A method for extracting geographical entity themes based on spatial association features of geographical entities

Abstract

The present invention discloses a method for extracting geographical entity themes based on spatial association features of geographical entities, which relates to the technical fields of urban planning, natural language processing, and urban transportation. First, geographical entity review data and geographical entity location data are screened. Secondly, a geographical entity association semantic network is constructed, and a set of sub-networks of the geographical entity association semantic network is calculated. Then, a keyword semantic network is constructed for each sub-network in the set of sub-networks of the geographical entity association semantic network. Finally, by extracting the keyword semantic network and the geographical entity association form in the sub-networks of the geographical entity association semantic network, the themes of the geographical entities in each sub-network are determined. The present invention can, on the basis of identifying the text semantic connection relationship between geographical entities, divide out spatial connection sub-networks, and then combine the geographical entity association form and the calculation results of the keyword semantic network evaluation indexes in each spatial connection sub-network to extract the themes of each spatial connection sub-network.

Description

Technical Field

[0001] The present invention relates to the technical fields of urban planning, tourism planning, urban traffic, and complex network modeling, and particularly to a method for extracting the themes of geographical entities based on the spatial association characteristics of geographical entities. Background Art

[0002] The extraction of the thematic content of geospatial entities based on review data has always been a research focus and application hotspot in the fields of urban tourism, geographic information systems, traffic geography, etc. Generally, the methods and ideas for theme extraction mainly adopt natural language analysis of text content to achieve text classification and theme extraction. However, when extracting the themes of geospatial entities, not only the review data related to geographical entities need to be considered, but also the spatial association relationships between geographical entities need to be considered. Since the text data related to geospatial entities contains a large number of association relationships between geographical entities, therefore, the spatial association relationships between geographical entities can be extracted based on the review text first, and the grouping between geographical entities can be achieved through the spatial association relationships. After further grouping, the thematic features of each group can be extracted through a certain keyword semantic modeling and the association patterns between geographical entities. Here, the association pattern between geographical entities refers to the type of co-occurrence relationship between the names of geographical entities extracted from the review text.

[0003] Overall, the present invention first extracts the spatial association relationships between geographical entities, groups the geographical entities, that is, the community division in the complex network; then extracts the themes from the grouped results through keyword semantic modeling and the association patterns between geographical entities. This has a certain degree of innovation compared with the traditional method of directly clustering from text semantics to obtain themes, and can group and mine the theme features on the basis of determining the spatial association relationships.

[0004] In addition, when mining the theme features, the present invention also innovatively considers using deep learning algorithms to extract the association patterns between geographical entities, so as to better reveal the theme content of a certain type of geographical entity from the connection mode between entities. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for extracting the themes of geographical entities based on the spatial association characteristics of geographical entities. The present invention can obtain the thematic features of the grouping between geographical entities based on text reviews on the basis of extracting the spatial association relationships between geographical entities.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A method for extracting the themes of geographical entities based on the spatial association characteristics of geographical entities according to the present invention includes the following steps:

[0008] Step 1: Extract the required geographical entity review data from the geographical entity review data by means of subject word screening;

[0009] Step 2: Based on the geographical entity review data and the data of geographical entity names, construct a geographical entity associated semantic network through the co-occurrence degree of geographical entity names in the geographical entity review data;

[0010] Step 3: Perform complex network community division calculation on the geographical entity associated semantic network to obtain a set of sub-networks of the geographical entity associated semantic network;

[0011] Step 4: Construct a keyword semantic network for each sub-network in the set of sub-networks of the geographical entity associated semantic network;

[0012] Step 5: Extract the geographical entity associated pattern names for each sub-network in the set of sub-networks of the geographical entity associated semantic network;

[0013] Step 6: Determine the themes of geographical entities in each sub-network through the extraction of keyword semantic networks and geographical entity associated pattern names of each sub-network.

[0014] As a further optimization scheme of the geographical entity theme extraction method based on the spatial association characteristics of geographical entities described in the present invention, each geographical entity has a unique name, which is called the geographical entity name; in Step 1, each geographical entity corresponds to a geographical entity location information.

[0015] As a further optimization scheme of the geographical entity theme extraction method based on the spatial association characteristics of geographical entities described in the present invention, in Step 1, the geographical entity review data is geographical entity review data with geographical entity location information, and the method for extracting geographical entity review data with geographical entity location information is as follows:

[0016] Step 1.1: Establish a one-to-one correspondence between the geographical entity review data and the geographical entity location information;

[0017] Step 1.2: Extract the required geographical entity review data by judging whether each comment in the geographical entity review data includes the specified subject word;

[0018] Step 1.3: Further screen out the finally required geographical entity review data with geographical entity location information according to the required geographical entity review data extracted in Step 1.2 and the one-to-one correspondence between the geographical entity review data and the geographical entity location information.

[0019] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, the one-to-one correspondence between geographical entity review data and geographical entity location information means that each review in the geographical entity review data corresponds to only one unique geographical entity.

[0020] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, step 2 includes the following steps:

[0021] Step 2.1: Use geographical entities as the nodes of the complex network;

[0022] Step 2.2: If the review of a geographical entity A includes the name of another geographical entity B, it is considered that there is a first complex network connection edge between A and B; if the review of A includes the names of two different geographical entities C and D, it is considered that there is a second complex network connection edge between C and D;

[0023] Step 2.3: Based on the complex network theory, establish a geographical entity association semantic network for the nodes and the first and second complex network connection edges in the complex network in steps 2.1 and 2.2.

[0024] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, in step 3, the sub-networks in the sub-network set refer to the respective communities obtained by community division calculation, and the number of network nodes included in each sub-network should be greater than 1.

[0025] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, in step 4, for each sub-network, extract the geographical entities included in each sub-network and the review data corresponding to the geographical entities, and construct a keyword semantic network. The specific steps are as follows:

[0026] Step 4.1: Extract the high-frequency keywords in the reviews of the sub-network;

[0027] Step 4.2: Use the high-frequency keywords as the nodes of the complex network to be constructed;

[0028] Step 4.3: If a certain high-frequency keyword appears in the same review, it is considered that there is a complex network connection edge between these two nodes;

[0029] Step 4.4: Based on the complex network theory, establish a keyword semantic network for the nodes and the complex network connection edges in the complex network in steps 4.2 and 4.3.

[0030] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, in step 5, the name of the geographical entity association pattern is obtained through calculation by a deep learning model, and the specific steps are as follows:

[0031] Step 5.1: Extract the comments with complex network connection edges from the geographical entity association semantic network, and add special symbols to identify the geographical entity names identified in the comments.

[0032] Step 5.2: Add the geographical entity name corresponding to the comment and the first delimiter M at the beginning of the comment in step 5.1 to obtain the processed data result of the comment.

[0033] Step 5.3: Divide the data results of part of the comments in step 5.2 into a training set and a test set.

[0034] Step 5.4: Add the preset geographical entity association pattern name at the beginning of the comments in the training set and the test set, and separate the geographical entity names on both sides of the comments processed in step 5.2 with the second delimiter N. The first delimiter M is different from the second delimiter N.

[0035] Step 5.5: Use the R-BERT model to train the pre-trained model with the data processed in step 5.4 to obtain the pre-trained model.

[0036] Step 5.6: Use the pre-trained model obtained in step 5.5 to calculate the geographical entity association pattern name for the data results of the remaining part of the comments in step 5.3.

[0037] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities according to the present invention, the method for determining the themes of geographical entities in each sub-network in step 6 includes the following steps:

[0038] Step 6.1: Count the top Y keywords with weighted degree centrality values in the keyword semantic network of each sub-network, where Y is an integer greater than 3 and less than 10.

[0039] Step 6.2: Count the top Y geographical entity association pattern names with the highest occurrence frequencies of the geographical entity association pattern names in each sub-network.

[0040] Step 6.3: Use the keywords and geographical entity association pattern names obtained by the statistics in steps 6.1 and 6.2 as the themes of the sub-network.

[0041] As a further optimization solution of the method for extracting geographical entity themes based on the spatial association characteristics of geographical entities described in the present invention, after step 6.3, step 6.4 is further included. Step 6.4: Visualize the spatial distribution of geographical entities in different sub-networks according to the geographical entity location information.

[0042] The present invention adopts the above technical solutions and has the following technical effects compared with the prior art:

[0043] (1) The present invention provides a method for extracting geographical entity themes based on the spatial association characteristics of geographical entities, which can obtain the theme characteristics of grouped geographical entities on the basis of extracting the spatial association relationships between geographical entities;

[0044] (2) Based on the association relationships between geographical entities in complex networks and text semantics, the present invention proposes a method for extracting geographical entity themes based on the spatial association characteristics of geographical entities. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic diagram of the overall process of the present invention.

[0046] Figure 2 is a schematic diagram for explaining the innovative features of the technical path of the present invention.

[0047] Figure 3 is a schematic diagram of the detailed technical process.

[0048] Figure 4 is a schematic diagram of the principle for constructing the geographical entity association semantic network; among them, (a) is that a geographical entity comment contains the name of another geographical entity, and (b) is that a geographical entity comment contains the names of two other geographical entities.

[0049] Figure 5 is a schematic diagram of the principle for constructing the keyword semantic network.

[0050] Figure 6 is a schematic diagram of adding the geographical entity name and separator before the original comment sentence.

[0051] Figure 7 is a schematic diagram of adding separators to geographical entities in the comment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The following further elaborates on the technical solutions of the present invention with reference to the accompanying drawings:

[0053] The core points of the present invention are mainly divided into two calculation processes. The first calculation process is to group geographical entities through text semantic relationships, that is, the sub-network division in the technology of the present invention. The result of the text semantic relationship grouping here is to group geographical entities spatially according to the spatial association strength relationship, which is different from general text classification. The second calculation process is to perform keyword semantic network analysis and geographical entity association pattern analysis on each sub-network after text semantic grouping. This step further mines the theme of each sub-network through natural language technology from the relationship of core keywords in the text and the spatial association pattern of geographical entities. This is also different from the general text theme mining idea, but combines the keyword semantic relationship and the spatial association pattern of geographical entities as the basis for judging the theme division of a group of geographically associated entities, providing a more comprehensive theme division scheme from a multi-dimensional perspective. In addition, when mining theme features, the use of deep learning algorithms to extract the association pattern between geographical entities is also considered to better reveal the theme content of a certain type of geographical entity from the connection method between entities.

[0054] See the appendix Figure 1 Step 1: From the geographical entity comment data, extract the required geographical entity comment data through the method of theme word screening. This step is achieved by checking whether the original geographical entity comment contains the specified theme word. For example, if a certain original geographical entity comment contains the keyword "city wall", it is considered that this original geographical entity comment is the comment data to be screened out. Each geographical entity has a unique name, called the geographical entity name. Geographical entities generally refer to entities with geographical spatial positions. To distinguish geographical entities in the analysis, each geographical entity should be assigned a unique geographical entity name.

[0055] Step 1 also includes that from the geographical entity location data, the required geographical entity location data can also be extracted through the method of theme word screening. Geographical entity location data also refers to geographical entity location information.

[0056] Generally speaking, the steps to extract the required geographical entity comment data and geographical entity location data in Step 1 are as follows:

[0057] Step 1.1: Establish a one-to-one correspondence between the geographical entity comment data and the geographical entity location data. That is, each geographical entity comment data corresponds to a specific geographical entity, and each geographical entity corresponds to a specific location data. In this way, although each comment data is text, it corresponds to a specific location data.

[0058] Step 1.2: Extract the required geographical entity comment data by judging whether each comment in the geographical entity comment data includes the specified theme word;

[0059] Step 1.3: Further, according to the required geographical entity review data extracted in Step 1.2 and the one-to-one correspondence between the geographical entity review data and the geographical entity location data, filter out the final required geographical entity location data.

[0060] Here, the one-to-one correspondence between the geographical entity review data and the geographical entity location data means that each comment in the geographical entity review data corresponds to only one unique geographical entity. That is to say, one geographical entity can correspond to multiple comment data.

[0061] Step 2: Based on the geographical entity review data and the geographical entity name data, construct a geographical entity associated semantic network through the co-occurrence degree of the geographical entity name in the geographical entity review data;

[0062] See Appendix Figure 2 In Path 1, generally, topic extraction is achieved through topic extraction in natural language analysis, such as the common Latent Dirichlet Allocation (LDA) technology. This method generally only realizes topic extraction for text semantics and lacks consideration of spatial location. The present invention adopts Path 2 in the attached figure. First, model the text semantic relationship, classify geographical entities from a spatial perspective, and then extract the topic of each classification result based on the technical method combining natural language and complex network. It can be seen that the present invention first performs topic extraction based on spatial classification and grouping, and can fully explore the topic features of each classification on the basis of understanding spatial relationships.

[0063] Further, see Appendix Figure 3 , and this study gives a detailed flowchart of the present invention taking geographical entities of scenic spots as an example.

[0064] Step 2.1: Use the geographical entity as a complex network node, that is, use the geographical entity of the scenic spot as a node of the complex network. Because it has specific spatial location information, it can reflect the relationship between geographical entities in the geographical space.

[0065] Step 2.2: See Appendix Figure 4 , Figure 4 is a schematic diagram of the principle for constructing a geographical entity associated semantic network; Figure 4 In (a), a geographical entity review contains the name of another geographical entity, Figure 4 In (b), a geographical entity review contains the names of two other geographical entities. If the review of a geographical entity A includes the name of another geographical entity B, it is considered that there is a first complex network connection edge between A and B.

[0066] If the comment of A includes the names of two different geographical entities C and D, then it is considered that there is an edge of the second complex network connection between C and D.

[0067] Step 2.3: According to the complex network theory, establish a geographical entity associated semantic network for the nodes of the complex network and the first and second complex network connection edges in Steps 2.1 and 2.2.

[0068] Step 3: Perform complex network community partition calculation on the geographical entity associated semantic network to obtain a set of sub-networks of the geographical entity associated semantic network;

[0069] The sub-networks in the set of sub-networks refer to the respective communities obtained by the community partition calculation, and the number of network nodes included in each sub-network should be greater than 1.

[0070] Currently, for the geographical entity associated semantic network, the combination situation between those geographical entities with close semantic connections can be solved through the community partition algorithm. Since geographical entities carry spatial location information, the spatial distribution state of geographical entities within each community can be drawn.

[0071] Step 4: Construct a keyword semantic network for each sub-network in the set of sub-networks of the geographical entity associated semantic network;

[0072] Furthermore, it is necessary to analyze the characteristics of each sub-network in the set of sub-networks to extract the detailed characteristics of each sub-network, and then give a richer content of theme characteristics. Specifically, for each sub-network, extract the geographical entities included in each sub-network and the corresponding comment data, and construct a keyword semantic network. The specific steps are as follows:

[0073] Step 4.1: High-frequency keywords can reflect the main core content and viewpoints in the comments. Therefore, first extract the high-frequency keywords in the comments of the sub-network.

[0074] Step 4.2: Since high-frequency keywords are the core in this step, use the high-frequency keywords as the nodes of the complex network to be constructed.

[0075] Step 4.3: In the present invention, the co-occurrence relationship of high-frequency keywords is used to comprehensively calculate which high-frequency keywords have relatively more important roles. See Appendix Figure 5 , if a certain high-frequency keyword appears in the same comment, then it is considered that there is an edge of the complex network connection between these two nodes. Step 4.4: According to the complex network theory, establish a keyword semantic network for the nodes and complex network connection edges of the complex network in Steps 4.2 and 4.3.

[0076] Step 5: Extract the geographical entity association pattern names for each sub-network in the sub-network set of the geographical entity association semantic network. Relation classification is an important NLP task, and its main goal is to extract the relationships between entities. In recent years, BERT-based relation extraction models have received extensive attention, and the R-BERT model is a BERT-based relation extraction method with excellent results. Therefore, the geographical entity association pattern names in this step are calculated through the deep learning model R-BERT. The specific steps are as follows:

[0077] Step 5.1: When using the R-BERT model for relation extraction, specific processing of the comments is required to achieve better deep learning model effects. Based on the characteristics of the comment texts of geographical entities in this study, the following specific implementation steps are proposed. See Appendix Figure 6 , extract the comments with complex network connection edges from the geographical entity association semantic network, and add special symbols to identify the geographical entity names identified in the comments. This step is to limit the geographical entities for which relation extraction is to be performed through specific identification symbols, so that the training of the R-BERT model is more targeted.

[0078] Step 5.2: See Appendix Figure 7 , add the geographical entity name corresponding to the comment and the first delimiter M at the beginning of the comment in Step 5.1 to obtain the processed comment data result. This step is to ensure that each comment contains the geographical entity name, because although some comments are known to belong to a specified geographical entity, the comment sentences may not contain the name of the geographical entity.

[0079] Step 5.3: Divide the comment data results in part of Step 5.2 into a training set and a test set;

[0080] Step 5.4: Add the preset geographical entity association pattern names at the beginning of the comments in the training set and the test set, and separate the geographical entity names on both sides of the comments processed in Step 5.2 with the second delimiter N to obtain the processed data; the first delimiter M is different from the second delimiter N. This step is to distinguish the geographical entity association pattern names from the geographical entity names, so that the model can recognize the difference between the two to achieve better model training effects.

[0081] Step 5.5: Use the R-BERT model to train the pre-trained model on the data processed in Step 5.4 to obtain the pre-trained model.

[0082] Step 5.6: Use the pre-trained model obtained in Step 5.5 to calculate the geographical entity association pattern names for the remaining comment data results in Step 5.3.

[0083] Step 6: Determine the themes of the geographical entities in each sub-network by extracting the keyword semantic networks and the names of geographical entity association patterns of each sub-network.

[0084] The method for determining the themes of the geographical entities in each sub-network in this step includes the following steps:

[0085] Step 6.1: Count the top Y keywords in the keyword semantic network of each sub-network in terms of weighted degree centrality value, where Y is an integer greater than 3 and less than 10;

[0086] Step 6.2: Count the top Y geographical entity association pattern names in terms of the occurrence frequency of the geographical entity association pattern names in each sub-network;

[0087] Step 6.3: Take the keywords and geographical entity association pattern names obtained from the statistics in Step 6.1 and Step 6.2 as the themes of the sub-network.

[0088] Step 6.4: Visualize the spatial distribution of the geographical entities in different sub-networks according to the geographical entity location data.

[0089] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A geographic entity topic extraction method based on the spatial association characteristics of geographic entities, characterized in that: The following steps are involved: Step 1: Extract the required geographic entity comment data from the geographic entity comment data by filtering the keywords; Step 2: Based on the geographic entity comment data and the geographic entity name data, a geographic entity association semantic network is constructed through the co-occurrence degree of the geographic entity name in the geographic entity comment data; Step 3: Performing complex network community partition calculation on the geographic entity associated semantic network to obtain a subnetwork set of the geographic entity associated semantic network; Step 4: construct a keyword semantic network for each sub-network in the sub-network set of the geographic entity associated semantic network; Step 5: extracting geographic entity association pattern names from each sub-network in the sub-network set of the geographic entity association semantic network; Step 6: Determine the theme of each sub-network geographic entity by extracting the keyword semantic network of each sub-network and the name of the geographic entity association pattern.

2. A geographic entity topic extraction method based on geographic entity spatial association features according to claim 1, characterized in that: Each geographic entity has a unique name, called the geographic entity name; in step 1, each geographic entity corresponds to a geographic entity location information.

3. A geographic entity topic extraction method based on geographic entity spatial association features according to claim 2, characterized in that: In step 1, the geographic entity comment data is the geographic entity comment data with the geographic entity location information. The method for extracting the geographic entity comment data with the geographic entity location information is as follows: Step 1.1, establishing a one-to-one correspondence between geographic entity comment data and geographic entity location information; Step 1.2, extracting the required geographic entity comment data by determining whether each comment in the geographic entity comment data includes a specified keyword; Step 1.3: further filter out the required geographic entity comment data with geographic entity location information based on the required geographic entity comment data extracted in step 1.2 and the one-to-one correspondence between the geographic entity comment data and the geographic entity location information.

4. A geographic entity topic extraction method based on geographic entity spatial association features according to claim 3, characterized in that: The one-to-one correspondence between geographic entity comment data and geographic entity location information means that each comment in the geographic entity comment data corresponds to only one geographic entity.

5. The method for extracting geographic entity topics based on spatial association features of geographic entities according to claim 1, characterized in that: Step 2 includes the following steps: Step 2.1, taking geographic entities as nodes of complex networks; Step 2.2: If a comment of a geographic entity A includes the name of another geographic entity B, it is considered that there is a first complex network connection edge between A and B; if a comment of A includes the names of two different geographic entities C and D, it is considered that there is a second complex network connection edge between C and D; Step 2.3: Based on the complex network theory, a geographic entity association semantic network is established for the nodes of the complex networks in steps 2.1 and 2.2 and the edges connecting the first and second complex networks.

6. A geographic entity topic extraction method based on geographic entity spatial association features according to claim 1, characterized in that: In step 3, the subnetworks in the subnetwork set refer to the communities obtained by community division calculation, and the number of network nodes included in each subnetwork must be greater than 1.

7. The method for extracting geographic entity topics based on spatial association features of geographic entities according to claim 1, characterized in that: Step 4: For each sub-network, extract the geographic entities included in each sub-network and the comment data corresponding to the geographic entity, and construct a keyword semantic network. The specific steps are as follows: Step 4.1, extract high-frequency keywords from the comments of the sub-network; Step 4.2: Use high-frequency keywords as nodes of the complex network to be constructed; Step 4.3: If a high-frequency keyword appears in the same comment, it is considered that there is a complex network connection edge between the two nodes; Step 4.4: Based on the complex network theory, a keyword semantic network is established for the nodes and edges of the complex networks in steps 4.2 and 4.

3.

8. The method for extracting geographic entity topics based on spatial association features of geographic entities according to claim 1, characterized in that: In step 5, the name of the geographic entity association pattern is calculated by the deep learning model. The specific steps are as follows: Step 5.1, extracting comments with complex network connection edges from the geographic entity association semantic network, and adding special symbols to the geographic entity names identified in the comments for identification; Step 5.2, adding the name of the geographic entity corresponding to the comment and the first separator M to the beginning of the comment in step 5.1, to obtain the data result of the processed comment; Step 5.3: Divide the data results of some comments in step 5.2 into a training set and a test set; Step 5.4, add the preset geographic entity association pattern name at the beginning of the comments in the training set and the test set, and separate the geographic entity names on both sides of the comments processed in step 5.2 with the second separator N to obtain the processed data; the first separator M is different from the second separator N; Step 5.5: Use the R-BERT model to perform pre-training model training on the data processed in step 5.4 to obtain a pre-training model; Step 5.6: Use the pre-trained model obtained in step 5.5 to calculate the geographic entity association pattern name based on the data results of the remaining comments in step 5.

3.

9. The method for extracting geographic entity topics based on spatial association features of geographic entities according to claim 1, characterized in that: The method for determining the subject of each sub-network geographic entity in step 6 includes the following steps: Step 6.1: Count the top Y keywords with weighted degree centrality values ​​in the keyword semantic network in each sub-network, where Y is an integer greater than 3 and less than 10; Step 6.2, counting the geographic entity association pattern names with the top Y appearance frequencies in each sub-network; Step 6.3: The keywords and geographic entity association pattern names obtained by statistics in steps 6.1 and 6.2 are used as the topics of the sub-network.

10. A method for extracting geographic entity topics based on spatial association features of geographic entities according to claim 9, characterized in that: Step 6.3 is followed by step 6.4, in which, according to the location information of the geographic entities, the spatial distribution of the geographic entities in different sub-networks is visualized.

Images