An emergency knowledge graph-based public health emergency scenario deduction method

Abstract

The application provides a sudden public health event scenario deduction method based on an emergency knowledge graph, comprising a knowledge graph module and a scenario deduction module. The knowledge graph module establishes a knowledge graph ontology according to sudden public health event data, and performs entity extraction, attribute extraction and graph fusion construction on the collected data. The scenario deduction module combines the emergency knowledge graph, and aims at the problems of complex correlation, variable scenario and rapid transmission of sudden public health cascade disasters, and provides a sudden public health event cascade disaster deduction method based on the knowledge graph. The sudden public health event information is aggregated, so as to mine the development possibility of the sudden public health event and deduce the development of the cascade sudden public health event.

Description

Technical Field

[0001] This invention relates to the application of knowledge graphs in the field of public health events, specifically the process of extracting and analyzing public health emergencies and performing event deductions using knowledge graphs. Background Technology

[0002] Major public health emergencies are characterized by complex interconnections, dynamic and ever-changing scenarios, and rapid disaster transmission. Cascade disasters differ from general disasters, involving the combined effects of multiple hazards, resulting in more potential and uncontrollable consequences and greater losses. In recent years, an increasing number of disasters have exhibited characteristics of cascade disasters, a typical example being the 2011 Great East Japan Earthquake, which triggered a tsunami and a nuclear power plant explosion. For cascade disasters in public health emergencies, the lack of readily available precursory information makes the "prediction-response" approach ineffective in predicting cascade disasters and prone to misleading decisions. An emergency simulation model based on "scenario-response" integrates historical cascade development processes of public health emergencies, predicts the aftermath of cascade disasters, and provides decision support for public health emergencies.

[0003] Knowledge graphs can store structured semantic information and enable computers to understand this information. They have become an indispensable resource for many artificial intelligence applications, but are affected by information gaps. Knowledge Graph Completion (KGC) emerged to address this issue. Its purpose is to predict the missing knowledge in the knowledge graph based on the existing knowledge. Furthermore, it extrapolates the probability of a subsequent event occurring under the given circumstances based on the existing sets of precursor and consequent events and their probabilities in the knowledge graph.

[0004] This patent is based on a deduction technology that integrates temporal knowledge graphs and contextual knowledge. It breaks through the bottleneck of fragmented, isolated and disordered information, adapts to the needs of high-precision and cross-domain emergency scenario changes, and constructs an orderly, efficient and timely hierarchical deduction system for major public health emergencies, realizing the cascading controllability of emergencies in complex factor coupling scenarios. Summary of the Invention

[0005] To address the aforementioned issues, this invention is based on the cascading disasters of emergencies, incorporating scenarios of major emergencies to construct an emergency knowledge graph oriented towards the evolution mechanism of emergencies. It meets the needs of both emergency scenarios under major emergencies and ordinary scenarios under normal conditions. This improves dispatching and decision-making capabilities in emergency scenarios, providing a guarantee for responding to emergencies.

[0006] The data sources for this invention rely on urban logistics data, urban material data, and urban public opinion data. It integrates multi-source, heterogeneous emergency logistics data through unstructured data fusion methods. When dealing with large volumes of data (major emergencies, logistics data, urban perception data), a unified paradigm can be used to express these large volumes of data through node and relationship formats: <node ID, attribute 1, attribute 2, ...>; <node ID1, node ID2, relationship type, relationship weight>.

[0007] The emergency knowledge graph-based method for scenario simulation of public health emergencies is characterized by including a knowledge graph module and a scenario simulation module.

[0008] The knowledge graph module includes at least four dimensions for constructing the knowledge graph ontology, namely: a person profile graph ontology, an emergency organization graph ontology, an emergency logistics graph ontology, and an emergency event logic graph ontology.

[0009] The scenario simulation module includes at least information on the transformation elements of the scenario situation, information on the combination of scenario situations, and a knowledge ontology network of the characteristic relationships of scenario elements. Based on the knowledge graph module, and addressing the challenges of complex, rapidly changing, and rapidly transmitting cascading public health emergencies, it overcomes the bottleneck of fragmented, isolated, and disordered information. Adapting to the needs of high-precision, cross-domain emergency scenario changes, it constructs an orderly, efficient, and timely hierarchical and tiered scenario simulation system for major public health emergencies, enabling the cascading controllability of emergencies under complex factor coupling scenarios.

[0010] The emergency knowledge graph-based method for simulating public health emergencies is characterized by including a knowledge graph module and a scenario simulation module.

[0011] The attributes of the portrait atlas include: age, workplace, education level, occupation, work address, and gender;

[0012] The emergency response agency map is organized by province / city / district, and its attributes include: area under its jurisdiction, agency responsibilities, address, agency size, and agency attributes.

[0013] The ontological attributes of the emergency knowledge graph include: personnel information, main tasks, transportation capacity information, road information, response units, and coordination agencies;

[0014] The emergency knowledge graph ontology includes emergency events, derivative disasters, response agencies, and response strategies.

[0015] The information on the transformation elements of the situation includes a basic description of the type of emergency, the degree of disaster, its scope, and its evolution.

[0016] The scenario situation combination information includes a set of data and information on the disaster-causing factors, related disaster-bearing bodies, response bodies, and disaster-prone environmental characteristics of emergency events, as well as their trends;

[0017] The knowledge ontology network of the relationship between the features of the scenario elements includes a situation ontology, a network ontology, and an evaluation ontology for scenario situation deduction, which correspond to scenario situation knowledge, situation evolution network knowledge, and evolution probability evaluation knowledge, respectively.

[0018] This invention, based on a constructed knowledge graph ontology, performs entity extraction, attribute extraction, and graph fusion construction on collected data. It comprises two parts: node processing and edge processing. The node processing of this invention preprocesses the raw emergency logistics data, removing noise information from individual nodes, emergency organization nodes, emergency logistics nodes, and event nodes. Taking emergency organization nodes as an example, nodes can be divided into four categories: material storage organizations, material production organizations, material transportation organizations, and material allocation organizations. These four types of nodes are extracted from the data, primarily stored as a database row record with multiple feature fields. Edge processing mainly relies on the dependency relationships between nodes, such as dependency relationships, to associate edges with nodes in the knowledge graph.

[0019] Based on the aforementioned emergency knowledge graph, this paper proposes a knowledge graph-based method for simulating cascading disasters in public health emergencies, addressing the challenges of complex interconnections, dynamic scenarios, and rapid transmission in such events.

[0020] The scenario deduction method according to the present invention includes:

[0021] 1. Before calculating the probability of specific scenario combinations, prepare relevant knowledge resources based on the scope of the research and the boundaries of the system, and determine the basic path of the deduction.

[0022] 2. When conducting scenario transformation simulations, the first step should be to determine the object of the simulation, namely the transformation path and transformation elements that need to be simulated.

[0023] 3. The transformation path determination is a roadmap for the scenario situation transformation and deduction, containing fixed scenario elements, including precursor scenario elements and follow-up scenario elements.

[0024] The transformation and evolution of scenarios vary across regions. Identifying representative transformation paths and elements in a given region can save on simulation costs and help decision-makers pinpoint key points in scenario simulation. On the other hand, potential or less obvious scenario transformation paths and elements can be generated based on historical cases and data analysis.

[0025] 4. Based on existing knowledge resources in the knowledge graph, extract the scenario combination patterns or scenario association mechanisms involved in the existing cascaded public health cases, and use them as the objects for subsequent scenario combination probability analysis.

[0026] 5. After identifying the transformation elements of the situation, narrow the research granularity and focus on the specific element characteristics involved in the transformation of situation elements and the complex relationships between these characteristics.

[0027] 6. A knowledge ontology network of contextual element feature relationships was constructed, and feature filtering was implemented based on the selected transformation elements.

[0028] 7. Set up scenario combinations for the situations of interest, select historical cases to disseminate information to each scenario combination, and infer the membership function of each scenario combination as its probability of occurrence.

[0029] 8. After determining the transformation elements and selecting their features, the variables in the transformation process of the scenario elements can be identified. These variables are divided into input variables (various types of knowledge that support the transformation deduction to determine the correlation between the features of the scenario elements during the transformation process), state variables (descriptive information of the scenario situation at each time segment during the evolution of the scenario situation), and output variables (the feature values ​​of the scenario elements determined after derivation, mainly the feature values ​​of the transformed elements). A Bayesian network model of scenario situation transformation is established and solved using the Markov chain Monte Carlo algorithm.

[0030] 9. Based on the set of precursor events and inference techniques, the probability of occurrence of each precursor event in the set of follow-up events can be obtained. The node with the highest similarity is found based on the similarity between the target public opinion event and each node in the graph. If the maximum similarity is less than a preset threshold, it indicates that there is no corresponding node for the target event in the graph, and inference cannot be performed. Otherwise, events that may occur in reality are inferred based on subsequent nodes in the graph. If a node has multiple successors, the probability of cascading events is calculated based on the edge weight coefficients.

[0031] Compared with existing technologies, this invention offers the following advantages: It provides a system for identifying urban risk areas during public health emergencies. First, in the infectious disease model module, a model more closely aligned with real-world scenarios is constructed. Second, regarding public health event information, the information reconstruction module focuses on the impact of unobservable information on future event evolution, and combined with a specifically designed infectious disease model, it can better reconstruct unobservable information. Furthermore, by using fine-grained regional information, the graph fusion module, based on the fusion of intra- and inter-graph information, can effectively acquire fine-grained features, thereby accurately identifying the risk level of fine-grained regions. This system can be fully utilized in public health event scenarios, and the graph information fusion module can be more broadly applied to event simulations within urban areas. Attached Figure Description

[0032] Figure 1 1. Structural diagram of the present invention;

[0033] Figure 2 The knowledge graph model of this invention. Detailed Implementation

[0034] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, and the terms "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0035] To further illustrate how this invention enables the deduction process of public health emergencies, the following is a detailed description of the knowledge graph-based deduction process for public health emergencies.

[0036] Step 1: Extract target features from the constructed scenario targets in the knowledge graph, and match the extracted feature values ​​with scenario elements in the knowledge graph scenario library.

[0037] Before calculating the probability of specific scenario combinations, relevant knowledge resources should be prepared based on the scope of the research and the boundaries of the system to determine the basic path of the deduction.

[0038] Identify the objects of the simulation, namely the transformation path and transformation elements to be simulated. Based on the transformation path, determine the roadmap for the scenario transformation simulation, including precursor scenario elements and follow-up scenario elements.

[0039] First, locate the set of scenario elements that are consistent with the scenario objective, and then further extract scenario elements through node matcher, constraint matcher and annotation matcher to form a new event scenario.

[0040] This patent adopts a hierarchical matching principle of "class-item-value", utilizing the ontological principle of... <i-n-c-a>This represents a set of scenario features for cascading public health events, corresponding to four types of features: Issues, Nodes, Constraints, and Annotations. Within the context of scenario situation deduction, the relationships between different scenario elements are considered. <i-n-c-a>Framework extended to <i-n-c-a-r>The framework adds relational components.

[0041] By comparing matching items in historical and current contexts based on public health emergency scenario elements located in the knowledge graph, and for successfully matched items in the public health emergency scenario, the values ​​of elements in the historical and current contexts are compared to determine if they are consistent. Finally, a similarity score is used. This represents the ratio of the number of identical values ​​under each scenario element to the total number of values ​​for the same scenario element in the new scenario, i.e.:

[0042]

[0043] In the formula, This represents the number of identical values ​​in each of the target matcher, node matcher, constraint matcher, and annotation matcher, and the number of values ​​in the target matcher in the new scenario. The number of values ​​in the node matcher The number of values ​​in the constraint matcher The number of values ​​in the annotation matcher , Each corresponds to a weight for the matching type. Finally, a similarity score is calculated using a similarity function to cluster context elements with the same characteristics.

[0044] Step 2: Construct a hypothesis correlation diagram. The hypothesis correlation diagram is constructed by considering the cascading relationship and temporal progression between precursor and consequent scenarios.

[0045] Let the set of scenario elements of a public health emergency be represented as:

[0046]

[0047] The set of precursor scenario elements is obtained based on the hypothesis correlation diagram. and the set of aftereffect scenario elements .Depend on express The function of a precursor scenario. express The action function is composed of a series of aftereffect scenarios. It is generated from the set of precursor scenarios and their action functions. If There is a cascading relationship between the set of precursor scenario elements and the set of consequent scenario elements. By combining them according to the cascading relationship, the cascading type and relationship of the scenarios in the set of precursor scenario elements and the set of consequent scenario elements can be determined.

[0048] Step 3: Calculate the probability of occurrence of a given public health emergency by using the set of precursor scenario elements, the set of consequent scenario elements, and the causal relationship between precursor and consequent scenarios.

[0049] The precursor scenario is obtained from step two. and aftermath The aftermath of the event. The probability of occurrence is expressed as ,but Aftermath when it occurs The probability of occurrence is Using the law of total probability, we obtain:

[0050]

[0051] As shown in the above formula, calculating the probability of a public health emergency requires mining the situation combination inference rules and calculating the probability of the occurrence of precursor scenarios. .

[0052] Record the premonitory situation as Then the domain of the premonitory situation is

[0053]

[0054] Let the situational characteristic observations of the situation be represented as: Based on scenario observations Diffusion of the observation to the universe of discourse Among all scenario feature points, the contained scenario information is diffused throughout the scenario set. The membership degree of a single observation in each scenario within the universe of discourse is shown in the following formula:

[0055]

[0056] In the formula, This represents the diffusion coefficient, and its value is determined by the maximum and minimum values ​​of the scenario features and the number of scenario feature points.

[0057] The total membership degree of the scenario observations is obtained based on the diffusion coefficient and the scenario information diffusion function:

[0058]

[0059] The membership function of a fuzzy subset is obtained from the above formula and the membership degree of the observed values:

[0060]

[0061] ,make

[0062]

[0063] The above formula represents the expression based on the observed values. After information is disseminated, if the observed values ​​for a public health emergency can only be taken from... One of them, when When it is a real observed sample point, Observed values The number of sample points, then let

[0064]

[0065] All samples can be used It means that when hour:

[0066]

[0067] This indicates that the probability is:

[0068] The probability of a public health emergency precursor scenario occurring can be obtained from the above formula.

[0069] Step 4: Under the condition that the precursor scenario of a public health emergency occurs, the probability of the consequence scenario occurring can be obtained by solving the probability of the simultaneous occurrence of the precursor scenario and the consequence scenario of a public health emergency, and the number of public health emergencies occurring under the consequence scenario. In the middle, the number of warning scenarios The ratio of the two can be derived from... express:

[0070]

[0071] The above formula yields the conversion probability from precursor scenario elements to consequent scenario elements.

[0072] Step 5: Determine whether the probability of a post-public health emergency scenario transformation exceeds the occurrence threshold defined in the knowledge graph. Let the occurrence threshold be [value missing]. The critical value of the probability of the aftermath scenario transition is ,when The scenario does not meet the conditions for transformation. If The evolution of public health event scenarios and their combinations.

[0073] Step Six: Repeat the above steps until all subsequent aftermath scenarios of the cascading disasters under the current precursor scenario of the public health emergency are obtained.

[0074] Step 7: Based on the inference results, update the triples of the knowledge graph.

[0075] First, multiple embedding spaces are created for the knowledge graph.

[0076] When a new triple is added to the knowledge graph, a new embedding space is created for the new triple, and the context information of the embedding space is updated. The knowledge graph is then updated based on the updated embedding space.

[0077] When new triples are added to the initial knowledge graph, the model is automatically updated online in the multi-element embedding space. The embedding space is constructed sequentially from left to right, with the leftmost embedding space representing the embedding space at a specific point in time. After the knowledge graph is updated for the first time, the updated content is used to construct a new multi-element embedding space according to the above method. Similarly, after the knowledge graph is updated for the second time, the updated content is used to construct a new multi-element embedding space according to the above method, and so on.

[0078] According to the dynamic updating method of knowledge graphs in embodiments of the present invention, when data changes, the model can adapt to dynamic data changes by generating a new embedding space. Furthermore, this study investigates an incremental updating model for knowledge graphs oriented towards relation prediction without requiring retraining, enabling the knowledge graph to adapt to dynamic data changes, particularly addressing many practical application scenarios involving unregistered entities, thus improving the dynamics of knowledge graph completion.

[0079] Compared with existing technologies, this invention offers the following advantages: It provides a method for scenario simulation of public health emergencies based on an emergency knowledge graph. Specifically, in the knowledge graph module, a knowledge graph ontology is built based on public health emergency data, and entity extraction, attribute extraction, and graph fusion are performed on the collected data. Secondly, in the scenario simulation module, combined with the emergency knowledge graph, and addressing the issues of complex cascading disaster associations, changing scenarios, and rapid transmission in public health emergencies, this invention provides a method for cascading disaster simulation of public health emergencies based on a knowledge graph. It aggregates information on public health emergencies to explore their potential development and to simulate the development of cascading public health emergencies.

[0080] The technical means disclosed in this invention are not limited to those disclosed in the above embodiments, but also include technical solutions composed of any combination of the above technical features.

Claims

1. A scenario simulation system for public health emergencies based on emergency knowledge graphs, characterized in that, It includes a knowledge graph module and a scenario inference module; the knowledge graph module includes at least four dimensions of knowledge graph ontology: a person profile graph ontology, an emergency organization graph ontology, an emergency logistics graph ontology, and an emergency event logic graph ontology; the scenario inference module includes at least a knowledge ontology network of scenario situation transformation element information, scenario situation combination information, and scenario element feature relationships; in the knowledge graph module, based on the constructed knowledge graph ontology, entity extraction, attribute extraction, and graph fusion construction are performed on the collected data, including node processing and edge processing; the nodes... The processing method preprocesses the raw data, removing noise from individual nodes, emergency organization nodes, emergency logistics nodes, and event nodes. The emergency organization nodes are categorized into four types: material storage organizations, material production organizations, material transportation organizations, and material allocation organizations. These four types of nodes are extracted from the data and stored as database records with multiple feature fields. The edge processing method associates edges between nodes in the knowledge graph based on their hierarchical relationships. The ontological attributes of the person profile graph include: age, work unit, education level, occupation, work address, and gender. The emergency response agency ontology is organized by province / city / district, with attributes including: responsible area, agency responsibilities, address, agency size, and agency attributes. The emergency logistics ontology attributes include: personnel information, main tasks, transportation capacity information, road information, response units, and connecting agencies. The emergency event causal ontology includes emergency events, derivative disasters, response agencies, and response strategies. The scenario situation transformation element information includes basic descriptions of the type, severity, scope, and evolution of the emergency event. The scenario situation combination information includes data and information sets and trends of the disaster-causing factors, related disaster-bearing bodies, response bodies, and disaster-prone environmental characteristics of the emergency event. The knowledge ontology network of scenario element characteristic relationships includes a situation ontology, network ontology, and assessment ontology for scenario situation deduction, corresponding to scenario situation knowledge, situation evolution network knowledge, and evolution probability assessment knowledge, respectively.

2. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 1, characterized in that: Before calculating the probability of specific scenario combinations, relevant knowledge resources should be prepared based on the scope of the research and the boundaries of the system, and the basic path of the deduction should be determined. The scenario situation transformation deduction should first determine the object of the deduction, that is, the transformation path and transformation elements to be deduced. Based on the transformation path, the roadmap of the scenario situation transformation deduction should be determined, which contains fixed scenario elements, including precursor scenario elements and follow-up scenario elements.

3. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 1, characterized in that: Based on existing knowledge resources in the knowledge graph, and on the basis of existing cascaded public health cases, the scenario combination patterns or scenario association mechanisms involved are extracted as objects for subsequent scenario combination probability analysis.

4. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 1, characterized in that: After identifying the transformation elements of the scenario situation, the research granularity is narrowed to focus on the specific element characteristics involved in the transformation of scenario elements and the complex relationships between these characteristics; a knowledge ontology network of scenario element characteristic relationships is constructed, and feature screening is achieved based on the selected transformation elements.

5. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 1, characterized in that: Scenario combinations are set up for the scenarios of interest, historical cases are selected to disseminate information to each scenario combination, and the membership function of each scenario combination is inferred as its probability of occurrence.

6. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 4, characterized in that: After determining the transformation elements and selecting their features, we can identify the variables in the scenario transformation process, classify them into input variables, state variables, and output variables, establish a Bayesian network model for scenario transformation, and solve it using the Markov chain Monte Carlo algorithm.

7. The emergency public health event scenario simulation system based on emergency knowledge graph as described in claim 1, characterized in that: The probability of occurrence of each subsequent event in the set of subsequent events is obtained based on the set of precursor events and inference techniques. Matching is performed with the node of highest similarity based on the similarity between the public health emergency and the scenario feature nodes in the knowledge graph. Scenario inference is then performed based on the subsequent nodes of the corresponding subsequent scenarios in the knowledge graph to obtain the event with the highest probability of occurrence under the precursor scenario. If a precursor event has multiple corresponding subsequent scenarios, the probability of occurrence of the subsequent event is inferred based on the edge weight coefficients of the relationships between different scenarios in the knowledge graph.

8. A method for scenario simulation of public health emergencies using the system described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Extract target features from the contextual goals of the knowledge graph, and match the extracted feature values ​​with the contextual elements in the knowledge graph contextual library to locate the set of contextual elements that are consistent with the contextual goals; Step 2: Construct a hypothesis correlation diagram by considering the cascading relationship and temporal progression of the precursor and aftereffect scenarios, determine the sets of elements for the precursor and aftereffect scenarios, and identify the cascading type and relationship between them. Step 3: Based on the set of precursor scenario elements, the set of aftermath scenario elements, and the causal relationship between them, calculate the probability of occurrence of the precursor scenario through information diffusion; Step 4: Under the condition that the precursor scenario occurs, calculate the conversion probability from precursor scenario elements to subsequent scenario elements; Step 5: Determine whether the probability of the aftermath scenario transformation exceeds the occurrence threshold defined in the knowledge graph, and determine whether the scenario situation has evolved accordingly. Step Six: Repeat steps two through five until all subsequent aftermath scenarios of the cascading disaster under the current precursor scenario are obtained; Step 7: Based on the inference results, create a new embedding space for the new triples in the knowledge graph, and update the context information of the embedding space to update the knowledge graph.

Images