Association matching method for genealogy information and geographic information
By constructing and integrating genealogy and geographical knowledge graphs, the accuracy problem of matching genealogy information with geographical information was solved, deep correlation analysis was achieved, the accuracy and efficiency of data display were improved, and multi-scenario applications were supported.
Patent Information
- Application Number
- CN202511484140.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-17
- Publication Date
- 2026-01-16
AI Technical Summary
Existing technologies, when displaying the location trajectories of family members in a genealogy, suffer from unclear or overloaded data display due to changes in blood relations and scale, reducing the accuracy of matching genealogy information with geographical information.
By preprocessing genealogical and geographic information, a knowledge graph is constructed, entity alignment and fusion are performed, and deep association analysis is achieved by combining representation learning and reasoning computation.
It achieves unified matching of genealogical information and geographic information, improves the accuracy and efficiency of data display, and supports data support for scenarios such as academic research, cultural heritage preservation, and social governance.
Smart Images
Figure CN121350641A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of genealogy data management, in particular to a method for associating and matching genealogy information and geographic information. BACKGROUND
[0002] Genealogy not only contains important social, cultural, geographical and other information, but also is a reference for many disciplines and a part of Chinese culture. By analyzing the association between the information data of each person in the genealogy obtained through management and the geographic information, the migration trajectory of the clan over time can be obtained, which is of great significance to the culture and development of the clan.
[0003] In the process of displaying the location trajectory of the sample data of the people in the clan in the prior art, the location changes in the sample data of the people in the genealogy are usually clustered and analyzed, so as to display the migration trajectory of the sample data in the cluster. However, since the blood relationship and the like will affect the migration decision of the people, and when the scale of the map is different, if the required data precision does not change accordingly, it will cause unclear data display or data overload, etc. Therefore, relying solely on the location changes of the people will reduce the accuracy of the matching management of the genealogy information and the geographic information, and the result is not representative. SUMMARY
[0004] The purpose of the present application is to provide a method for associating and matching genealogy information and geographic information to solve the problems in the background art.
[0005] To achieve the above-mentioned purpose, the present application provides a method for associating and matching genealogy information and geographic information, comprising the following steps: S1, preprocessing the collected genealogy information and geographic information to obtain standardized data; S2, extracting entities, relationships and attributes from the standardized data to construct a genealogy initial knowledge graph and a geographic initial knowledge graph, respectively; S3, aligning the entities of the genealogy initial knowledge graph and the geographic initial knowledge graph; S4, based on the result of the entity alignment, merging the genealogy initial knowledge graph and the geographic initial knowledge graph into a unified and consistent knowledge graph; S5, performing representation learning and reasoning calculation on the merged knowledge graph to realize deep association analysis of the genealogy information and the geographic information, and outputting the matching result.
[0006] Preferably, in S2, the entities include people, families, events in the genealogy and places, regions and terrains in the geography; the relationships include the kinship between people, the migration relationship between people and places, the containing relationship between places and regions, etc.; and the attributes include the basic information of the entities, such as the birth date of the people and the longitude and latitude of the places.
[0007] Preferably, the steps for constructing the initial genealogical knowledge graph and the initial geographical knowledge graph in S2 are as follows: Standardized data is split into semantic units according to records, and then input into a pre-trained RoBERTa model for word segmentation, part-of-speech tagging, and semantic parsing to obtain semantic representation vectors for each text segment, enabling the machine to understand the semantics of the data. The semantic representation vector is input into the Seq2Seq framework of the BiLSTM-Attention-CRF network structure to identify named entities based on the data content; The identified named entities will be used as entity nodes in the genealogy or geographical knowledge graph, respectively. The pre-trained syntactic dependency parsing model is invoked to process standardized data fragments containing entities and generate syntactic dependency tree data. Within the syntactic dependency tree, the shortest dependency path between named entity pairs is extracted to clarify the core syntactic logic of entity associations; Based on the shortest dependency path and combined with domain knowledge, a relationship template is constructed; Using the constructed relation template, the named entity pairs are used to extract the relation data; Using entity nodes as the foundation of the knowledge graph, entity relationships as edges connecting nodes, and entity attributes attached to corresponding nodes, a complete initial knowledge graph for genealogy or initial geographical knowledge graph is constructed.
[0008] Preferably, the content of S3 is as follows: The initial genealogical knowledge graph and the initial geographical knowledge graph are embedded into the same preset vector space to obtain the first vector of all entities and relations in the initial genealogical knowledge graph and the second vector of all entities and relations in the initial geographical knowledge graph. Based on the first entity relation matrix corresponding to the initial genealogy knowledge graph and the second entity relation matrix corresponding to the initial geographic knowledge graph, respectively, construct the first KNN tree corresponding to the initial genealogy knowledge graph and the second KNN tree corresponding to the initial geographic knowledge graph. The first and second vectors are normalized based on the first and second KNN trees, respectively. Then, the processed first and second vectors are weighted based on the weight values of the first and second vectors, respectively, to obtain the first and second final alignment vectors corresponding to the initial genealogy knowledge graph and the initial geographic knowledge graph, respectively. Based on the first and second final alignment vectors, entity similarity is compared between the initial genealogical knowledge graph and the initial geographical knowledge graph. Alignment entities are determined in the initial genealogical and geographical knowledge graphs based on the comparison results. Entity alignment is then performed on the initial genealogical and geographical knowledge graphs based on these alignment entities to obtain the target knowledge graph, including: The entity similarity between entities in the initial genealogy knowledge graph and entities in the initial geographic knowledge graph is determined based on the first and second final alignment vectors. The entity similarity is compared with a second preset threshold, and entities with an entity similarity value greater than the second preset threshold are identified as aligned entities.
[0009] Preferably, before constructing the first KNN tree corresponding to the initial genealogical knowledge graph and the second KNN tree corresponding to the initial geographical knowledge graph, the following steps are also included: Collect relational predicates of all entities and relationships in the initial genealogical knowledge graph and the initial geographical knowledge graph; obtain the scoring information of relational predicates generated based on semantic information based on the preset scoring collection interface, and perform weight assignment processing on the relational predicates to obtain the weight values of the first vector and the second vector; or, perform knowledge representation learning and knowledge embedding on the initial genealogical knowledge graph and the initial geographical knowledge graph respectively based on the preset knowledge embedding algorithm to obtain the first calculation vector and the second calculation vector, and calculate the weight values of the relational predicates according to the preset weight value calculation formula and the first calculation vector and the second calculation vector to obtain the weight values of the first vector and the second vector.
[0010] Preferably, constructing the first KNN tree corresponding to the initial genealogical knowledge graph and the second KNN tree corresponding to the initial geographical knowledge graph includes: constructing a first initial relation matrix corresponding to the initial genealogical knowledge graph and a second initial relation matrix corresponding to the initial geographical knowledge graph based on all entities and relations in the initial genealogical knowledge graph and the initial geographical knowledge graph; comparing the weight values of the first vector and the second vector with a first preset threshold, and adjusting all entities and relations in the first initial relation matrix and the second initial relation matrix according to the comparison results to obtain a first entity relation matrix and a second entity relation matrix; and constructing the first KNN tree corresponding to the initial genealogical knowledge graph and the second KNN tree corresponding to the initial geographical knowledge graph based on the central entity and the target neighbor entities of the central entity in the first entity relation matrix and the second entity relation matrix, respectively. Traverse the first entity relation matrix and the second entity relation matrix to determine the neighbor entities corresponding to the central entities in the first entity relation matrix and the second entity relation matrix; sort the similarity of the vector products between the central entities and their corresponding neighbor entities, and determine the target neighbor entities corresponding to each central entity based on the sorting results; construct KNN trees based on the target neighbor entities corresponding to the central entities to obtain the first KNN tree corresponding to the initial genealogy knowledge graph and the second KNN tree corresponding to the initial geographic knowledge graph.
[0011] Preferably, the content of S4 is as follows: Knowledge representation learning is performed based on genealogical knowledge graph and geographical knowledge graph respectively to obtain each first initial vector and each second initial vector; Based on the pre-acquired reference vector set, each first initial vector and each second initial vector are mapped to the reference vector space to obtain each first mapped vector and each second mapped vector; Based on each first mapping vector and each second mapping vector, the knowledge in the genealogical knowledge graph and the geographical knowledge graph is fused to obtain the fused knowledge graph.
[0012] Preferably, the content of S5 is as follows: Based on the fused knowledge graph, vector representations of entities and relations, a reasoning rule base is constructed, and spatiotemporal constraints are defined; Traverse the entities in the genealogy knowledge graph and extract the geographically relevant descriptions from their attributes. For each descriptive text, query the geographic knowledge graph, match the rules in the rule base, and record the hard match results; For soft matching, the low-dimensional vectors generated by S4 are used. For hard-matched, unrelated genealogical entities, the cosine similarity between their vectors and the vectors of all candidate geographic entities is calculated. Set a similarity threshold ,filter The candidate association pairs are then validated a second time, and irrelevant associations are filtered out by combining the information above and below. For the same genealogy entity, if hard matches and soft matches point to the same geographic entity, they are merged into one result; if they point to different entities, they are processed for conflict resolution. The confidence level of the association results is evaluated, and those with hard matches are considered definitive associations. In soft matching Furthermore, the secondary validation passed, indicating a high-confidence association; in soft matching... For associations with medium confidence, it is recommended to verify the label. If secondary verification is questionable, the association is pending confirmation and should be marked as requiring manual confirmation. Based on the rule-based reasoning results and the candidate list retrieved by vector, all candidate associations are traversed, the spatiotemporal attributes of relevant entities are extracted from the knowledge graph, and the results are verified according to the defined spatiotemporal constraints. Output structured matching results.
[0013] The preferred formula for calculating cosine similarity is as follows: .
[0014] The conflict resolution process is as follows: When a genealogical entity matches multiple geographic entities: prioritize hard matches; if no hard match is found, select the entity with the highest similarity from the soft matches; combine external data to assist decision-making.
[0015] Therefore, the above-mentioned method for associating and matching genus information with geographic information has the following beneficial effects: (1) Through standardized processing, knowledge graph construction and association matching, the two types of heterogeneous data are integrated into a unified "genealogy-geography" knowledge network, which solves the problem of "family information without spatial dimension" and "geographical information lacking humanistic connection". (2) Combining inference rules and vector similarity, the accuracy and coverage of matching are balanced; (3) Automatically process entity associations in standardized data; batch matching of inference rules and vector calculations can quickly process large-scale data and improve work efficiency; (4) The correlation matching results can provide data support for academic research, cultural heritage, social governance and other scenarios.
[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0017] Fig. 1 This is a flowchart illustrating a method for associating and matching genus information and geographic information according to the present invention. Fig. 2 This is a schematic flowchart of the entity alignment method according to an embodiment of the present invention; Fig. 3 This is a schematic diagram illustrating the specific entity alignment method of the present invention. Detailed Implementation
[0018] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0019] Please seeFigs. 1-3 A method for associating and matching ancestral information with geographic information includes the following steps: S1. Preprocess the collected genealogical and geographical information to obtain standardized data.
[0020] S2. Extract entities, relationships, and attributes from standardized data to construct initial genealogical knowledge graphs and initial geographical knowledge graphs, respectively.
[0021] Entities include people, families, and events in genealogy, and locations, regions, and terrains in geography; relationships include kinship between people, migration between people and locations, and inclusion relationships between locations and regions; attributes include basic information about entities, such as a person's date of birth and the latitude and longitude of a location.
[0022] The steps for constructing the initial genealogical knowledge graph and the initial geographical knowledge graph are as follows: Standardized data is split into semantic units according to records, and then input into a pre-trained RoBERTa model for word segmentation, part-of-speech tagging, and semantic parsing to obtain semantic representation vectors for each text segment, enabling the machine to understand the semantics of the data. The semantic representation vector is input into the Seq2Seq framework of the BiLSTM-Attention-CRF network structure to identify named entities based on the data content; The identified named entities will be used as entity nodes in the genealogy or geographical knowledge graph, respectively. The pre-trained syntactic dependency parsing model is invoked to process standardized data fragments containing entities and generate syntactic dependency tree data. Within the syntactic dependency tree, the shortest dependency path between named entity pairs is extracted to clarify the core syntactic logic of entity associations; Based on the shortest dependency path and combined with domain knowledge, a relationship template is constructed; Using the constructed relation template, the named entity pairs are used to extract the relation data; Using entity nodes as the foundation of the knowledge graph, entity relationships as edges connecting nodes, and entity attributes attached to corresponding nodes, a complete initial knowledge graph for genealogy or initial geographical knowledge graph is constructed.
[0023] S3. Perform entity alignment on the initial genealogy knowledge graph and the initial geographical knowledge graph.
[0024] The initial genealogical knowledge graph and the initial geographical knowledge graph are embedded into a predefined vector space to obtain the first vector of all entities and relations in the initial genealogical knowledge graph and the second vector of all entities and relations in the initial geographical knowledge graph. A predefined minimum margin loss formula is used to train the first and second initial graph convolutional neural network models to obtain corresponding first and second vector extraction models. The first and second initial graph convolutional neural network models share a weight matrix. The first vector extraction model extracts the first vector of all entities and relations in the first knowledge graph, and the second vector extraction model extracts the second vector of all entities and relations in the second knowledge graph. That is, the first and second knowledge graphs contain some known aligned entities and relations; these triples can effectively serve as a bridge for entity alignment in the first and second knowledge graphs. Aligned triples in the first and second knowledge graphs are then replaced with identical triples. The GCN-Align (Cross-lingual Knowledge Graph Alignment via Graph Convolutional Networks) algorithm is used, employing two GCN (Graph Convolutional Networks) models with shared weight matrices. The GCN model is trained by minimizing the margin loss, as shown in the following formula:
[0025]
[0026] in, and This refers to a small number of known aligned entities and relations in two graphs, where h(e) is the extracted vector. f is the loss value, and d is the vector dimension. It uses the L2 norm. Through the trained vector representations, the entities of the first and second knowledge graphs are embedded into a unified vector space. The entities and relations of the first and second knowledge graphs obtain low-dimensional vector representations in the same representation space. The training process for entity and relation vector representations is optimized using L2 distance minimization.
[0027] Based on the first entity relation matrix corresponding to the initial genealogy knowledge graph and the second entity relation matrix corresponding to the initial geographic knowledge graph, respectively, construct the first KNN tree corresponding to the initial genealogy knowledge graph and the second KNN tree corresponding to the initial geographic knowledge graph. The first and second vectors are normalized based on the first and second KNN trees, respectively. Then, the processed first and second vectors are weighted based on the weight values of the first and second vectors, respectively, to obtain the first and second final alignment vectors corresponding to the initial genealogy knowledge graph and the initial geographic knowledge graph, respectively. Based on the first and second final alignment vectors, entity similarity is compared between the initial genealogical knowledge graph and the initial geographical knowledge graph. Alignment entities are determined in the initial genealogical and geographical knowledge graphs based on the comparison results. Entity alignment is then performed on the initial genealogical and geographical knowledge graphs based on these alignment entities to obtain the target knowledge graph, including: The entity similarity between entities in the initial genealogy knowledge graph and entities in the initial geographic knowledge graph is determined based on the first and second final alignment vectors. The entity similarity is compared with a second preset threshold, and entities with an entity similarity value greater than the second preset threshold are identified as aligned entities.
[0028] Before constructing the first KNN tree corresponding to the initial genealogical knowledge graph and the second KNN tree corresponding to the initial geographical knowledge graph, the following steps are also included: Collect relational predicates of all entities and relationships in the initial genealogical knowledge graph and the initial geographical knowledge graph; obtain the scoring information of relational predicates generated based on semantic information based on the preset scoring collection interface, and perform weight assignment processing on the relational predicates to obtain the weight values of the first vector and the second vector; or, perform knowledge representation learning and knowledge embedding on the initial genealogical knowledge graph and the initial geographical knowledge graph respectively based on the preset knowledge embedding algorithm to obtain the first calculation vector and the second calculation vector, and calculate the weight values of the relational predicates according to the preset weight value calculation formula and the first calculation vector and the second calculation vector to obtain the weight values of the first vector and the second vector.
[0029] Constructing a first KNN tree corresponding to the initial genealogical knowledge graph and a second KNN tree corresponding to the initial geographical knowledge graph includes: constructing a first initial relation matrix corresponding to the initial genealogical knowledge graph and a second initial relation matrix corresponding to the initial geographical knowledge graph based on all entities and relations in the initial genealogical knowledge graph and the initial geographical knowledge graph; comparing the weight values of the first vector and the second vector with a first preset threshold, and adjusting all entities and relations in the first initial relation matrix and the second initial relation matrix according to the comparison results to obtain a first entity relation matrix and a second entity relation matrix; and constructing the first KNN tree corresponding to the initial genealogical knowledge graph and the second KNN tree corresponding to the initial geographical knowledge graph based on the central entity and the target neighbor entities of the central entity in the first entity relation matrix and the second entity relation matrix, respectively. Traverse the first entity relation matrix and the second entity relation matrix to determine the neighbor entities corresponding to the central entities in the first entity relation matrix and the second entity relation matrix; sort the similarity of the vector products between the central entities and their corresponding neighbor entities, and determine the target neighbor entities corresponding to each central entity based on the sorting results; construct KNN trees based on the target neighbor entities corresponding to the central entities to obtain the first KNN tree corresponding to the initial genealogy knowledge graph and the second KNN tree corresponding to the initial geographic knowledge graph.
[0030] S4. Based on the entity alignment results, the initial genealogy knowledge graph and the initial geographic knowledge graph are merged into a unified and consistent knowledge graph.
[0031] Knowledge representation learning is performed based on genealogical knowledge graph and geographical knowledge graph respectively, to obtain each first initial vector and each second initial vector.
[0032] The classic translation model TransE can be used. TransE is a distributed vector representation based on entities and relations, utilizing the translation invariance of word vectors to map entities and relations to a low-dimensional vector space. TransE treats the relation in the triple instance "entity-relation-entity" (head-relation-tail) as a vector between the head entity (head) and the tail entity (tail). By continuously adjusting h, r, and t (the vectors corresponding to head, relation, and tail), (h+r) is made as equal to t as possible, i.e., htr=t. It should be noted that in a triplet instance, either of the two entities can be used as the head entity and the other as the tail entity.
[0033] The distance function of TransE can be set as d(h+r,t), which measures the distance between (h+r) and t. The smaller the function value, the more reasonable the triplet. TransE defines the following objective function using the maximum margin method:
[0034]
[0035] in, S is the interval value, S is the triple from the knowledge base, and S' is a manually constructed negatively sampled triple. The h or t is replaced by a random value. or However, they are not replaced simultaneously. For S, the smaller the distance function (h+r,t), the better, while for... As for (h+r,t), the larger the better.
[0036] Based on the pre-acquired reference vector set, each first initial vector and each second initial vector are mapped to the reference vector space to obtain each first mapped vector and each second mapped vector; Based on each first mapping vector and each second mapping vector, the knowledge in the genealogy knowledge graph and the geographical knowledge graph is fused to obtain the fused knowledge graph. The first initial vector is the vector representation of the entity or relation in the genealogical knowledge graph in the first vector space.
[0037] The second initial vector is the vector representation of entities or relationships in the geographic knowledge graph in the second vector space.
[0038] A reference vector is the vector representation of an entity in a reference knowledge graph within a reference vector space.
[0039] Genealogical knowledge graphs, geographical knowledge graphs, and reference knowledge graphs are built based on different data sources.
[0040] For any first instance in the first genealogical knowledge graph and any second instance in the second geographical knowledge graph, if it is determined that the tail entities in any first instance and any tail entities in any second instance have a relationship in the reference knowledge graph, then they are treated as a pre-fused entity pair. Subsequently, bidirectional supervised training is performed on each initial vector based on the reference vector set to obtain mapping vectors. For each pre-fused entity pair, the first distance between the reference vectors of the two tail entities and the second distance between the mapping vectors of the two head entities are calculated. If the difference between the two is less than a preset threshold, then in the fused graph, it is determined that the two head entities have the same relationship as the tail entity pair.
[0041] S5. Perform representation learning and reasoning computation on the fused knowledge graph to achieve deep correlation analysis between genealogical information and geographical information, and output matching results.
[0042] Based on the fused knowledge graph, entity, and relation vector representations, a reasoning rule base is constructed, and spatiotemporal constraints are defined. The reasoning rule base defines rules for genealogical and geographical associations based on domain knowledge, with core association types such as: genealogical figures. Geographical location (e.g., the connection between "Zhang San" and "Jinan City" as "ancestral home"); genealogy incidents Geographical region / topography (e.g., the connection between "migration during the Guangxu era" and "destination" in "North China"); family genealogy Geographical location (such as the "origin" connection between the "Zhang family" and "Zhangjiazhuang").
[0043] Traverse the entities in the genealogy knowledge graph and extract the geographically relevant descriptions from their attributes. For each descriptive text, query the geographic knowledge graph, match the rules in the rule base (as shown in Table 1), and record the hard match results; Table 1 contains rules.
[0044] For soft matching, the low-dimensional vectors generated by S4 are used. For hard-matched, unrelated genealogical entities, the cosine similarity between their vectors and the vectors of all candidate geographic entities is calculated. Set a similarity threshold ,filter The candidate association pairs are then validated a second time, and irrelevant associations are filtered out by combining contextual information. The cosine similarity calculation formula is as follows: .
[0045] For the same genealogical entity, if hard matches and soft matches point to the same geographic entity, they are merged into a single result; if they point to different entities, conflict resolution is initiated. The conflict resolution process is as follows: When a genealogical entity matches multiple geographic entities: prioritize hard matches; if no hard match is found, select the entity with the highest similarity from the soft matches; combine external data to assist decision-making.
[0046] The confidence level of the association results is evaluated, and those with hard matches are considered definitive associations. In soft matching Furthermore, the secondary validation passed, indicating a high-confidence association; in soft matching... For associations with medium confidence, it is recommended to verify the label. If secondary verification is questionable, the association is pending confirmation and should be marked as requiring manual confirmation. Based on the rule-based reasoning results and the candidate list retrieved by vectors, all candidate associations are traversed, the spatiotemporal attributes of relevant entities are extracted from the knowledge graph, and the results are verified according to the defined spatiotemporal constraints.
[0047] The output structured matching results are shown in Table 2.
[0048] Table 2 Output Results
[0049] Therefore, the present invention adopts the above-mentioned method for associating genealogical information with geographic information. This method achieves deep integration of "genealogical and cultural information" and "geospatial information" through technological innovation. It has significant advantages in enhancing data value, reducing processing costs, and supporting multi-scenario applications, and provides strong technical support for historical research, cultural inheritance and social application.
[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for associating and matching genus information with geographic information, characterized in that, The method comprises the following steps: S1, preprocessing the collected genealogy information and geographical information to obtain standardized data; S2, extracting entities, relationships and attributes from the standardized data to construct a genealogy initial knowledge graph and a geographical initial knowledge graph respectively; S3, performing entity alignment on the genealogy initial knowledge graph and the geographical initial knowledge graph; S4, based on the results of entity alignment, fusing the genealogy initial knowledge graph and the geographical initial knowledge graph into a unified and consistent knowledge graph; S5, performing representation learning and reasoning calculation on the fused knowledge graph to realize deep correlation analysis of genealogy information and geographical information and output matching results.
2. The method of claim 1, wherein the method further comprises: In S2, the entities include persons, families, events in genealogy and places, regions, terrains in geography; the relationships include kinship between persons, migration relationship between persons and places, and inclusion relationship between places and regions; and the attributes include basic information of the entities.
3. The method of claim 2, wherein the geographical information is obtained from a geographical information system. In S2, the steps of constructing the genealogy initial knowledge graph and the geographical initial knowledge graph are as follows: The standardized data is split into semantic units according to records, input into a pre-trained RoBERTa model, and subjected to word segmentation, part-of-speech tagging and semantic parsing processing to obtain semantic representation vectors of each text segment, so that the machine understands the data semantics; The semantic representation vectors are input into a BiLSTM-Attention-CRF network structure Seq2Seq framework to identify named entities according to data content; The identified named entities are taken as entity nodes of the genealogy or geographical knowledge graph respectively; A pre-trained syntactic dependency analysis model is called to process the standardized data segments containing entities to generate syntactic dependency tree data; In the syntactic dependency tree, the shortest dependency path between pairs of named entities is mined to clarify the core syntax logic of entity correlation; Based on the shortest dependency path, relationship templates are constructed in combination with domain knowledge; The constructed relationship templates are used to extract relationships from pairs of named entities to obtain entity relationship data; The entity nodes are taken as the basis of the knowledge graph, the entity relationships are taken as the edges connecting the nodes, and the entity attributes are mounted to the corresponding nodes to construct a complete genealogy initial knowledge graph or a geographical initial knowledge graph.
4. The method of claim 3, wherein the geographical information is obtained from a geographical information database. The content of S3 is as follows: Embed the genealogy initial knowledge graph and the geographical initial knowledge graph into a preset same vector space to obtain first vectors of all entities and relationships in the genealogy initial knowledge graph and second vectors of all entities and relationships in the geographical initial knowledge graph; Based on each center entity and the target neighbor entity of the center entity in the first entity relationship matrix corresponding to the genealogy initial knowledge graph and the second entity relationship matrix corresponding to the geographical initial knowledge graph, construct a first KNN tree corresponding to the genealogy initial knowledge graph and a second KNN tree corresponding to the geographical initial knowledge graph; Based on the first KNN tree and the second KNN tree, normalize the corresponding first vector and second vector respectively, and then perform vector weighting processing on the processed first vector and the processed second vector based on the weight values of the first vector and the weight values of the second vector respectively to obtain a first final alignment vector and a second final alignment vector corresponding to the genealogy initial knowledge graph and the geographical initial knowledge graph respectively; According to the first final alignment vector and the second final alignment vector, the entity similarity between the entities in the genealogy initial knowledge graph and the entities in the geographical initial knowledge graph is determined. The entity similarity is compared with the second preset threshold, and the entity with the entity similarity value greater than the second preset threshold is determined as the aligned entity. The genealogy initial knowledge graph and the geographical initial knowledge graph are embedded into a preset same vector space to obtain a first vector of all entities and relationships in the genealogy initial knowledge graph and a second vector of all entities and relationships in the geographical initial knowledge graph, including:
5. The method of claim 4, wherein the geographical information is obtained from a geographical information database. The first initial graph convolutional neural network model and the second initial graph convolutional neural network model are trained by using a preset minimum interval loss formula to obtain corresponding first vector extraction models and second vector extraction models; the first initial graph convolutional neural network model and the second initial graph convolutional neural network model share a weight matrix; the first vector extraction model is used to extract the first vector of all entities and relationships in the genealogy initial knowledge graph, and the second vector extraction model is used to extract the second vector of all entities and relationships in the geographical initial knowledge graph; The vector product similarity between all vectors in the first vector and the second vector is determined. Before constructing the first KNN tree corresponding to the genealogy initial knowledge graph and the second KNN tree corresponding to the geographical initial knowledge graph, the method further includes:
6. The method of claim 4, wherein the geographical information is obtained from a geographical information system. Collecting the relationship predicates of all entities and relationships in the genealogy initial knowledge graph and the geographical initial knowledge graph; obtaining scoring information of the relationship predicates generated based on semantic information based on a preset scoring collection interface, performing weight assignment processing on the relationship predicates to obtain weight values of the first vector and the second vector; or, performing knowledge representation learning and knowledge embedding on the genealogy initial knowledge graph and the geographical initial knowledge graph based on a preset knowledge embedding algorithm to obtain first calculation vectors and second calculation vectors, and calculating the weight values of the relationship predicates according to a preset weight value calculation formula and the first calculation vectors and the second calculation vectors to obtain the weight values of the first vector and the second vector. 7. The method of claim 6, wherein the geographical information is obtained from a geographical information system (GIS) database. The first KNN tree corresponding to the initial knowledge graph of the genealogy and the second KNN tree corresponding to the initial knowledge graph of the geography are constructed, including: constructing a first initial relationship matrix corresponding to the initial knowledge graph of the genealogy and a second initial relationship matrix corresponding to the initial knowledge graph of the geography based on all entities and relationships in the initial knowledge graph of the genealogy and the initial knowledge graph of the geography; comparing the weight values of the first vector and the weight values of the second vector with a first preset threshold, and adjusting all entities and relationships in the first initial relationship matrix and the second initial relationship matrix according to the comparison results, to obtain a first entity relationship matrix and a second entity relationship matrix; based on the center entities in the first entity relationship matrix and the second entity relationship matrix and the target neighbor entities of the center entities, the first KNN tree corresponding to the initial knowledge graph of the genealogy and the second KNN tree corresponding to the initial knowledge graph of the geography are constructed; The center entities corresponding to the neighbor entities in the first entity relationship matrix and the second entity relationship matrix are determined by traversing the first entity relationship matrix and the second entity relationship matrix; the vector product similarity between the center entities and the corresponding neighbor entities is sorted, and the target neighbor entities corresponding to each center entity are determined according to the sorting results; and the KNN tree is constructed according to the target neighbor entities corresponding to the center entities to obtain the first KNN tree corresponding to the initial knowledge graph of the genealogy and the second KNN tree corresponding to the initial knowledge graph of the geography.
8. The method of claim 4, wherein the geographical information is obtained from a geographical information system. S4 includes the following content: Perform knowledge representation learning based on the genealogy knowledge graph and the geographical knowledge graph to obtain each first initial vector and each second initial vector; Map each first initial vector and each second initial vector to a reference vector space based on a reference vector set obtained in advance to obtain each first mapping vector and each second mapping vector; Fuse the knowledge in the genealogy knowledge graph and the geographical knowledge graph according to each first mapping vector and each second mapping vector to obtain a fused knowledge graph.
9. The method of claim 1, wherein the method further comprises: S5 includes the following content: Construct a reasoning rule library based on the fused knowledge graph, vector representation of entities and relationships, and define a space-time constraint; Traverse the entities in the genealogy knowledge graph to extract the geographical related descriptions in the attributes thereof, For each description text, query the geographical knowledge graph, match the rules in the rule library, and record the hard matching results; Soft matching is performed, the low-dimensional vector generated by S4 is called to associate the genealogy entity not associated by hard matching, and the cosine similarity of the vector and all candidate geographical entity vectors is calculated , a similarity threshold is set , the candidate associated pairs are screened , and secondary verification is performed on the candidate associated pairs, and irrelevant associations are filtered in combination with upper and lower information For the same genealogy entity, if the hard matching and the soft matching point to the same geographical entity, merge them into one result; if they point to different entities, enter the conflict processing; Evaluate the confidence of the associated results, and the hard matching results are determined associations; In soft matching and secondary verification passes is a high-confidence association; in soft matching is a medium-confidence association; or secondary verification is questionable is a pending association; According to the rule reasoning results and the vector retrieval candidate list, traverse all candidate associations, extract the space-time attributes of the related entities from the knowledge graph, and verify them according to the defined space-time constraint; Output the structured matching results.
10. The method of claim 9, wherein the geographical information is obtained from a geographical information system (GIS) database. The cosine similarity calculation formula is as follows: ； The conflict processing process is as follows: When the genealogy entity matches multiple geographical entities: prefer the hard matching result; if there is no hard matching, select the entity with the highest similarity in the soft matching; and make a decision with the aid of external data.