Intelligent event evaluation method and device based on artificial intelligence technology

Abstract

The invention discloses an intelligent event assessment method and device based on an artificial intelligence technology, and the method comprises the steps: carrying out the clustering analysis of an alarm-related address through a DBSCAN density clustering algorithm, and assessing the geographic space risk of the alarm-related address; recursively extracting an alarm-related family knowledge graph by using a BFS breadth-first search algorithm of a graph database, and analyzing the risk of an alarm-related family relationship network; performing text vector representation on the received information text and the processed information text by using a transform neural network, identifying similar event conditions based on the distance between text vectors, further identifying repeated event conditions and series-parallel case clues, and analyzing event condition association risks; by establishing a multi-dimensional scoring system, weighted summation is carried out on factors including geographic space risks, family relation network risks and event condition association risks, and a comprehensive score of event condition risk degrees is obtained. According to the method, an open source big data platform is used as a basic framework, artificial intelligence components such as machine learning, knowledge graph and natural language processing are integrated, a unified data service port is packaged, and risk assessment of historical event conditions and specific event conditions is achieved.

Description

Technical Field

[0001] This invention relates to an intelligent event assessment method and apparatus based on artificial intelligence technology, belonging to big data analysis technology related to public safety management. Background Technology

[0002] In today's society, incident assessment is of paramount importance; its accuracy directly impacts the rational allocation of resources, efficiency, and the safeguarding of social security. For a long time, relevant departments have primarily relied on traditional methods for incident assessment, such as platform searches and SQL queries. Staff manually sift through a vast incident database, performing both full-scale screening and targeted screening for specific issues, aiming to identify potential conflicts and then assign and monitor related tasks. However, these traditional methods have gradually revealed numerous serious limitations in practical application, resulting in significant blind spots in governance and posing potential threats to social security.

[0003] First, the problem of relying solely on existing methods is particularly prominent. Traditional methods heavily depend on manual labor, using database query languages ​​such as SQL to manually filter and process massive amounts of event data. This approach is not only extremely inefficient, but also, with societal development, the volume of event data has exploded, making it difficult to handle such a heavy workload solely through manpower. Staff often get bogged down in tedious data filtering, consuming a great deal of time and energy, yet failing to achieve the desired results.

[0004] Secondly, the accuracy is severely lacking. Incident descriptions typically contain rich and complex information, including unstructured data such as diverse and uncertain geographical locations and complex relationships among involved personnel. Traditional methods utilize this data very little. The weak connection between this data and the external environment leads to the overlooking of many crucial clues. For example, cross-regional incident correlations, similar incidents with long time intervals, location difficulties due to vague descriptions, and recurring incidents involving complex relationships between different people in the same household are difficult to effectively detect and identify. This masks potential risks, preventing timely handling and response.

[0005] Furthermore, early warning capabilities are extremely weak. Traditional methods primarily focus on events that have already occurred, lacking effective mechanisms for predicting potential risks. They cannot proactively anticipate potential risk trends and prevent problems before they escalate or cause serious consequences. Often, police only take notice when conflicts have intensified or even resulted in disturbances or other serious incidents. By then, the optimal time for intervention may have been missed, leading to irreparable damage to public safety.

[0006] With the continuous development of science and technology, several event assessment algorithms have emerged. However, most of these algorithms rely primarily on simple keyword retrieval or statistics, such as counting the number of times two identical individuals are involved in the same event, searching for keywords related to the location of an event to determine the number of events at a particular location, and counting the occurrence frequency of each type of event. These methods fall short when faced with complex situations such as different individuals residing in the same household, non-standard location names, and insufficient similarity in the content of the events. They are unable to effectively identify and address these complex risk factors, resulting in the failure to identify and eliminate risk points in a timely manner, thus creating potential threats to public safety.

[0007] Fortunately, the rapid development of artificial intelligence technology, particularly the significant advancements in natural language processing, machine learning, and big data analytics, has brought new hope and solutions for intelligent event assessment. These advanced technologies can greatly improve the automation of event assessment, rapidly analyzing and processing massive amounts of data through intelligent algorithms, significantly reducing the burden on human resources. Simultaneously, they can deeply mine and analyze potential information within event data, improving the accuracy of assessments and leaving no subtle risk clue unchecked. More importantly, with powerful predictive capabilities, they can identify potential risks in advance, enabling proactive prevention and thus more effectively preventing and addressing social security issues, providing a more solid guarantee for social stability and public safety. Summary of the Invention

[0008] Purpose of the invention: In order to overcome the shortcomings of the existing technology, the present invention provides an intelligent event assessment method and device based on artificial intelligence technology. By utilizing cutting-edge technologies such as graph databases and natural language processing, it comprehensively considers multiple dimensions of factors such as geospatial risks, family relationship network risks, and event situation correlation risks, and fully explores decentralized and hidden social conflicts, security risks, and clues to illegal and criminal activities, providing strong support for relevant departments to carry out accurate analysis and proactive prevention.

[0009] Technical solution: To achieve the above objectives, the technical solution adopted by this invention is as follows: An intelligent event assessment method based on artificial intelligence technology is proposed. It utilizes the DBSCAN density clustering algorithm to cluster addresses involved in police incidents and assess their geospatial risk. It recursively extracts a knowledge graph of families involved in police incidents using the Breadth-First Search (BFS) algorithm in a graph database to analyze the risk of family relationship networks. It employs a transformer neural network to represent received and processed information texts as text vectors, identifying similar events based on the distance between text vectors, thereby identifying repeated events and clues to linked cases, and analyzing the associated risks of events. Finally, it establishes a multi-dimensional scoring system, weighted and summing factors including geospatial risk, family relationship network risk, and event association risk to obtain a comprehensive score for the degree of risk of the event.

[0010] Specifically, the method includes the following steps: Step 1: Data Acquisition and Preprocessing: Collect multi-source heterogeneous historical event data, including received information text, dispatch information text, and processing information text. Extract key fields such as police-related addresses, police-related personnel, and received / processed information. The received / processed information includes received / processed time, event type, and event description. Step 2, High-Frequency Risk Location Mining: The event address API is used to convert police-related addresses into event addresses and their corresponding latitude and longitude coordinates. These event addresses and coordinates are added to the historical database of police-related addresses. Different descriptions of the same latitude and longitude coordinates are identified, and the frequency of police involvement for each event address is statistically analyzed. The DBSCAN density clustering algorithm is used to cluster the latitude and longitude coordinates, and the Haversine formula is used to calculate the cluster radius. The centroid of each cluster is designated as a high-frequency risk location, and these locations are marked in the historical database of police-related addresses. Step 3: High-risk family identification: Using police-involved personnel as the root node, trace the family member nodes, associate all historical event data related to the family member nodes, and construct a multi-layered relationship network of police-involved personnel-family members-event situations; based on the Neo4j graph database, use the breadth-first search algorithm (BFS) to recursively extract the knowledge graph of police-involved families from the multi-layered relationship network of police-involved personnel-family members-event situations, and add the knowledge graph of police-involved families to the historical database of police-involved families; Step 4: Similar Event Mining: Using the transformer neural network architecture and the m3e Chinese semantic similarity model, the received information text and the processed information text are represented as text vectors. The text vectors are added to the event vector history library, the similarity of different text vectors is measured, and similar event situations (event situations with similar content, similar methods, or other related information) are identified, thereby identifying repeated event situations and clues to linked cases. Step 5, Multi-dimensional Scoring System: For information on an event on a given day, the addresses involved in the incident are converted into event addresses and their corresponding latitude and longitude coordinates and added to the historical database of addresses involved in the incident. Based on the frequency of the incident involving the address and its distance from high-frequency risk addresses, the geospatial risk of the address involved in the incident is assessed. Simultaneously, the received and processed information texts are represented as text vectors and added to the historical database of event vectors. Similar events are identified, and it is determined whether they are repeated events or have clues of parallel cases, analyzing the associated risks of the events. At the same time, the database of families involved in the incident searches for the individuals involved in the incident. If they exist, the knowledge graph of the families involved in the incident is extracted; if not, the knowledge graph of the families involved in the incident is constructed and added to the historical database of families involved in the incident, analyzing the risk of the family relationship network. The factors, including geospatial risk, family relationship network risk, and associated risks of the event, are weighted and summed to form a comprehensive score of the risk level of the event on that day.

[0011] Specifically, in Step 4, Euclidean distance is used to measure the similarity of different text vectors.

[0012] Specifically, in Step 5, the Elasticsearch distributed search engine is used to search for data in the historical database of police-related addresses, the historical database of police-related families, and the historical database of event vectors, achieving millisecond-level associative retrieval. At the same time, a UI interface can be designed to realize cross-analysis and chart display of event data from multiple dimensions such as time, space, and personnel, making it convenient for police officers to accurately assess the security situation in their jurisdiction from multiple levels, including macro, meso, and micro perspectives.

[0013] Specifically, in Step 1, for the collected historical event data, which includes received information text, dispatch information text, and processing information text, Python is first used to clean and transform the data, and then key fields, including police-related addresses, police-related personnel, and processing information, are extracted.

[0014] Specifically, in Step 3, when constructing the multi-layered relationship network of police officers, family members, and incident details, the population database is first imported, and a household association edge is added to each police officer node using the same address and household rules, linking it to the police officer's family members; then, based on the roles of the police officers and their family members in their respective incident details, association edges are established between the personnel and the incident details, and the personnel, association edges, and incident details are stored in the Neo4j graph database to form a multi-layered relationship network of police officers, family members, and incident details.

[0015] Specifically, in Step 3, based on the Neo4j graph database, the Breadth-First Search (BFS) algorithm is used to start with the person involved in the case and reach their family members in the same household in one hop, and reach the event situation involving the family member in two hops. This process is recursively repeated until six hops. The person involved in the case and their family members are treated as nodes, and the role relationship between the nodes and the event situation is treated as an edge. All nodes and edges are extracted to form a knowledge graph of families involved in the case with the person involved in the case as the core. The risk value of the knowledge graph of families involved in the case with the police is evaluated based on the number and frequency of the event situation.

[0016] Specifically, in Step 4, an event situation corpus is constructed and incrementally trained to obtain an event situation semantic understanding model suitable for the scenario. The event situation semantic understanding model is then used to encode the received information text and the processed information text to obtain text vectors.

[0017] Specifically, in Step 4, the dimensions of the text vector include the time of the crime, the location of the crime, the perpetrator, and the method of the crime. Events with similarity greater than a set threshold are grouped into one category to form similar event cases.

[0018] An apparatus for implementing an intelligent event assessment method based on artificial intelligence technology includes a data acquisition and preprocessing module, a high-frequency risk location mining module, a high-risk family mining module, a similar event situation mining module, and a multi-dimensional scoring module; The data acquisition and preprocessing module collects and summarizes structured and unstructured multi-source heterogeneous event data, cleans and standardizes the event data, and extracts keywords including police-related addresses, police-related personnel, and handling information. The high-frequency risk location mining module calls the event address API to convert police-related addresses into event addresses and corresponding latitude and longitude coordinates, builds a historical database of police-related addresses, and uses statistical and clustering methods to identify high-frequency risk locations. The high-risk family mining module uses graph databases to mine the social relationship networks of people involved in police affairs, extracts knowledge graphs of families involved in police affairs, and constructs a historical database of families involved in police affairs. The similar event mining module uses a transformer neural network architecture and the m3e Chinese semantic similarity model to represent event data as text vectors and build an event vector history library. The multi-dimensional scoring module, for a given day's event information, calls the high-frequency risk location mining module to assess the geospatial risk of the police-related address, the high-risk family mining module to analyze the risk of the police-related family relationship network, and the similar event situation mining module to analyze the event situation association risk. It then performs a weighted summation of factors including geospatial risk, family relationship network risk, and event situation association risk to obtain a comprehensive score for the degree of event situation risk.

[0019] Beneficial Effects: The event intelligent assessment method and device based on artificial intelligence technology provided by this invention have the following advantages compared with existing technologies: 1. It introduces advanced artificial intelligence technologies such as machine learning and knowledge graphs, breaking through the timeliness and accuracy bottlenecks of traditional manual investigation methods, and realizing automated, intelligent, and relational mining of massive event data; 2. The high-frequency risk location mining module uses clustering analysis technology to discover high-risk locations hidden behind vague addresses; the high-risk family mining module uses graph database analysis technology to reveal the complex social relationship network of police-related personnel and discover families frequently involved in police affairs; the similar event situation mining module uses semantic analysis technology to understand the inherent logical relationship between similar event situations; 3. The multi-dimensional scoring module can connect key elements such as time, space, people, and cases to form a three-dimensional and systematic ability to analyze and judge the jurisdiction's security situation, providing decision support for relevant departments to eliminate hidden dangers in a timely manner and accurately combat crime. Attached Figure Description

[0020] Figure 1 This describes the construction process of the police-related address history database, police-related family history database, and event situation vector history database in this invention; Figure 2 A schematic diagram of the process for identifying high-risk locations; Figure 3 This refers to the process of scoring information about a specific event from multiple dimensions. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0022] An intelligent event assessment device based on artificial intelligence technology includes a data acquisition and preprocessing module, a high-frequency risk location mining module, a high-risk family mining module, a similar event situation mining module, and a multi-dimensional scoring module. The data acquisition and preprocessing module collects and summarizes structured and unstructured multi-source heterogeneous event situation data, cleans and standardizes the event situation data, and extracts keywords including police-related addresses, police-related personnel, and handling information. The high-frequency risk location mining module calls an event address API to convert police-related addresses into event addresses and corresponding latitude and longitude coordinates, constructs a historical database of police-related addresses, and uses statistical and clustering methods to identify high-frequency risk locations. The high-risk family mining module uses a graph database to mine the social data of police-related personnel. The system utilizes a relationship network to extract a knowledge graph of families involved in police affairs and constructs a historical database of such families. The similar event mining module employs a transformer neural network architecture and the m3e Chinese semantic similarity model to represent event data as text vectors, constructing a historical database of event vectors. The multi-dimensional scoring module, for a given day's event information, calls the high-frequency risk location mining module to assess the geospatial risk of the address involved in police affairs, the high-risk family mining module to analyze the relationship network risk of families involved in police affairs, and the similar event mining module to analyze the association risk of the event. It then performs a weighted summation of factors including geospatial risk, family relationship network risk, and event association risk to arrive at a comprehensive score for the degree of event risk.

[0023] A method for intelligent event assessment using the aforementioned device is based on an open-source big data platform, integrating artificial intelligence components such as machine learning, knowledge graphs, and natural language processing, and encapsulating a unified data service port. This method can automatically mine multi-dimensional security risks from massive event data, significantly improving analysis efficiency and accuracy to assess risks in both historical and specific events, thereby better assisting relevant departments in their practical work. The method utilizes the DBSCAN density clustering algorithm to cluster police-related addresses and assess their geospatial risk; it recursively extracts a knowledge graph of police-related families using the Breadth-First Search (BFS) algorithm in a graph database to analyze the risk of family relationship networks; it uses a transformer neural network to represent received and processed information text as text vectors, identifying similar events based on the distance between text vectors, thereby identifying repeated events and clues to linked cases, and analyzing the associated risks of events; and it establishes a multi-dimensional scoring system, weighting and summing factors including geospatial risk, family relationship network risk, and event association risk to obtain a comprehensive score for the degree of event risk.

[0024] The present invention will be further described below with reference to specific implementation steps.

[0025] Step 1: Data Acquisition and Preprocessing like Figure 1 As shown, the historical event database is queried using SQL to collect multi-source heterogeneous historical event data, including received information text, dispatch information text, and processing information text. First, Python is used to clean and transform the data, and then key fields such as police-related addresses, police-related personnel, and handling information are extracted. The handling information includes handling time, event type, and event description.

[0026] Step 2: High-frequency risk location identification The system calls the event address API to convert police-related addresses into event addresses and their corresponding latitude and longitude coordinates. These event addresses and coordinates are then added to the police-related address history database (Elasticsearch database). Different descriptions of the same latitude and longitude coordinates are identified, and the frequency of police involvement for each event address is statistically analyzed. The DBSCAN density clustering algorithm is used to cluster the latitude and longitude coordinates, and the Haversine formula is used to calculate the cluster radius. The centroid of each cluster is designated as a high-frequency risk location, and these high-frequency risk locations are marked in the police-related address history database.

[0027] like Figure 2 As shown, based on different event types, such as theft, fraud, and disorderly conduct, the locations involved in the incidents are first grouped according to their latitude and longitude coordinates. For each group, spatial clustering is performed using the DBSCAN density clustering algorithm (ε=200m, samples=5). The centroid of each cluster is taken as the high-frequency risk location, and the density of the cluster and the number of samples within the cluster represent the risk level of that high-frequency risk location. The spherical distance between any two points within a cluster is calculated using the Haversine formula to find the coverage area of ​​the high-frequency risk location. Outliers not in any cluster are removed, and finally, a heat map and statistical report of the distribution of high-frequency risk locations are generated.

[0028] Step 3: Identifying High-Risk Families Using police-related personnel as the root node, this method traces back to family member nodes and associates all historical event data related to these family member nodes to construct a multi-layered relationship network of police-related personnel, family members, and event situations. To construct this network, a population database is first imported. Using the same-address, same-household rule, a household association edge is added to each police-related personnel node, linking it to their family members. A node is created for each member, with attributes including name, ID number, registered address, and permanent address. Then, based on the roles of the police-related personnel and their family members in their respective event situations, association edges are established between personnel and event situations, forming the multi-layered relationship network of police-related personnel, family members, and event situations.

[0029] Based on the Neo4j graph database, a knowledge graph of families involved in police cases is recursively extracted from a multi-layered relationship network of police-family members-event details using the breadth-first search (BFS) algorithm. Starting with the police-involved individual, the algorithm proceeds with a hop to their family members in the same household, a hop to the event details involving the family member, and so on, up to a hop of 6. The police-involved individual and their family members are treated as nodes, and the role relationships between nodes and event details are treated as edges. All nodes and edges are extracted to form a knowledge graph of families involved in police cases with the police-involved individual at its core.

[0030] Adding a knowledge graph of families involved in police affairs to the history database of families involved in police affairs (Neo4j graph database) allows for the evaluation of the risk value of the knowledge graph based on the number and frequency of events.

[0031] Step 4: Similar Event Scenario Mining Employing a transformer neural network architecture and the M3E Chinese semantic similarity model, the received and processed information texts are represented as 768-dimensional text vectors. These text vectors are added to an event situation vector history database (Elasticsearch database) to construct an event situation corpus and perform incremental training, resulting in an event situation semantic understanding model (M3E) suitable for various scenarios. The event situation semantic understanding model is then used to encode the received and processed information texts, yielding text vectors. The dimensions of these text vectors include the time of the crime, the location of the crime, the perpetrators, and the method of the crime. Similarity is evaluated by measuring the Euclidean distance between different text vectors. Event situations with similarity greater than a set threshold are clustered together to identify similar events (event situations with similar content, similar methods, or other related information), thereby identifying duplicate events and clues for linking cases.

[0032] The event situation corpus needs to be updated and grouped in a timely manner. Each event situation group contains situations that describe the same event, but with different wording. For example: Event Situation ①: There was noise pollution from square dancing at xx square; Event Situation ②: The loudspeakers at xx square were very loud. These two events would be grouped into the "Passing Event Situation" group.

[0033] Text vectors that express the situation of an event should be able to maintain compatibility with mainstream pre-trained models while also fully expressing the semantic information of the event.

[0034] Step 5: Multi-dimensional scoring system like Figure 3As shown, for information on an event on a given day, the addresses involved in the incident are converted into event addresses and their corresponding latitude and longitude coordinates and added to the historical database of addresses involved in the incident. Historical events within a specified range are searched, and the geospatial risk of the addresses involved in the incident is assessed based on their frequency of involvement and their distance from high-frequency risk addresses. Simultaneously, the received and processed information texts are represented as text vectors and added to the historical database of event vectors. Similar events are identified, and it is determined whether they are duplicate events or related to other cases, analyzing the associated risks of the events. Furthermore, the database of families involved in the incident searches for individuals involved in the incident. If they exist, a knowledge graph of their families is extracted; otherwise, a knowledge graph of their families is constructed and added to the historical database of families involved in the incident, analyzing the risk of family relationship networks. Finally, a weighted sum of factors including geospatial risk, family relationship network risk, and event association risk is calculated to form a comprehensive score for the risk level of the event on that day.

[0035] In addition, importing structured event information elements (time, address, people, category), event information text, cluster analysis results, etc. into Elasticsearch allows for the aggregation and analysis of spatial hotspots by processing location, the aggregation and analysis of personnel panoramas by the identity, age, and place of residence of police personnel involved, and the aggregation and analysis of the distribution of event information elements by category and keywords. The analysis results can be visualized through scatter plots and heat maps, which can help in the compilation of comprehensive judgment reports, the assessment of the overall security situation in the jurisdiction, the prediction of future development trends, and the provision of reference for decision-making.

[0036] In this method, DBSCAN clustering parameters, BFS search depth, and similarity thresholds can be dynamically adjusted according to application requirements and data scale. This embodiment fully utilizes multi-source data and employs artificial intelligence technologies such as machine learning, knowledge graphs, and natural language processing to mine hidden security risks from massive event data from multiple perspectives, including time, space, and personnel, forming actionable and interpretable auxiliary decision-making suggestions to provide precise guidance for frontline work. Through continuous accumulation and optimization, this method can form a standardized and systematic big data intelligent analysis method, providing a universal solution for practical applications.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. An event intelligent evaluation method based on artificial intelligence technology, characterized in that: The DBSCAN density clustering algorithm is used for clustering analysis of the police-related addresses to evaluate the geographic spatial risk of the police-related addresses; the BFS breadth-first search algorithm of the graph database is used for recursive extraction of the police-related family knowledge graph to analyze the family relationship network risk of the police-related families; the transformer neural network is used for text vector representation of the received information text and the processed information text, the distance between the text vectors is measured to identify similar event situations, and then repeated event situations and serial and parallel case clues are identified to analyze the event situation association risk; a multi-dimensional scoring system is established to weight and sum the factors including the geographic spatial risk, the family relationship network risk and the event situation association risk to obtain a comprehensive score of the event situation risk degree. 2.The event intelligent assessment method based on artificial intelligence technology according to claim 1, characterized in that: The method comprises the following steps: Step 1, data acquisition and preprocessing: collect multi-source heterogeneous historical event situation data including received information text, police-out information text and processed information text, extract key fields including police-related addresses, police-related personnel and received and processed information, and the received and processed information includes received and processed time, event situation type and event situation description; Step 2, high-frequency risk place mining: call the event address API to convert the police-related addresses into event addresses and corresponding latitude and longitude coordinates, add the event addresses and corresponding latitude and longitude coordinates to the police-related address history library, and count the police-related frequencies of each event address; use the DBSCAN density clustering algorithm to cluster the latitude and longitude coordinates, calculate the clustering radius by combining the Haversine formula, take the centroid of each cluster as a high-frequency risk place, and mark the high-frequency risk places in the police-related address history library; Step 3, high-risk family mining: take the police-related personnel as a root node, trace the family member nodes, associate all historical event situation data related to the family member nodes, and construct a multi-layer relationship network of police-related personnel-family member-event situation; based on the Neo4j graph database, use the BFS breadth-first search algorithm to recursively extract the police-related family knowledge graph from the multi-layer relationship network of police-related personnel-family member-event situation, and add the police-related family knowledge graph to the police-related family history library; Step 4, similar event situation mining: adopt the transformer neural network architecture and the m3e Chinese semantic similarity model to represent the received information text and the processed information text into text vectors, add the text vectors to the event situation vector history library, measure the similarity of different text vectors, identify similar event situations, and then identify repeated event situations and serial and parallel case clues. Step 5, multi-dimensional scoring system: for a certain daily event information, convert the police-related address into an event address and corresponding latitude and longitude coordinates and add it to the police-related address history library. Based on the police-related frequency of the police-related address and the distance from each high-frequency risk address, the geographical space risk of the police-related address is evaluated. At the same time, the received information text and the processed information text are represented as text vectors and added to the event situation vector history library. Similar event situations are identified, and it is determined whether it is a repeated event situation or there is a serial and parallel case clue to analyze the event situation associated risk. At the same time, the police-related personnel are searched in the police-related family history library. If there are, the police-related family knowledge graph of the police-related personnel is extracted. If not, the police-related family knowledge graph of the police-related personnel is constructed and added to the police-related family history library to analyze the risk of the family relationship network. The factors including geographical space risk, family relationship network risk and event situation associated risk are weighted and summed to form a comprehensive score of the risk degree of the event situation on that day. 3.The event intelligent evaluation method based on artificial intelligence technology according to claim 2, characterized in that: In Step 4, the Euclidean distance is used to measure the similarity of different text vectors. 4.The method of claim 2, wherein the method further comprises: determining a first event type of the first event; determining a second event type of the second event; and determining the first event type and the second event type based on the first event type and the second event type. In Step 5, the Elasticsearch distributed search engine is used to search data in the police-related address history library, the police-related family history library, and the event situation vector history library. 5.The event intelligent assessment method based on artificial intelligence technology according to claim 2, characterized in that: In Step 1, for the collected multi-source heterogeneous historical event situation data including received information text, police information text and processed information text, first use python for data cleaning and conversion, and then extract key fields including police-related address, police-related personnel and processing information. 6.The event intelligent assessment method based on artificial intelligence technology according to claim 2, characterized in that: In Step 3, when constructing the multi-layer relationship network of police-related personnel-family members-event situation, first import the population database, and use the same address and same household rule to add the same household association edge for each police-related personnel node, and associate to the family members of the police-related personnel; then according to the roles of the police-related personnel and family members in their respective event situations, establish the association edge between the personnel and the event situation, and store the personnel, association edge and event situation into the Neo4j graph database to form the multi-layer relationship network of police-related personnel-family members-event situation. 7.The event intelligent assessment method based on artificial intelligence technology according to claim 6, characterized in that: In Step 3, based on the Neo4j graph database, the BFS breadth-first search algorithm is used to start from the police-related personnel, reach the family members of the same household in 1 hop, reach the event situations involving the family members in 2 hops, and recursively perform the process until 6 hops; the police-related personnel and family members are taken as nodes, and the role relationship between the nodes and the event situation is taken as an association edge. All nodes and association edges are extracted to form a police-related family knowledge graph centered on the police-related personnel, and the risk value of the police-related family knowledge graph is evaluated according to the number and frequency of event situations. 8.The event intelligent assessment method based on artificial intelligence technology according to claim 2, characterized in that: In Step 4, an event situation corpus is constructed and incrementally trained to obtain an event situation semantic understanding model suitable for the scene. The received information text and the processed information text are encoded using the event situation semantic understanding model to obtain text vectors. 9.The event intelligent assessment method based on artificial intelligence technology according to claim 2, characterized in that: In Step 4, the dimensions of the text vector include the crime time, crime location, crime personnel, and crime method, and events with a similarity greater than a set threshold are clustered into a category to form similar event cases. 10.An event intelligent evaluation device based on artificial intelligence technology, characterized in that: The system comprises a data collection and preprocessing module, a high-frequency risk location mining module, a high-risk family mining module, a similar event case mining module, and a multi-dimensional scoring module. The data collection and preprocessing module collects and aggregates structured and unstructured multi-source heterogeneous event case data, cleans and standardizes the event case data, and extracts keywords such as police-related addresses, police-related personnel, and handling information. The high-frequency risk location mining module converts police-related addresses into event addresses and corresponding latitude and longitude coordinates by calling an event address API, constructs a police-related address history library, and identifies high-frequency risk locations using statistical and clustering methods. The high-risk family mining module uses a graph database to mine the social relationship network of police-related personnel, extracts a police-related family knowledge graph, and constructs a police-related family history library. The similar event case mining module uses a transformer neural network architecture and an m3e Chinese semantic similarity model to represent event case data as a text vector and construct an event case vector history library. The multi-dimensional scoring module evaluates the geographical space risk of police-related addresses by calling the high-frequency risk location mining module, analyzes the relationship network risk of police-related families by calling the high-risk family mining module, and analyzes the event case correlation risk by calling the similar event case mining module. The geographical space risk, family relationship network risk, and event case correlation risk are weighted and summed to obtain a comprehensive score of the event case risk level.

Images