Social group emergency spatio-temporal knowledge graph construction method

Abstract

The invention relates to the technical field of knowledge maps, in particular to a social group emergency spatio-temporal knowledge map construction method, which comprises the following steps of: constructing a social group emergency knowledge expression model considering spatio-temporal characteristics; constructing a social group emergency spatio-temporal knowledge graph based on the multi-source heterogeneous data; and evaluating the space-time mapping knowledge domain of the social group emergencies. According to the method, semantic and spatio-temporal expression of the social group emergencies is realized in a spatio-temporal knowledge graph form, so that an emergency decision maker can know spatio-temporal semantic development veins of the social group emergencies, and effective auxiliary support is provided for making an emergency decision scheme of the social group emergencies; and meanwhile, the defect that a space-time knowledge graph lacks a social group emergency space-time knowledge graph is overcome.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, specifically a method for constructing a spatiotemporal knowledge graph for sudden social group events. Background Technology

[0002] In today's era of deep integration of digitalization and information technology, spatiotemporal knowledge graphs, as a technological means to effectively integrate and present complex information, are widely and deeply penetrating numerous research and application fields. Within the scope of research on social mass emergencies, while relevant academic exploration and technical practices have made some progress, existing problems remain significant, severely restricting the accurate understanding and efficient response to such events, as well as the in-depth development of related fields.

[0003] Existing methods for modeling social mass emergencies lack the ability to express spatiotemporal evolution: Currently, ontology-based methods dominate knowledge modeling for social mass emergencies. Ontology models focus on semantic evolution, describing the basic characteristics, concepts, and semantic relationships of events, thus achieving a structured representation of events to some extent. However, this method has significant limitations. Social mass emergencies are dynamic processes with extremely complex and closely related changes in time and space. Ontology models struggle to comprehensively and meticulously depict the specific evolution of events at different time points and geographical locations, failing to effectively present the spatiotemporal dynamics of events. For example, consider a resident protest triggered by urban planning. Temporally, the protest may progress from small-scale gatherings to large-scale demonstrations, and then gradually subside or transform into other forms. Spatially, the protests may begin around the planned area and gradually spread to the city center, continuously expanding their impact. However, existing ontology models can only record basic information about events, such as event type and participants, and it is difficult to accurately present the dynamic changes in these spatiotemporal dimensions. This leads to emergency decision-makers lacking a deep understanding of the spatiotemporal evolution of events when formulating response strategies, affecting the scientific nature and timeliness of decision-making.

[0004] Current research on the construction of spatiotemporal knowledge graphs for social mass emergencies is limited. In this field, current research primarily focuses on natural disasters (such as earthquakes and floods) and accidents (such as traffic accidents and industrial accidents). While these studies provide strong support for responding to these types of disasters, relatively little attention has been paid to social mass emergencies. Social mass emergencies differ fundamentally from natural disasters and accidents, involving complex social relationships, conflicts of interest, and human factors. For example, social mass emergencies may be triggered by a variety of factors, including unequal distribution of economic benefits, improper policy implementation, and cultural conflicts. Furthermore, the development of these events is often strongly influenced by social factors such as public opinion and public sentiment. These unique characteristics make it difficult to directly apply spatiotemporal knowledge graph construction methods used for natural disasters and accidents to social mass emergencies. The lack of research specifically on spatiotemporal knowledge graph construction for social mass emergencies hinders the rapid and accurate integration and analysis of relevant information when facing such events. It also makes it difficult to reveal the inherent laws and development trends of events from a spatiotemporal perspective, thus affecting the ability to prevent, monitor, and respond to social mass emergencies.

[0005] The heterogeneity of data sources increases the difficulty of research: Data related to social mass emergencies comes from a wide range of sources and formats, encompassing government statistics, news media reports, social media information, academic research literature, and on-site survey data. This data exhibits significant heterogeneity, with substantial differences in structure, format, quality, and semantics across different sources. Government data is typically in structured tabular form, exhibiting strong data standardization but potentially slow update speeds; news media reports are primarily text-based, containing rich event details but may contain subjective biases; social media information is fragmented, highly real-time, and of inconsistent quality. This heterogeneous data characteristic makes data acquisition, integration, and analysis exceptionally difficult. Constructing a spatiotemporal knowledge graph requires substantial human and material resources to address data compatibility and consistency issues; otherwise, the accuracy and completeness of the knowledge graph will be affected, hindering in-depth research and effective responses to social mass emergencies.

[0006] Lack of unified standards and norms: Currently, there are no unified standards and norms in the field of constructing spatiotemporal knowledge graphs for social mass emergencies. This leads to differences in concept definitions, data representations, and relationship modeling among knowledge graphs constructed by different research teams or institutions, making it difficult to achieve knowledge sharing and integration. For example, the definition of "social mass emergencies" may vary in scope among different studies; the formats and precision used to represent the time and location of events also differ. This lack of unified standards not only limits the application effectiveness of knowledge graphs in different scenarios but also hinders the exchange and promotion of research results in this field, making it difficult to form a systematic research framework and affecting the overall development of spatiotemporal knowledge graph construction technology for social mass emergencies.

[0007] In light of the aforementioned issues, research on the construction methods of spatiotemporal knowledge graphs for social mass emergencies is urgently needed. By organically integrating social mass emergencies with spatiotemporal knowledge graph theory and methods, we can fully mine the semantic information within the events and deeply study their evolutionary patterns and processes across the spatiotemporal dimensions. This can not only effectively compensate for the shortcomings of previous research but also provide emergency decision-makers with comprehensive and accurate information support, helping them formulate more scientific and effective emergency decision-making plans, thereby improving society's ability to respond to and manage mass emergencies. Summary of the Invention

[0008] The purpose of this invention is to provide a method for constructing a spatiotemporal knowledge graph of social group emergencies, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a spatiotemporal knowledge graph for social group emergencies, comprising the following steps:

[0010] S1: Construct a knowledge representation model for social group emergencies that takes into account spatiotemporal characteristics. From a hierarchical framework perspective, the knowledge representation model for social group emergencies is divided into five levels from top to bottom: concept layer, object layer, state layer, attribute layer, and relationship layer;

[0011] S2: Based on the knowledge representation model of social group emergencies with spatiotemporal characteristics described in S1, Protégé is selected for ontology construction, and Neo4j database is selected for ontology storage.

[0012] S3: Using the knowledge representation model of social group emergencies constructed in S2 as the knowledge framework, a spatiotemporal knowledge graph of social group emergencies is constructed based on multi-source heterogeneous data through data acquisition and preprocessing, knowledge extraction, knowledge fusion and knowledge storage.

[0013] S4: Evaluate the coverage and accuracy of the spatio-temporal knowledge graph of social group emergencies generated in S3.

[0014] Preferably, in step S1, based on ontology reuse and self-defined concepts, establish the concept layer of social group emergencies; according to the constituent elements and spatio-temporal evolution laws of social group emergencies, study and establish the object layer and state layer of social group emergencies; summarize and establish the attribute layer for different state characteristic attributes; model the relationships existing in the knowledge graph. Finally, establish a knowledge expression model of social group emergencies that takes into account the spatio-temporal evolution process.

[0015] Preferably, in step S2, establish corresponding nodes in Neo4j and define node types and node attributes; node types include concept nodes, object nodes, state nodes, and attribute nodes; node attributes include the node name, i.e., the name, definition, and spatio-temporal characteristics in the classification system of social group emergencies. The node name is the unique identifier of the node. The relationship between nodes represents a subordinate relationship, expressed by inclusion and being included.

[0016] Preferably, in step S3, for data acquisition, propose a data acquisition method of multi-strategy learning. Obtain event elements and attribute knowledge in the structured data table through direct mapping; for the semi-structured data of encyclopedia web pages, by parsing the web page structure, design a web page element template matching model, and combine with web crawler technology to obtain domain-related data; for unstructured data such as text on news websites, use a search engine, set retrieval keywords, and use web crawlers to obtain data.

[0017] Preferably, in step S3, for data preprocessing, the preprocessing of structured data includes data cleaning, screening and other preprocessing operations; the preprocessing of unstructured text data includes sentence splitting, word segmentation, and stop word filtering. Sentence splitting uses a regular expression sentence splitting method based on punctuation marks to split sentences by identifying punctuation marks such as full stops, question marks, and exclamation marks; word segmentation adopts HanLP for word segmentation of text sentences. Stop word filtering uses the Harbin Institute of Technology stop word list to filter special symbols such as "de", "le", "zhe", and "#$@" in the text.词性标注。利用Lable Studio工具对经过前期筛选停用词过滤后的数据进行人工标注,并导出为BIO词性标注。

[0018] Preferably, in step S3, for knowledge extraction, it includes entity recognition and relationship extraction. Entity recognition uses the deep learning BERT-BiLSTM-CRF model method to identify the results of词性标注,获取知识文本数据中的人名、组织 / 机构名、地名、时间、行为、诱因、类型等;关系提取采用规则与语法框架的方法抽取实体间的语义关系,时空关系。

[0019] Preferably, in step S3, knowledge fusion is performed based on the cosine similarity calculation of word vectors, using clustering and a threshold merging method. The cosine similarity can be expressed as:

[0020]

[0021] The Similarity(A,B) value ranges from [0,1]. The closer the similarity is to 1, the more similar the two texts are semantically. If the similarity is higher than the threshold, the two texts are considered identical and are merged.

[0022] Preferably, in step S3, knowledge storage is achieved by obtaining entities and relationships between entities through the aforementioned claims 4-7. Source data with different structures is transformed into structured knowledge triple data in the form of (entity, relation, entity) or (entity, attribute, attribute value) and stored using the graph database Neo4j.

[0023] Preferably, in step S4, the generated spatiotemporal knowledge graph of social group emergencies is evaluated using two metrics: coverage and accuracy. Raw data from different sources is randomly sampled, knowledge triples are manually constructed, and then compared with the knowledge transformed from the current data in the knowledge graph. Coverage can be represented as:

[0024]

[0025] The dataset of triples extracted from the knowledge graph is K, and the dataset of triples constructed manually is T. The higher the coverage, the better the quality of the spatiotemporal knowledge graph of social group emergencies.

[0026] Accuracy is the percentage of correct triples extracted from the knowledge graph out of the total number of triples, and can be expressed as:

[0027]

[0028] The number of random samples is T, and the number of errors is T. wrong The higher the accuracy, the better the quality of the spatiotemporal knowledge graph for social group emergencies.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] 1. This method for constructing a spatiotemporal knowledge graph for social mass emergencies provides comprehensive and accurate basis for emergency decision-making: The spatiotemporal knowledge graph for social mass emergencies constructed in this invention presents the development trajectory of the event in time and space dimensions, as well as the complex relationships between various entities, in an intuitive and structured form. Emergency decision-makers can use this graph to quickly grasp the overall picture of the event and clearly understand its spatiotemporal distribution characteristics at different stages, such as the starting point, propagation path, and scale changes at different times. This helps them accurately predict the development trend of the event, thereby formulating more targeted and timely emergency decision-making plans. When facing mass protests triggered by a policy, by using the graph to display the location of the event, the changes in the number of participants over time, and the related information with surrounding areas, decision-makers can accurately assess the risk of the event's spread, allocate resources in a timely manner, and take effective guidance or control measures to prevent further escalation and minimize negative impacts.

[0031] 2. This method for constructing a spatiotemporal knowledge graph for social mass emergencies fills a gap in academic research and technological application: Currently, the application of spatiotemporal knowledge graphs in the field of social mass emergencies is still in its early stages, and related research results are relatively scarce. The construction method innovatively proposed in this invention fills this gap and lays a solid foundation for subsequent research. On the one hand, at the academic research level, it provides a new theoretical framework and research perspective for the spatiotemporal evolution analysis of social mass emergencies, promotes the development of interdisciplinary research, and facilitates the deep integration of computer science, sociology, management, and other disciplines in this field. On the other hand, in terms of technological application, it provides key technical support for the development of relevant emergency management systems and platforms, which helps to improve the level of information management of social mass emergencies and enhance society's emergency response capabilities.

[0032] 3. This method for constructing a spatiotemporal knowledge graph for social mass emergencies effectively integrates multi-source heterogeneous data: Data sources for social mass emergencies are extensive and diverse, including structured government statistics, semi-structured web page information, and a large amount of unstructured news reports and social media content. This invention employs a multi-strategy learning data acquisition method and a systematic data preprocessing workflow, which can efficiently extract valuable information from various data sources and transform it into unified structured knowledge triples, stored in the graph database Neo4j. This not only solves the problem of multi-source heterogeneity of data but also ensures the integrity and accuracy of the knowledge graph. By integrating these scattered data, the knowledge graph can more comprehensively reflect the true situation of the event, providing a rich data foundation for in-depth analysis and research. When analyzing a large-scale strike, by integrating government-released labor relations data, news media reports, and workers' demands expressed on social media, the knowledge graph can reveal the complex interest relationships, social opinion trends, and potential risk factors behind the event.

[0033] 4. This method for constructing a spatiotemporal knowledge graph for social group emergencies possesses excellent scalability and adaptability: The knowledge representation model constructed in this invention adopts a layered architecture, from the concept layer, object layer, state layer, attribute layer to the relationship layer, with clear logic and interrelationships between each layer. This layered structure makes the model highly scalable. When facing new types of social group emergencies or new research needs, new concepts, objects, states, attributes, or relationships can be easily added to the corresponding layers. When dealing with emerging online group events, relevant concepts can be added to the concept layer, objects such as network platforms and virtual accounts can be included in the object layer, and network propagation relationships can be defined in the relationship layer. Simultaneously, in the data processing flow, whether it is data acquisition, preprocessing, or knowledge extraction, fusion, and storage, a modular design concept is adopted. This allows for flexible adjustment and optimization of each module when facing different types of data or technological updates, ensuring that the entire construction method always maintains high efficiency and adaptability, meeting the ever-changing needs of practical application scenarios. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a general framework diagram of the present invention;

[0036] Figure 2This invention provides a knowledge representation model for social group emergencies that takes into account spatiotemporal characteristics. Detailed Implementation

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0038] Please see Figures 1-2 , the present invention provides a technical solution:

[0039] A method for constructing a spatiotemporal knowledge graph of social group emergencies includes the following steps:

[0040] S1: Constructing the conceptual layer of the knowledge representation model for social mass emergencies. This involves selectively reusing events and environments from the ABC ontology, types and roles from the SEM, and drawing on ontology models specific to the emergency domain to ultimately form the core concepts of social mass emergencies. The conceptual layer of the knowledge representation model for social mass emergencies includes event environment, event, and event roles. The event environment refers to the natural and social environment affected by or influencing the event; the event refers to the event itself, including event type, event triggers, and event level; and the event role refers to the people or things affected by the event. Event levels are divided into extremely serious social mass emergencies (Level I), serious social mass emergencies (Level II), relatively serious social mass emergencies (Level III), and general social mass emergencies (Level IV).

[0041] S2: Construct the object layer of the knowledge representation model for social mass emergencies. Further refine the included objects based on the concepts in S1. The objects of the knowledge representation model for social mass emergencies not only include the event itself, but also factors influencing the event's changes such as terrain, landforms, weather, population, and economy, as well as facilities affected by the event, including housing, transportation, communication, energy, and public services. If the social mass emergency is E, the event itself is e, and the environment of the social mass emergency is O. e O refers to various people and things affected by sudden social mass incidents. b Then it can be expressed as:

[0042] E = <e,O e O b >

[0043] S3: Constructing the state layer of a knowledge representation model for social group emergencies. Based on the specific changes, actions, or records of objects in S2 over time and space, these are decomposed into different states. The set of states represents the changes in the objects, and the different states of the objects represent the development process of the event itself. If O represents an object related to a social group emergency, S represents the state of object O, and i represents the number of states contained in the object, then it can be represented as:

[0044] O = <S1,S2,S3,…,S i >

[0045] The evolution of sudden social events is constituted by the spatiotemporal connection of different object states. Connecting different states of the same object sequentially in chronological order forms a state sequence of that object. Let P... O S is the state sequence of the object. Oj If it is an object state, it can be represented as:

[0046] P O = O1 ,S O2 ,S O3 ,…,S Oj >

[0047] The sequence of states of all objects constitutes the entire evolution of the event.

[0048] S4: Constructing the Attribute Layer of the Knowledge Representation Model for Social Group Emergencies. The attribute layer of the knowledge representation model for social group emergencies defines the time, location, attributes, and behaviors of the event. Time (T) and space (S) are the prerequisites for the existence of the object and also the basic framework for representing the object. Attribute A is used to characterize the properties inherent in the object itself, and behavior B is used to describe the various activities and actions performed by the object. Therefore, the attributes of the object are represented as follows:

[0049] C = <T,S,A,B>。

[0050] In sudden social events involving large groups, individuals exist in different spatiotemporal states. Their spatial location and attributes / behaviors change over time. Therefore, under specific temporal and spatial conditions, the individual's attributes and behavioral characteristics together constitute the different states of that individual. Let's assume they are at time f... t and space f s The specific properties f exhibited by the object a and behavior f a The state of an object can be represented as:

[0051] S1 = <f t ,f s ,f a ​,f b >

[0052] S5: Construct the relationship layer of the knowledge expression model for social group emergencies. In the knowledge graph, a relationship can be regarded as an edge between two nodes, which is used to connect concept nodes, element nodes and attribute nodes, and can be expressed as:

[0053] R = <O1, r, O2>.

[0054] Among them, O1 and O2 are different objects, and r represents the type of relationship existing between different objects. Social group emergencies mainly include time relationships, spatial relationships and semantic relationships, etc.

[0055] S6: Based on S1 - S5, select Protégé to construct the knowledge expression model for social group emergencies, and select the Neo4j database for storage. Establish corresponding nodes in Neo4j, and define node types and node attributes; node types include concept nodes, object nodes, state nodes, and attribute nodes; node attributes include the node name, that is, the name, definition, and spatio-temporal characteristics in the classification system of social group emergencies. The node name is the unique identifier of the node, and the relationship between nodes represents a subordination relationship, which is expressed by inclusion and being included.

[0056] S7: Collect data such as social group emergency case databases, encyclopedia data, news website data, case compilations, and literature data. For knowledge sources with multiple structures, propose a data acquisition method for multi-strategy learning. Obtain event elements and attribute knowledge in structured data tables through direct mapping; for semi-structured data on encyclopedia web pages, design a web element template matching model by parsing the web page structure, and combine web crawler technology to obtain domain-related data; for unstructured data such as text on news websites, use a search engine, set retrieval keywords, and use web crawlers to obtain data.

[0057] S8: For structured data processing, it includes preprocessing operations such as data cleaning and screening; for preprocessing of unstructured text data, it includes sentence splitting, word segmentation, and stop word filtering. Sentence splitting for text uses a regular expression sentence splitting method based on punctuation marks, which splits sentences by identifying punctuation marks such as full stops, question marks, and exclamation marks; word segmentation uses HanLP for word segmentation of text sentences. Stop word filtering uses the Harbin Institute of Technology stop word list to filter special symbols such as "de", "le", "zhe", and "#$@" in the text.词性标注. Use the LableStudio tool to manually annotate the data after pre-filtering stop words in the previous stage, and export it as BIO词性标注.

[0058] S9: Entity Recognition. For structured and semi-structured data, the format is converted into a triplet structure that can be unified by the model framework. For a large amount of unstructured text data, the deep learning BERT-BiLSTM-CRF model is used to recognize the part-of-speech tagging results, extracting names of people, organizations / institutions, places, times, behaviors, causes, types, etc. from the knowledge text data. First, the labeled corpus is processed by a BERT pre-trained language model to obtain corresponding word vectors. Then, the word vectors are input into the BiLSTM module for further processing. Finally, the CRF module decodes the output of the BiLSTM module to obtain a predicted labeled sequence. Then, the entities in the sequence are extracted and classified, thus completing the entire entity recognition process.

[0059] S10: Extract semantic and spatiotemporal relationships between entities using a rule-based and grammatical framework. First, define the set of relationships to be extracted; second, use HanLP part-of-speech tagging to extract entities and word phrases with relationships; finally, when a word is encountered in the relationship set, select the entity that is closest to the left and right of that word.

[0060] S11: Cosine similarity calculation based on word vectors, employing methods such as clustering and threshold merging for knowledge fusion. First, entities are segmented into n words in the semantic space, and the word frequency of each word is calculated to form word vectors A and B for the entity, where A and B are both n-dimensional vectors. The cosine similarity between A and B can then be expressed as:

[0061]

[0062] The Similarity(A,B) value ranges from [0,1]. The closer the similarity is to 1, the more similar the two texts are semantically. If the similarity is higher than the threshold, the two texts are considered identical and are merged.

[0063] S12: Through S7-S11, entities and relationships between entities are obtained, and source data with different structures are transformed into structured knowledge triple data in the form of (entity, relation, entity) or (entity, attribute, attribute value) and stored using the graph database Neo4j.

[0064] S13: The generated spatiotemporal knowledge graph of social group emergencies is evaluated using two metrics: coverage and accuracy. Raw data from different sources is randomly sampled, knowledge triples are manually constructed, and then compared with the knowledge transformed from the current data in the knowledge graph. Coverage can be represented as:

[0065]

[0066] The dataset of triples extracted from the knowledge graph is K, and the dataset of triples constructed manually is T. The higher the coverage, the better the quality of the spatiotemporal knowledge graph of social group emergencies.

[0067] Accuracy is the percentage of correct triples extracted from the knowledge graph out of the total number of triples, and can be expressed as:

[0068]

[0069] The number of random samples is T, and the number of errors is T. wrong The higher the accuracy, the better the quality of the spatiotemporal knowledge graph for social group emergencies.

[0070] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0071] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a spatiotemporal knowledge graph for sudden social group events, characterized in that, Includes the following steps: S1: Construct a knowledge representation model for social group emergencies that takes into account spatiotemporal characteristics. From the perspective of hierarchical framework, the knowledge representation model for social group emergencies is divided into five levels from top to bottom: concept layer, object layer, state layer, attribute layer and relation layer. S2: Based on the knowledge representation model of social group emergencies with spatiotemporal characteristics described in S1, Protégé is selected for ontology construction, and Neo4j database is selected for ontology storage. S3: Using the knowledge representation model of social group emergencies constructed in S2 as the knowledge framework, a spatiotemporal knowledge graph of social group emergencies is constructed based on multi-source heterogeneous data through data acquisition and preprocessing, knowledge extraction, knowledge fusion and knowledge storage. S4: Evaluate the coverage and accuracy of the spatiotemporal knowledge graph of social group emergencies generated by S3.

2. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In step S1, a conceptual layer for social mass emergencies is established based on the reuse of the subject matter and self-defined concepts; an object layer and a state layer for social mass emergencies are established according to the constituent elements and spatiotemporal evolution laws of social mass emergencies; an attribute layer is established by summarizing the characteristic attributes of different states; and the relationships existing in the knowledge graph are modeled. Finally, a knowledge representation model for social mass emergencies that takes into account the spatiotemporal evolution process is established.

3. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In S2, corresponding nodes are established in Neo4j, and node types and node attributes are defined. Node types include concept nodes, object nodes, state nodes, and attribute nodes. Node attributes include the node name, which is the name, definition, and spatiotemporal characteristics in the classification system of social group emergencies. The node name is the unique identifier of the node. The relationship between nodes represents a subordinate relationship, which is represented by inclusion and being included.

4. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In S3, regarding data acquisition, a multi-strategy learning method is proposed to acquire data. Event elements and attribute knowledge in structured data tables are obtained through direct mapping. For semi-structured data from encyclopedia web pages, the web page structure is parsed, a web page element template matching model is designed, and web crawler technology is combined to acquire domain-related data. For unstructured text-based data such as news websites, a search engine is used, search keywords are set, and web crawlers are used to acquire data.

5. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In S3, data preprocessing includes preprocessing operations such as data cleaning and filtering for structured data; and preprocessing for unstructured text data includes sentence segmentation, word segmentation, and stop word filtering. Text sentence segmentation uses a regular expression-based sentence segmentation method based on punctuation marks, and segments sentences by recognizing punctuation marks such as periods, question marks, and exclamation marks. HanLP is used for word segmentation of text sentences.

6. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In S3, knowledge extraction includes entity recognition and relation extraction. Entity recognition uses the deep learning BERT-BiLSTM-CRF model to identify the part-of-speech tagging results and obtain the names of people, organizations / institutions, places, times, behaviors, causes, types, etc. in the knowledge text data. Relation extraction uses a rule and grammatical framework method to extract the semantic and spatiotemporal relationships between entities.

7. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In step S3, knowledge fusion is performed based on the cosine similarity calculation of word vectors. Clustering and a threshold-based merging method are used for knowledge fusion, where cosine similarity can be expressed as: The Similarity(A,B) value ranges from [0,1]. The closer it is to 1, the more similar the two texts are semantically. If the similarity is higher than the threshold, the two texts are judged to be the same and are merged.

8. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In S3, knowledge storage, through the aforementioned claims 4-7, achieves the acquisition of entities and relationships between entities, transforming source data of different structures into structured knowledge triple data in the form of (entity, relation, entity) or (entity, attribute, attribute value) and storing it using the graph database Neo4j.

9. The method for constructing a spatiotemporal knowledge graph for sudden social group events according to claim 1, characterized in that: In step S4, the generated spatiotemporal knowledge graph of social group emergencies is evaluated using two metrics: coverage and accuracy. Raw data from different sources is randomly sampled, knowledge triples are manually constructed, and then compared with the knowledge transformed from the current data in the knowledge graph. Coverage can be represented as: The dataset of triples extracted from the knowledge graph is K, and the dataset of triples constructed manually is T. The higher the coverage, the better the quality of the spatiotemporal knowledge graph of social group emergencies. Accuracy is the percentage of correct triples extracted from the knowledge graph out of the total number of triples, and can be expressed as: The number of random samples is T, and the number of errors is T. wrong The higher the accuracy, the better the quality of the spatiotemporal knowledge graph for social group emergencies.

Images