An emergency knowledge graph construction method based on bidirectional empowerment of page rank algorithm and louvain algorithm

Abstract

The present application relates to the technical field of wisdom emergency management, knowledge graph analysis and algorithm fusion, and particularly relates to an emergency knowledge graph construction method based on bidirectional empowerment of PageRank algorithm and Louvain algorithm. The method constructs an "entity-relation-entity" triple graph, calculates the global importance score of the entity by the PageRank algorithm, converts it into the Louvain algorithm weight to divide the emergency theme community, and outputs the "community-entity-priority" three-dimensional result through bidirectional empowerment optimization. The method innovatively realizes bidirectional empowerment of the two algorithms, solves the problem of insufficient adaptability of a single algorithm and disconnection between the graph and application, improves the accuracy of entity analysis, supports intelligent generation of emergency disposal reports, and greatly improves the efficiency of emergency disposal and the scientific nature of decision-making.

Description

Technical Field

[0001] This invention relates to the fields of intelligent emergency management, knowledge graph analysis and algorithm fusion technology, specifically to a knowledge graph construction method that achieves accurate analysis of emergency entities through bidirectional empowerment of PageRank and Louvain algorithms, and an intelligent emergency response report generation system applying this method. Background Technology

[0002] In emergency management scenarios, emergency knowledge graphs are a core foundation supporting resource allocation and decision-making, but their construction and application face multiple technical bottlenecks:

[0003] 1. Insufficient adaptability of single algorithms for constructing knowledge graphs: In existing technologies, PageRank can only achieve global importance ranking of emergency entities, but cannot reflect the relative priority in a specific scenario, which can easily lead to misjudgment in cross-scenario resource allocation; Louvain algorithm lacks core entity guidance when dividing entity communities, which can easily mix non-core auxiliary entities with core handling entities, and is out of sync with the hierarchical logic of "core-auxiliary" in emergency handling, resulting in knowledge graph analysis results that do not match actual needs.

[0004] 2. Knowledge graphs are disconnected from emergency application scenarios: Traditional emergency knowledge graphs only stay at the level of entity association construction. Their analysis results are difficult to directly transform into the core content of emergency response reports. They have not formed a closed loop of "entity analysis - report generation - decision support", which limits the practical value of knowledge graphs.

[0005] 3. Low efficiency and low quality in emergency report generation: Existing emergency response reports mostly rely on manual writing, which requires manually collecting and organizing information from heterogeneous data sources. This is not only time-consuming (often several hours), but also prone to problems such as data omissions, input errors, and format confusion due to staff fatigue and the complexity of information. Furthermore, fixed templates are difficult to adapt to the differentiated needs of different types and levels of emergency events, resulting in insufficient report credibility and linkage response efficiency.

[0006] The aforementioned problems prevent emergency knowledge graphs from fully playing their role in decision support, and make it difficult to guarantee the efficiency and accuracy of emergency response. There is an urgent need for an emergency knowledge graph construction method that balances the accuracy of entity analysis with application adaptability. Summary of the Invention

[0007] To address the limitations of existing technologies that rely on a single algorithm to construct emergency knowledge graphs, the disconnect between knowledge graphs and emergency report generation, and the low efficiency and poor adaptability of emergency report generation, this invention provides an emergency knowledge graph construction method based on the bidirectional empowerment of Page Ranking Algorithm and Luwan Algorithm. This method enables scenario-based and precise analysis of emergency entities and supports the intelligent generation of emergency response reports, forming a complete closed loop of "graph construction - analysis - application".

[0008] To solve the above-mentioned technical problems, the present invention adopts the following solution:

[0009] (I) Core Steps in the Construction of Emergency Knowledge Graphs

[0010] 1. Constructing a basic framework for an emergency knowledge graph

[0011] The entities in emergency scenarios (including core response resources, response objects / scenarios, auxiliary resources, etc.) and the relationships between entities (responsibility, linkage, inclusion, supply, etc.) are sorted out to form an emergency knowledge graph in the form of "entity-relationship-entity" triples, clarifying the type attributes and related logic of each entity.

[0012] 2. Calculate the global importance score of the entity.

[0013] Based on historical emergency case data, initial weight biases are set for each entity. The relationship strength of triples is labeled in conjunction with emergency business rules and converted into initial edge weights. The PageRank algorithm formula with core entity gain coefficient is used to iteratively calculate the global importance score (PR value) of all entities until the score change is <0.0001 and convergence is achieved. The top 70% of entities in terms of global importance score (PR value) are selected as the core entity candidate set.

[0014] 3. Divide into emergency-themed communities

[0015] The global importance score (PR value) of entities is normalized to the range of 0-1 and used as the node weight of the Louvain algorithm. The edge weights are adjusted based on the difference between the global importance scores (PR values) of two nodes and the initial relationship weights. Edges with a weight lower than 0.3 after adjustment are set to 0 to cut off non-core associations. The preset seed nodes of the Louvain algorithm are bound to the core entity candidate set. A modular formula with entity type penalty terms is used to divide the topic communities that conform to the emergency logic through the Louvain algorithm.

[0016] 4. Achieve bidirectional empowerment between PageRank and Louvain algorithms.

[0017] PageRank uses the global importance score of entities to guide the Louvain algorithm to prioritize the aggregation of core entities and avoid non-core entities from interfering with the community structure. The Louvain algorithm normalizes the global importance score (PR value) within the community by dividing the emergency topic community, scaling it to the relative importance score within the scene according to the original score ratio, thus clarifying the priority of each entity in the same scene.

[0018] 5. Output 3D analysis results

[0019] Integrate the community segmentation results with the scenario-based priority scores to output a three-dimensional analysis result of "community-entity-priority", including thematic entity clusters, core entity distribution, and entity allocation priority comparison table under each scenario.

[0020] (II) Application of the method in the intelligent generation system of emergency response reports

[0021] The aforementioned construction method is applied to an intelligent emergency response report generation system. This system adopts a layered architecture of "data layer - processing layer - application layer," horizontally comprising six core functional modules. Each module achieves data interoperability through standardized data interfaces.

[0022] Multi-source data acquisition module: Collects structured data, unstructured text, and image / video data through hardware interfaces, software integration, and manual data entry. The collected data is automatically tagged with "data type + timestamp + source identifier" and then pushed to the data cleaning and fusion module.

[0023] Data cleaning and fusion module: Uses BERT-BiLSTM-CRF model to extract key entities, completes data deduplication, contradiction detection and format standardization, and generates structured data to support knowledge graph construction and report generation.

[0024] Emergency Knowledge Graph Module: Based on the aforementioned construction method, the emergency knowledge graph is constructed and bidirectional empowerment analysis is performed, providing entity priority and community-related data support for template matching and report generation.

[0025] The template dynamic matching module extracts five core features of emergency events: type, level, location, scope of impact, and handling stage. It uses the SBERT algorithm to calculate the matching degree with each template in the template library. If there are content gaps in the matched template, it automatically calls the entity-related content of the corresponding topic community from the emergency knowledge graph to supplement the template.

[0026] The report auto-generation module uses the NLG engine's "data-text mapping" technology to convert structured data into natural language descriptions, automatically inserts images / videos to form a dual evidence chain of "text + audiovisual", and performs automated review from three dimensions: format compliance, information completeness, and logical consistency.

[0027] Linked response interaction module: Using the WebSocket protocol, it pushes the report to the terminals of each linked department within 10 seconds after the report is generated, supporting cross-departmental two-way feedback and iterative report updates.

[0028] The beneficial effects of this invention are as follows:

[0029] 1. Improved Entity Analysis: By leveraging both PageRank and Louvain algorithms, the limitations of a single algorithm are overcome. The Louvain algorithm improves the entity relevance and functional consistency within the community by more than 40%, while the PageRank score is adapted to specific scenarios, avoiding misjudgments in cross-scenario decision-making.

[0030] 2. Support for intelligent report generation: The three-dimensional results of the knowledge graph, namely "community-entity-priority", directly provide a basis for report template matching and content filling, ensuring that the report highlights the key points and adapts to the needs of the scenario, breaking down the barrier of "disconnect between knowledge graph analysis and report generation".

[0031] 3. Increased efficiency in report generation: The system automates data collection, processing, and report writing, reducing the report generation cycle from several hours to within 10 minutes, significantly improving the timeliness of emergency response.

[0032] 4. Improved report quality and credibility: Data cleaning and automated review mechanisms ensure that the accuracy of report compliance verification is ≥95%, reducing human error; the dual evidence chain of "text + audiovisual" strengthens the objectivity and persuasiveness of the report.

[0033] 5. Enhanced Emergency Response Collaboration: Real-time reporting and two-way feedback mechanisms break down cross-departmental data barriers, improving the timeliness and effectiveness of emergency response. Attached Figure Description

[0034] Figure 1 This is a flowchart of an emergency knowledge graph construction method based on the bidirectional empowerment of Page ranking algorithm and Luwan algorithm according to the present invention.

[0035] Figure 2 This is a block diagram of the intelligent emergency response report generation system of the present invention. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0037] like Figure 1 As shown, Figure 1 This is a flowchart of an emergency knowledge graph construction method based on the bidirectional empowerment of Page ranking algorithm and Luwan algorithm, according to the present invention. The method includes the following steps:

[0038] (1) Organize the entities and relationships between entities in emergency scenarios, and construct an emergency knowledge graph in the form of "entity-relationship-entity" triples;

[0039] (2) Calculate the global importance score of each entity in the emergency knowledge graph using the Page ranking algorithm;

[0040] (3) The global importance score is converted into node weights and edge weights of the Luwan algorithm, and the emergency topic community is divided by the Luwan algorithm;

[0041] (4) Based on the emergency topic community, the global importance score of entities is optimized in a scenario-based manner to achieve bidirectional empowerment of Page's ranking algorithm and Luwan's algorithm;

[0042] (5) Output the three-dimensional analysis results of “community-entity-priority”.

[0043] Preferably, in step (2), an initial weight bias is set for entities based on historical emergency case data. Combined with emergency business rules, the strength of the "entity-relationship-entity" triple relationship is marked and converted into initial edge weights. The Page ranking algorithm formula containing the core entity gain coefficient is used to iteratively calculate the global importance score of the entity until the score change is <0.0001. The top 70% of entities with the highest global importance scores are selected as the core entity candidate set.

[0044] Preferably, in step (3), the global importance score of the entity is normalized to the 0-1 interval as the node weight of the Luwan algorithm, and the edge weight is corrected based on the difference between the global importance scores of the two nodes and the initial relation weight. Edges with a weight lower than 0.3 after correction are set to 0. The preset seed nodes of the Luwan algorithm are bound to the core entity candidate set, and the emergency topic community is divided using a module formula containing entity type penalty terms.

[0045] Preferably, in step (4), the two-way empowerment specifically includes: Page ranking algorithm guides Luwan algorithm to prioritize the aggregation of core entities through the global importance score of entities, and Luwan algorithm performs "intra-community normalization" on the global importance score through the divided emergency theme community, scaling it to the relative importance score within the scene according to the original score ratio.

[0046] like Figure 2 As shown, Figure 2 This is a block diagram of the intelligent emergency response report generation system of the present invention. The method is applied to the intelligent emergency response report generation system. The system adopts a layered architecture of "data layer - processing layer - application layer". Horizontally, it includes a multi-source data acquisition module, a data cleaning and fusion module, an emergency knowledge graph module, a template dynamic matching module, an automatic report generation module, and a linkage response interaction module. Each module achieves data interoperability through standardized data interfaces.

[0047] Preferably, the multi-source data acquisition module acquires structured data, unstructured text, and image / video data through three methods: hardware interface, software integration, and manual data entry. The acquired data is automatically tagged with "data type + timestamp + source identifier" and then pushed to the data cleaning and fusion module. The data cleaning and fusion module uses the BERT-BiLSTM-CRF model to extract key entities and completes data deduplication, contradiction detection, and format standardization.

[0048] Preferably, the template dynamic matching module extracts five core features of the emergency event: type, level, location, scope of impact, and handling stage, and uses the SBERT algorithm to calculate the matching degree with each template in the template library; if there are content gaps in the matched template, it automatically calls the entity-related content of the corresponding topic community from the emergency knowledge graph to supplement the template.

[0049] Preferably, the automatic report generation module uses the "data-text mapping" technology of the NLG engine to convert structured data into natural language descriptions, automatically inserts images / videos to form a dual evidence chain of "text + audiovisual", and performs automated review from three dimensions: format compliance, information integrity, and logical consistency.

[0050] Preferably, the linkage response interaction module adopts the WebSocket protocol and pushes it to the terminals of each linkage department within 10 seconds after the report is generated, supporting cross-departmental two-way feedback and iterative report updates.

[0051] The following uses an "urban fire emergency scenario" as an example to explain in detail the implementation process of this invention:

[0052] Step 1: Construct the basic framework of the emergency knowledge graph

[0053] Identify entities and triples in emergency scenarios:

[0054] Entity types and specific entities: core response resources (main fire brigade 1, backup fire brigade 2, emergency center A), response targets / scenes (fire core area, fire edge area, trapped personnel group 1), auxiliary resources (temporary supply point, logistics group, traffic police guidance group, media group);

[0055] Example of a three-unit group: Main Fire Brigade 1 - Responsible for - Fire Core Area, Emergency Center A - Linked with - Main Fire Brigade 1, Fire Core Area - Includes - Trapped Personnel Group 1, Temporary Supply Point - Replenishment - Main Fire Brigade 1.

[0056] Step 2: Calculate the global importance score of the entity.

[0057] Initial weight configuration: Based on historical fire cases, the initial weight of the main fire brigade 1 and the core fire area is set to 1.2, and the initial weight of other entities is set to 1.0; the strength of the triplet relationship is marked (e.g., "Main Fire Brigade 1 - Responsible for - Core Fire Area" strength 9, "Temporary Material Point - Supply - Main Fire Brigade 1" strength 7).

[0058] PR value calculation: Page ranking algorithm formula with core entity gain coefficient is used for iterative calculation. The scores after convergence are as follows: Main fire brigade 1 (9.2 points), fire core area (8.8 points), emergency center A (7.5 points), backup fire brigade 2 (5.3 points), trapped personnel group 1 (4.8 points), fire edge area (4.5 points), temporary material point (2.1 points), logistics group (1.8 points), media group (1.2 points);

[0059] Core entity candidate set: Filter the top 70% of entities by PR value, namely the main fire brigade 1, the core fire zone, emergency center A, backup fire brigade 2, trapped personnel group 1, and the fire edge zone.

[0060] Step 3: Divide into emergency-themed communities

[0061] Weight adjustment: Normalize the PR value of each entity to the 0-1 range (e.g., Main Fire Brigade 1 0.92, Media Group 0.12), adjust the edge weights (increase the weight of the associated edge between Main Fire Brigade 1 and the core fire area by 20%), and set the associated edge of "Media Group - Logistics Group" with a weight lower than 0.3 to 0;

[0062] Seed node binding: Bind the main fire brigade 1, the fire core area, and the emergency center A in the core entity candidate set as the preset seed nodes of the Luwan algorithm;

[0063] Community partitioning: The Luwan algorithm was run using a modular formula with entity type penalties, ultimately resulting in 3 topic communities:

[0064] Core response team: Main fire brigade 1, core fire zone, emergency center A, and trapped personnel group 1;

[0065] Backup support teams: 2 backup fire brigades, fire edge zone support team, and traffic police evacuation team;

[0066] Support Groups: Temporary Supply Points, Logistics Group, Media Group.

[0067] Step 4: Two-way empowerment optimization

[0068] Page's ranking algorithm empowers Luwan's algorithm: core entities cluster to form functionally distinct communities, avoiding the problem of "temporary supply points, media groups and core entities being mixed together";

[0069] Luwan's algorithm empowers PageRank's algorithm: It normalizes PR values ​​within each community to obtain relative priority scores within the scenario.

[0070] Core response team: Main fire brigade 1 (10 points), emergency medical center A (8.2 points), trapped personnel group 1 (5.2 points);

[0071] Backup support teams: 2 backup fire brigades (10 points), fire edge zone (8.5 points), traffic police diversion team (6.1 points);

[0072] Support Groups: Temporary Supply Point (10 points), Logistics Group (8.6 points), Media Group (5.7 points).

[0073] Step 5: Output 3D analysis results

[0074] A “community-entity-priority” comparison table was created, clearly defining that the core response group has the main fire brigade 1 as the highest priority, the backup support group has the backup fire brigade 2 as the highest priority, and the auxiliary support group has the temporary material point as the highest priority.

[0075] Step 6: The system application generates an emergency report.

[0076] Data Acquisition and Fusion: The multi-source data acquisition module collects sensor fire data (level 3 in the core fire area), monitoring images, and manually reported information on trapped personnel (12 people). The data is then processed by the data cleaning and fusion module to generate structured data.

[0077] Dynamic template matching: The dynamic template matching module extracts event features (type: fire, level: major, stage: under control), matches the "major fire emergency response report template" using the SBERT algorithm, and calls the "main fire brigade 1 - fire core area - rescue" related module from the core response group to supplement the template;

[0078] Report generation and delivery: The automatic report generation module converts structured data and 3D analysis results into natural language reports, inserts on-site monitoring images, and after automated review, pushes them to fire, medical, and transportation departments within 10 seconds through the linkage response interaction module; after receiving feedback from the transportation department, it supplements road congestion information and updates the report.

[0079] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An emergency knowledge graph construction method based on the bidirectional empowerment of PageRank algorithm and Louvain algorithm, applied to the urban fire emergency scene, characterized in that, The method comprises the following steps: (1) combing the entities and the correlation between the entities in the urban fire emergency scene, wherein the entities include core disposal resources, disposal objects or scenes, and auxiliary resources; the core disposal resources include a main fire brigade, a backup fire brigade, and an emergency center A; the disposal objects or scenes include a fire core area, a fire edge area, and a trapped personnel group 1; the auxiliary resources include a temporary material point, a logistics group, a traffic police diversion group, and a media group; the correlation between the entities includes responsibility, linkage, inclusion, and supply, and an urban fire emergency knowledge graph in the form of a triple of “entity - relationship - entity” is constructed; (2) calculating the global importance score of each entity in the urban fire emergency knowledge graph by using the PageRank algorithm; (3) converting the global importance score into the node weight and the edge weight of the Louvain algorithm, and dividing the urban fire emergency theme community by using the Louvain algorithm; (4) optimizing the global importance score of the entity based on the urban fire emergency theme community, and realizing the bidirectional empowerment of the PageRank algorithm and the Louvain algorithm; the bidirectional empowerment specifically includes: the PageRank algorithm guiding the Louvain algorithm to preferentially aggregate core entities through the global importance score of the entity, and the Louvain algorithm performing “in-community normalization” on the global importance score through the divided urban fire emergency theme community, and scaling the global importance score into the relative importance score in the scene according to the original score ratio; (5) outputting a three-dimensional analysis result of “community - entity - priority”, and the result is used for technical decision of resource allocation of urban safety emergency disposal and intelligent generation of emergency disposal reports.

2. The emergency knowledge graph construction method based on the PageRank algorithm and the Louvain algorithm bidirectional empowerment according to claim 1, characterized in that, In step (2), the initial weight bias of the entity is set based on historical urban fire emergency case data, the relationship strength of the triple of “entity - relationship - entity” is marked in combination with urban fire emergency business rules, and the global importance score of the entity is iteratively calculated by using the PageRank algorithm formula containing a core entity gain coefficient until the score change is less than 0.0001, and the top 70% of the entities with the highest global importance scores are selected as a core entity candidate set. 3.The emergency knowledge graph construction method based on the two-way empowerment of PageRank algorithm and Louvain algorithm according to claim 1, wherein, In step (3), the global importance score of the entity is normalized to the interval of 0-1 as the node weight of the Louvain algorithm, the edge weight is corrected based on the difference between the global importance scores of two nodes and the initial relationship weight, and the edge with a corrected weight lower than 0.3 is set to 0; the Louvain algorithm preset seed node is bound to the core entity candidate set, and the urban fire emergency theme community is divided by using a module formula containing an entity type penalty term.

4. The emergency knowledge graph construction method based on the PageRank algorithm and the Louvain algorithm bidirectional empowerment of claim 1, characterized in that, The method is applied to an intelligent generation system of urban fire emergency disposal reports, and the system adopts a layered architecture of “data layer - processing layer - application layer”, and laterally comprises a multi-source data acquisition module, a data cleaning and fusion module, an urban fire emergency knowledge graph module, a template dynamic matching module, a report automatic generation module, and a linkage response interaction module, and the modules realize data interconnection through standardized data interfaces.

5. The emergency knowledge graph construction method based on the two-way empowerment of the PageRank algorithm and the Louvain algorithm according to claim 4, characterized in that, The multi-source data collection module collects structured data, unstructured text and image and video data through three ways of hardware interface, software docking and manual supplement, and pushes the collected data to the data cleaning and fusion module after automatic marking according to "data type + timestamp + source identification"; the data cleaning and fusion module extracts key entities by using a BERT-BiLSTM-CRF model, and completes data deduplication, contradiction detection and format standardization.

6. The emergency knowledge graph construction method based on the two-way empowerment of the PageRank algorithm and the Louvain algorithm according to claim 4, characterized in that, The template dynamic matching module extracts five core features of city fire emergency events, including type, level, location, influence range and disposal stage, calculates the matching degree with each template in the template library by using an SBERT algorithm, and automatically calls the entity associated content of the corresponding theme community from the city fire emergency knowledge graph to supplement the template if the matching template has content gaps.

7. The emergency knowledge graph construction method based on the two-way empowerment of the PageRank algorithm and the Louvain algorithm according to claim 4, characterized in that, The report automatic generation module converts structured data into natural language description by using the "data-text mapping" technology of the NLG engine, automatically inserts images or videos to form "text + audio-visual" dual evidence chain, and automatically audits from three dimensions of format compliance, information integrity and logical consistency.

8. The emergency knowledge graph construction method based on the two-way empowerment of the PageRank algorithm and the Louvain algorithm according to claim 4, characterized in that, The linkage response interaction module uses the WebSocket protocol to push the report to the terminal of each linkage department within 10 seconds after the report is generated, and supports cross-departmental two-way feedback and report iteration update.

Images