph structure composed of a series of nodes and relationships); S13. Determine the set E of geographical knowledge graph objects to be completed and its corresponding triple set from ECMKG, }), specifically as , where EN = 3, and the set E of geographical knowledge graph objects to be completed is determined as { }; S14. Determine the effective semantic enhancement element set. Since , according to the rule screen the classification knowledge, so the population urban classification rule can be incorporated into the semantic enhancement element set , so ; S15. Establish an index for each and S obtained in S14, that is, establish an index through semantics for the population urban classification rule and the triple associated with the geographical entity to be completed.

[0030] S2. Calculate and generate semantic enhancement nodes of the geographical knowledge graph S21. Input and =[(Quzhou City - Kaihua County, number of beds (unit: beds per thousand people), 3.2), (Diqing Tibetan Autonomous Prefecture - Shangri-La City, number of beds (unit: beds per thousand people), 8.4), (Chengdu City - Pengzhou City, number of beds (unit: beds per thousand people), 7.4)] (partial fields), and substitute them into the semantic enhancement node model of the geographical knowledge graph to output the set of completed node names .

[0031] S3. Generate semantic enhancement relationships of the geographical knowledge graph S31. Input and And (Partial fields), substitute them into the generated semantic enhanced relationship model of the geographical knowledge graph, and output the relationship set between each node .

[0032] S4. Fusion to generate a dense geographical knowledge graph S41. According to the set N of completed node names, use Neo4j to create geographical knowledge graph nodes one by one, merge them with ECMKG, and generate the total set nodeSet of completed nodes; S42. According to the set of key-value pairs <name, direction> of the G-completed relationship, find the corresponding geographical entities according to the direction in Neo4j, create geographical knowledge graph relationships one by one, merge them with ECMKG, and generate the total set relationSet of ECMKG relationships after completion.

[0033] S5. Evaluation of the semantic density effect of the knowledge graph S51. To verify the effectiveness of this method, select the function for calculating the global sparsity of the graph in graph theory to show the hint effect of this method on the graph density. The function for calculating the global sparsity of the graph is as follows:

[0034] where is the global sparsity of the graph; is the number of edges in the graph; is the number of nodes in the graph; S52. Count the number of nodes and edges in ECMKG and the number of nodes and edges in the ECMKG after completion, and input them into the function for calculating the global sparsity of the graph respectively. The original global sparsity value of ECMKG is 8.3333%, and the global sparsity value of the enhanced dense ECMKG is 9.3750%, which verifies the effectiveness of this method. For details, see Figure 2 .

[0035] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for densifying a geographical knowledge graph based on classification knowledge, characterized in that Including the following steps: S1. Construct a set of elements of classification knowledge; S2. Calculate and generate nodes with enhanced semantics of the geographical knowledge graph; S3. Generate relationships with enhanced semantics of the geographical knowledge graph; S4. Integrate and generate a dense geographical knowledge graph.

2. The method for densifying a geographical knowledge graph based on classification knowledge according to claim 1, wherein The specific steps of step S1 are as follows: S11. Obtain classification knowledge elements to form a classification knowledge set as follows: Among them, is the classification knowledge set; represents the i-th classification knowledge in the classification knowledge set; is the number of classification knowledge elements in the classification knowledge set; represents the semantic element corresponding to the i-th classification knowledge; represents the range of the value domain interval; " " indicates that the boundary is uncertain whether it is closed; represents the name corresponding to the value domain interval; represents the number of intervals of the attribute element; S12. Import the geographical knowledge graph, specifically as follows: Among them, is the set of triples of the geographical knowledge graph; is the index of the triples of the geographical knowledge graph; is the g-th triple in the set of triples of the geographical knowledge graph; are the geographical entities, relationships, and attributes of the geographical knowledge graph respectively; refers to the number of all triples in the geographical knowledge graph; S13. Determine the set E of geographical entities to be completed and the triples associated with the geographical entities to be completed from GeoKG associated triples , as follows: Among them, E is the set of geographical entities to be completed; CT is the set of triples associated with the geographical entities to be completed; k is the index of the geographical entity to be completed in GeoKG; is the geographical entity to be completed; is the triple associated with the geographical entity to be completed; is the number of geographical entities to be completed; are the geographical entities, relationships, and attributes of the geographical knowledge graph respectively; represents the number of geographical entities to be completed; S14. Screen an effective set of classification knowledge, and the specific screening rules are as follows: Among them, is a valid classification knowledge set selected from the classification knowledge set; S15. Establish an index between CT and S through semantic content Index: Among them, is a mapping established through semantic content between the triples associated with the geographical entity to be completed and the classification knowledge set.

3. The method for densifying a geographical knowledge graph based on classification knowledge according to claim 1, wherein The specific steps of step S2 are as follows: S21. Determine the of the classification knowledge in the effective classification knowledge set, and input to generate a semantic enhancement node. The specific generation function is as follows: Among them, is a semantic enhancement node of the geographical knowledge graph, is the j-th classification semantics of the classification knowledge with subscript i in the following, which is expressed as the node name of the newly added node, is the minimum and maximum values of the value range of the j-th classification semantics of the classification knowledge with subscript i in the following, is the attribute node value of the triple set associated with the geographical entity to be completed; represents an empty set and no node is generated.

4. A method for densifying a geographical knowledge graph based on classification knowledge according to claim 1, characterized in that, The specific steps of step S3 are as follows: S31. Determine the in the set of valid classification knowledge, and input to generate a semantic enhancement relationship, that is, an edge between the geographical entity with nodes to be complemented and the semantic enhancement nodes. The specific generation function is as follows: Among them, is the semantic enhancement relationship of the geographical knowledge graph; represents the semantics corresponding to the classification knowledge with subscript i in is the direction of the semantic enhancement relationship, and the direction is from the geographical entity to be completed pointing to the semantic enhancement node associated with in S21 ; represents an empty set and no relationship is generated.

5. A method for densifying a geographical knowledge graph based on classification knowledge according to claim 1, characterized in that The specific steps of step S4 are as follows: S41. Merge the semantically enhanced nodes generated in S21 with the in GeoKG to generate the total set of nodes in the completed geographical knowledge graph, as follows: Among them, is the total set of nodes in the completed geographical knowledge graph, is the set of nodes with enhanced semantics in the geographical knowledge graph, is the set of geographical entities in the geographical knowledge graph; S42. Merge the semantic enhancement relationships generated in S31 with the in GeoKG to generate the total set of relationships in the completed geographical knowledge graph, as follows: Among them, is the total set of completed geographical knowledge graph relationships, is the semantic enhancement relationship set of the geographical knowledge graph, is the geographical entity relationship set of the geographical knowledge graph.

Images