Alarm emergency degree intelligent scoring method and system
By extracting and engineering multi-dimensional attributes and features, and combining them with machine learning models, the accuracy problem of emergency urgency scoring was solved, achieving multi-angle and three-dimensional emergency urgency scoring, thus improving the accuracy of scoring and decision support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-23
- Publication Date
- 2026-03-27
AI Technical Summary
Existing deep learning-based methods for scoring the urgency of police incidents rely on large amounts of labeled data, which are costly and have poor model interpretability. Furthermore, scoring based on a single attribute leads to inaccurate results.
Multi-dimensional attribute feature extraction is adopted, including the nature and category of the incident, the location of the incident, the items involved, the personnel involved, keyword scores, contextual tone scores, and external correlation urgency factors. Multi-dimensional feature vectors are generated through feature engineering, and machine learning models are used for scoring.
It enables a multi-dimensional and comprehensive description of the urgency of police incidents, improves the accuracy of scoring, and provides intelligent and precise decision support.
Smart Images
Figure CN121744217A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of police case processing, and in particular to a police case emergency degree intelligent scoring method and system. BACKGROUND
[0002] Police case processing is one of the core links of public safety management. With the development of social economy, the number of police cases is increasing year by year, and the types of police cases are becoming more and more complex. The traditional police case processing relies on manual reading of text, judgment of nature and classification. This mode has been difficult to meet the needs of "fast response and accurate disposal", and the intelligentization and automation of police case processing have become the development trend of the industry.
[0003] Therefore, the prior art proposes an intelligent police case processing method by introducing deep learning. However, it still has the following technical problems: the intelligent police case processing method based on deep learning often needs a large amount of labeled data for training, and the labeling cost of actual police case data is high. At the same time, the model has poor interpretability and is difficult to meet the demand for decision transparency. In order to solve the above technical problems, the prior art also proposes an intelligent police case processing method based on a large model, which reduces the demand for labeled data, but when scoring the emergency degree of a police case, the attribute extraction is one-sided and only focuses on a single attribute such as the occurrence place or the involved place, resulting in the technical problem of inaccurate police case emergency degree score.
[0004] Therefore, it is urgent to provide a police case emergency degree intelligent scoring method and system that can improve the comprehensiveness of attribute extraction when scoring the emergency degree of a police case to improve the accuracy of police case emergency degree scoring. SUMMARY
[0005] Therefore, it is necessary to provide a police case emergency degree intelligent scoring method and system to solve the technical problem of low scoring accuracy caused by scoring the emergency degree of a police case based on a single attribute in the prior art.
[0006] In order to solve the above technical problems, in a first aspect, the present application provides a police case emergency degree intelligent scoring method, comprising: obtaining police case data to be evaluated; performing multi-dimensional attribute extraction on the police case data to be evaluated to obtain multi-dimensional attribute features, the multi-dimensional attribute features including a police case nature category, an occurrence place, involved articles, involved personnel, a keyword score, a context tone score, and an external correlation emergency degree factor obtained based on external historical situation information related to the police case data to be evaluated; performing feature engineering processing on the multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition interaction relationships between different attribute features; inputting the multi-dimensional feature vector into a trained police case emergency degree scoring model to obtain an emergency degree score of the to-be-evaluated police case data; the police case emergency degree scoring model is a machine learning model representing a mapping relationship between the multi-dimensional feature vector and the emergency degree score.
[0007] In a possible implementation, the extraction process of the police case nature category, the occurrence place, the involved article, and the involved personnel includes: constructing a label knowledge base including a police case nature category label, an occurrence place label, an involved article label, and an involved personnel label; inputting the to-be-evaluated police case data into a large language model, and guiding the large language model to extract entity description texts of the involved article, the involved personnel, and the occurrence place from the to-be-evaluated police case data based on a first prompt word template; inputting the entity description texts, the to-be-evaluated police case data, and the label knowledge base into the large language model, and guiding the large language model to output attribute labels conforming to the label knowledge base specification based on a second prompt word template, the attribute labels including the police case nature category, the occurrence place, the involved article, and the involved personnel.
[0008] In a possible implementation, the extraction process of the keyword score includes: constructing a sensitive keyword set containing a basic score value, and performing word segmentation processing on the to-be-evaluated police case data to obtain a plurality of to-be-evaluated words; the basic score value is determined by expert scoring; calculating the similarity between each to-be-evaluated word and each sensitive keyword in the sensitive keyword set; taking a to-be-evaluated word with a similarity greater than a similarity threshold value as an initial keyword; eliminating a keyword in the initial keyword that is semantically repetitive with the police case nature category, the occurrence place, the involved article, and the involved personnel to obtain a target keyword; determining a basic score value of the target keyword, and determining the keyword score based on the basic score value, the similarity, and the occurrence frequency of the target keyword in the to-be-evaluated police case data.
[0009] In a possible implementation, the external correlation emergency degree factor includes a repeated alarm factor, an public opinion influence factor, a social atmosphere influence factor, and a stable situation influence factor; and the extraction process of the external correlation emergency degree factor includes: determining whether there is historical correlation data corresponding to the to-be-evaluated police case data in a preset time range, the historical correlation data including at least one of historical correlation police cases, historical correlation public opinions, historical correlation social atmospheres, and historical correlation stable situations; if not, setting the external correlation emergency degree factor as a preset value. determining the repeated alarm factor, the public opinion influence factor, the social atmosphere influence factor and the stable atmosphere influence factor based on the historical related police cases, the historical related public opinions, the historical related social atmospheres and the historical related stable atmospheres respectively if they exist; determining the weights of the repeated alarm factor, the public opinion influence factor, the social atmosphere influence factor and the stable atmosphere influence factor based on the fuzzy analytic hierarchy process, and performing weighted summation on the repeated alarm factor, the public opinion influence factor, the social atmosphere influence factor and the stable atmosphere influence factor based on the weights to obtain the external correlation emergency degree factor.
[0010] In a possible implementation, the determining the repeated alarm factor, the public opinion influence factor, the social atmosphere influence factor and the stable atmosphere influence factor based on the historical related police cases, the historical related public opinions, the historical related social atmospheres and the historical related stable atmospheres respectively if they exist, comprises: determining the total alarm times of the historical related police cases and the first alarm in the historical related police cases, and determining the repeated alarm factor based on the alarm time interval between the first alarm and the receiving time of the to-be-evaluated police case data and the total alarm times; determining the propagation time length and the influence degree score of each of the historical related public opinions, and taking the maximum value of the product of the propagation time length and the influence degree score as the public opinion influence factor; respectively calculating the product of the incoming call times and the incoming call type label score of each historical related social atmosphere, and accumulating the products corresponding to all the historical related social atmospheres, and taking the accumulation result as the social atmosphere influence factor; determining the gathering number score and the place risk score of each of the historical related stable atmospheres, and taking the maximum value of the product of the gathering number score and the place risk score as the stable atmosphere influence factor.
[0011] In a possible implementation, the feature engineering processing of the multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition interaction relationships between different attribute features comprises: performing numerical value processing on the multi-dimensional attribute features to obtain multi-dimensional numerical value features; constructing cross-combination features between different dimensional numerical value features to obtain the multi-dimensional feature vector.
[0012] In a possible implementation, the numerical value processing of the police case nature categories comprises: mapping the police case nature categories into nature category numerical values based on a pre-constructed mapping relationship; obtaining a first frequency of the police case nature categories occurring within a first set time range and a second frequency of the police case nature categories occurring within a second set time range, the first set time range being smaller than the second set time range; The property category value is enhanced based on the first frequency and the second frequency, and an enhanced property category value is obtained.
[0013] In a possible implementation, the determining whether the historical correlation data corresponding to the to-be-evaluated police case data exists in the preset time range comprises: The to-be-evaluated police case data and the plurality of historical data texts are respectively converted into a to-be-evaluated semantic vector and a plurality of historical semantic vectors. The cosine similarity of the to-be-evaluated semantic vector and each historical semantic vector is determined, and whether the historical correlation data exists is determined based on the cosine similarity.
[0014] In a possible implementation, when the historical correlation data is the historical correlation police case and the historical correlation social case, the method further comprises: The current caller of the to-be-evaluated police case data and the historical caller of each historical data text are obtained. When the current caller and the historical caller are consistent, the historical data text is taken as historical supplementary correlation data. The determining whether the historical correlation data exists based on the cosine similarity comprises: When the cosine similarity is greater than a similarity threshold, it is determined that the historical correlation data exists, the historical data text with the cosine similarity greater than the similarity threshold is taken as historical candidate correlation data, and an intersection of the historical supplementary correlation data and the historical candidate correlation data is taken as the historical correlation data.
[0015] In a second aspect, the present application further provides a police case emergency degree intelligent scoring system, comprising: A police case data acquisition unit is configured to acquire to-be-evaluated police case data. A multi-dimensional attribute extraction unit is configured to perform multi-dimensional attribute extraction on the to-be-evaluated police case data to obtain multi-dimensional attribute features, wherein the multi-dimensional attribute features comprise a police case property category, an occurrence place, an involved article, an involved person, a keyword score, a context tone score, and an external correlation emergency degree factor obtained based on external historical situation information related to the to-be-evaluated police case data. A feature engineering unit is configured to perform feature engineering processing on the multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition interaction relationships among different attribute features. An emergency degree scoring unit is configured to input the multi-dimensional feature vector into a trained police case emergency degree scoring model to obtain an emergency degree score of the to-be-evaluated police case data, wherein the police case emergency degree scoring model is a machine learning model representing a mapping relationship between the multi-dimensional feature vector and the emergency degree score.
[0016] The beneficial effects of this invention are as follows: The intelligent emergency urgency scoring method provided by this invention extracts attribute features from seven dimensions: emergency nature category, location, involved items, involved personnel, keyword score, contextual tone score, and external correlation urgency factor. These seven dimensions of attribute features integrate the structured information of the emergency data to be evaluated (emergency nature category, location, involved items, involved personnel), deep semantic and emotional information (keyword score, contextual tone score), and external macro-correlation situation information (external correlation urgency factor), achieving a comprehensive description of emergency urgency from multiple angles and in a three-dimensional manner. Simultaneously, through feature engineering, a multi-dimensional vector containing risk superposition and interaction relationships between different attribute features is generated, further realizing the description of the correlation between attribute features. The emergency urgency score determined based on the multi-dimensional feature vector overcomes the shortcomings of single-dimensional assessment methods in terms of semantic depth understanding, implicit risk capture, and situational correlation analysis, thereby improving the accuracy of the obtained emergency urgency score and providing intelligent and precise core decision support for subsequent emergency handling and dispatch. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 A schematic flowchart of an embodiment of the intelligent scoring method for the urgency of police incidents provided by the present invention; Figure 2 This is a schematic flowchart of an embodiment of the present invention for extracting the nature, type, location, items involved, and personnel involved in an emergency. Figure 3 A schematic diagram of an embodiment of the present invention for extracting keyword scores; Figure 4 A schematic diagram of an embodiment of the present invention for extracting external correlation urgency factors; Figure 5 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 6 A schematic flowchart illustrating an embodiment of the present invention for numerical processing of the nature category of police incidents; Figure 7 For the present invention Figure 4 A schematic diagram of an embodiment of S401; Figure 8This is a schematic diagram of an embodiment of the intelligent scoring system for the urgency of police incidents provided by the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0020] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.
[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0022] This invention provides an intelligent scoring method and system for the urgency of police incidents, which will be described below.
[0023] Figure 1 A schematic flowchart of an embodiment of the intelligent scoring method for the urgency of police incidents provided by the present invention is shown below. Figure 1 As shown, the intelligent scoring method for the urgency of an emergency includes: S101. Obtain the data of the police situation to be evaluated.
[0024] Specifically, the method for obtaining the alarm data to be evaluated is to automatically receive the alarm data to be evaluated pushed by the alarm system through the front-end interface or message queue.
[0025] S102. Extract multi-dimensional attributes from the police incident data to be evaluated to obtain multi-dimensional attribute features. The multi-dimensional attribute features include the nature and category of the police incident, the location of the incident, the items involved, the personnel involved, the keyword score, the context and tone score, and the external correlation urgency factor obtained based on the analysis of external historical situation information related to the police incident data to be evaluated. S103. Perform feature engineering on the multi-dimensional attribute features to generate a multi-dimensional feature vector containing the risk superposition and interaction relationship between different attribute features. S104. Input the multidimensional feature vector into the trained emergency urgency scoring model to obtain the urgency score of the emergency data to be evaluated; the emergency urgency scoring model is a machine learning model that represents the mapping relationship between the multidimensional feature vector and the urgency score.
[0026] In a specific embodiment of the present invention, the emergency urgency scoring model is a machine learning regression model based on gradient boosting tree (LightGBM regression model).
[0027] It should be understood that before executing step S104, the emergency urgency scoring model needs to be trained based on a sample set to ensure its performance. Specifically, using historical emergency data and manually labeled urgency scores as training data, the model parameters are optimized to obtain the emergency urgency scoring model.
[0028] It should also be understood that the intelligent scoring method for emergency urgency in this embodiment of the invention can be implemented in any device based on the intelligent scoring method for emergency urgency, such as an emergency analysis device or an emergency dispatch device. Specifically, the intelligent scoring method for emergency urgency is stored in the aforementioned device as a pre-programmed program. When the device is started, the program is invoked, and the intelligent scoring method for emergency urgency is implemented.
[0029] Compared with existing technologies, the intelligent emergency urgency scoring method provided in this invention extracts seven dimensions of attribute features: emergency nature category, location, involved items, involved personnel, keyword score, contextual tone score, and external correlation urgency factor. These seven dimensions of attribute features integrate the structured information of the emergency data to be evaluated (emergency nature category, location, involved items, involved personnel), deep semantic and emotional information (keyword score, contextual tone score), and external macro-correlation situation information (external correlation urgency factor), achieving a comprehensive description of emergency urgency from multiple perspectives and in a three-dimensional manner. Simultaneously, feature engineering generates a multi-dimensional vector containing risk superposition and interaction relationships between different attribute features, further realizing the description of the correlation between attribute features. The emergency urgency score determined based on the multi-dimensional feature vector overcomes the shortcomings of single-dimensional assessment methods in semantic depth understanding, implicit risk capture, and situational correlation analysis, thereby improving the accuracy of the obtained emergency urgency score and providing intelligent and precise core decision support for subsequent emergency handling and dispatch.
[0030] It should be noted that the intelligent alarm urgency scoring method in this embodiment of the invention is based on a front-end + back-end architecture. The back-end service is built using Spring Boot, and the front-end page is constructed using Vue.js. The front-end receives alarm data to be evaluated and visualizes the urgency score. The back-end receives requests from the front-end, triggers model calls, and performs processes such as multi-dimensional attribute extraction, feature engineering, and scoring.
[0031] In some embodiments of the present invention, such as Figure 2 As shown, the process for retrieving the nature and type of the incident, the location, the items involved, and the personnel involved includes: S201. Construct a tag knowledge base, which includes tags for the nature and category of the incident, tags for the location of the incident, tags for the items involved, and tags for the people involved.
[0032] Because the nature of police incidents is categorized hierarchically, the first level includes criminal, public security, disputes, and requests for assistance, while the second level, corresponding to criminal incidents, includes subcategories such as theft, robbery, and fraud. The third level, corresponding to theft, includes burglary and pickpocketing.
[0033] Therefore, the tag knowledge base adopts a tree structure design, which corresponds to the hierarchical structure of the types of police incidents, including multi-level tags, to form a complete classification system.
[0034] S202. Input the alarm data to be evaluated into the big language model, and guide the big language model to extract entity description texts involving items, personnel and locations from the alarm data to be evaluated based on the first prompt word template.
[0035] Specifically, the large language model can be the DeepSeek-v3 large model.
[0036] The first prompt word template is as follows: The police report is as follows: {temp}. Please help me extract the items involved, the people involved, and the location where the incident occurred, and return them in the following JSON format, without any other description: {"object":"","person":"","site":""}.
[0037] The temp field should be filled with the data of the incident to be evaluated.
[0038] It should be noted that the output based on the first prompt word template is the original description in the police data to be evaluated, such as the output being: {"object": "cleaver", "person": "drunk man", "site": "Fifth Municipal Hospital"}.
[0039] S203. Input the entity description text, the alarm data to be evaluated, and the tag knowledge base into the large language model, and guide the large language model to output attribute tags that conform to the tag knowledge base specifications based on the second prompt word template. The attribute tags include alarm nature category, location of occurrence, items involved, and personnel involved.
[0040] Second-stage prompt word template: The incident involves an object ({object}), a person ({person}), and occurs at a location ({site}). The incident details are as follows: {temp}. Please return the result in the following format based on the tag knowledge base and the set rules (based on predefined tag system JSON data). Only return the JSON result in the following format; do not include any irrelevant descriptive information: {"property":"","category":"","person_list":[""],"occur_site":"","involve_object":""}.
[0041] The second-stage prompts yield tagged attribute labels, such as: {"property": "public security", "category": "picking quarrels and provoking trouble", "person_list": ["intoxicated persons"], "occur_site": "medical facilities", "involve_object": "controlled knives"}.
[0042] This invention introduces a large language model to extract the nature, location, items, and personnel involved in police incidents, replacing the traditional extraction mode that relies on human experience, is inefficient, and has inconsistent standards, thus greatly improving processing efficiency and standardization.
[0043] In some embodiments of the present invention, such as Figure 3 As shown, the keyword score extraction process includes: S301. Construct a set of sensitive keywords containing basic scores, and perform word segmentation on the police data to be evaluated to obtain multiple words to be evaluated. The basic scores are determined by expert scoring.
[0044] Specifically, the base score for each sensitive keyword in the sensitive keyword set is assigned according to its urgency and risk level.
[0045] For example, the set of sensitive keywords can be obtained by combining data from typical police incidents across the country over the past three years with laws and regulations such as the "Public Security Police Sensitive Word Database," the "Criminal Law Crimes Directory," and the "Public Security Administration Punishment Law." The base score is calculated by the Delphi method, with the average score taken from the scores given by experts from the criminal investigation, public security, and intelligence command centers.
[0046] HanLP can be used for word segmentation. It should be noted that other technologies besides HanLP can also be used for word segmentation, which will not be elaborated upon here.
[0047] S302. Calculate the similarity between each word to be evaluated and each sensitive keyword in the sensitive keyword set.
[0048] The similarity is calculated using cosine similarity. Before calculating the similarity, the word to be evaluated and the sensitive keywords need to be vectorized using Word2Vec to obtain the word vector and the sensitive keyword vector. Specifically, the formula for calculating the similarity is as follows: Sim = (V1 V2) / (||V1||×||V2||) In the formula, Sim represents the similarity; V1 represents the word vector to be evaluated; and V2 represents the sensitive keyword vector. is the dot product operation; || is the L2 norm.
[0049] S303. Use words to be evaluated with a similarity greater than the similarity threshold as initial keywords.
[0050] It should be understood that the similarity threshold can be set or adjusted according to the actual application scenario. In a specific embodiment of the present invention, the similarity threshold is 0.75.
[0051] S304. Eliminate keywords from the initial keywords that are semantically redundant with the nature, type, location, items, and personnel involved in the incident to obtain the target keywords.
[0052] For example, if the item in question is "firearms and ammunition" and the initial keywords include "firearms," then that initial keyword should be deleted.
[0053] S305. Determine the base score of the target keyword, and determine the keyword score based on the base score, similarity, and frequency of occurrence of the target keyword in the alarm data to be evaluated.
[0054] Specifically, the base score of the sensitive keywords corresponding to the target keyword is used as the base score of the target keyword.
[0055] Specifically, the keyword score is:
[0056]
[0057] In the formula, Keyword score; The score for the i-th target keyword; This is the base score for the i-th target keyword; Let be the similarity score of the i-th target keyword; The frequency of occurrence of the i-th target keyword; n The number of target keywords.
[0058] This invention improves the accuracy of intelligent scoring of emergency urgency by designing a deduplication mechanism for keywords and attribute tags, thus avoiding scoring deviations caused by repeated calculations of information.
[0059] In some embodiments of the present invention, the process of extracting contextual tone scores is as follows: Obtain a pre-trained BERT model that has been pre-trained on a general dataset, and then perform specialized training based on the specific domain of alarm. Fine-tune the model parameters of the pre-trained BERT model to obtain the target BERT model. Subsequently, simply input the alarm data to be evaluated into the target BERT model to obtain the context and tone score.
[0060] Specifically, the contextual tone score ranges from 0 to 100. Since the contextual tone score represents the emotional dimension, while the scores for the nature of the incident, the location, the items involved, the personnel involved, and the keywords represent the high-risk dimension of the facts, the contextual tone score must be independent of the scores for the nature of the incident, the location, the items involved, the personnel involved, and the keywords, and should not be counted repeatedly. This is to maximize the comprehensiveness of the attribute characteristics and thus ensure the accuracy of the incident urgency score.
[0061] In some embodiments of the present invention, such as Figure 4 As shown, the external correlation urgency factors include repeat alarm factors, public opinion influence factors, social sentiment influence factors, and stability influence factors; the extraction process of external correlation urgency factors includes: S401. Determine whether there is historical related data corresponding to the alarm data to be evaluated within a preset time range. Historical related data includes at least one of historical related alarms, historical related public opinion, historical related social conditions, and historical related stability conditions.
[0062] Considering the different lifecycles and signal validity of different types of historical correlation data, in order to make reasonable use of historical correlation data and improve the accuracy of external correlation urgency factors, the preset time range for historical correlation alerts is 24 hours, the preset time range for historical correlation social conditions is 1 month, and the preset time range for historical correlation public opinion and historical correlation stability is 1 year.
[0063] The method for obtaining historical related data is to query historical related data within a preset time range through the Elasticsearch database.
[0064] S402. If it does not exist, set the external association urgency factor to the preset value.
[0065] Specifically, the default value is zero.
[0066] S403. If they exist, then the repeated alarm factor, public opinion impact factor, social impact factor and stability impact factor shall be determined based on historical related police incidents, historical related public opinion, historical related social situation and historical related stability situation, respectively.
[0067] In practical applications, if only one or more of the following historical data—related police incidents, public opinion events, social conditions, and stability maintenance—exist, the factor corresponding to the missing historical data will be set to zero. Similarly, if no historical public opinion events exist, the public opinion impact factor will be set to zero.
[0068] S404. Based on the fuzzy hierarchical analysis method (FAHP), determine the weights of the repeated alarm factor, public opinion influence factor, social situation influence factor, and stability situation influence factor. Based on the weights, perform a weighted summation of the repeated alarm factor, public opinion influence factor, social situation influence factor, and stability situation influence factor to obtain the external correlation urgency factor.
[0069] To ensure the rationality of the determined weights, after obtaining the weights, the weights need to be verified for consistency using a fuzzy consistency index. If the consistency verification passes, multiple factors are weighted and summed based on the weights to obtain the external correlation urgency factor.
[0070] In a specific embodiment of the present invention, the weights of the repeated alarm factor, public opinion influence factor, social situation influence factor, and stability influence factor are 0.1805, 0.4000, 0.1000, and 0.3195, respectively.
[0071] In some embodiments of the present invention, step S402 specifically includes: Determine the total number of alarms in historically related alarms and the first alarm in historically related alarms, and determine the duplicate alarm factor based on the alarm time interval between the first alarm and the reception time of the alarm data to be evaluated and the total number of alarms.
[0072] Specifically, when the total number of alarms is 3 or more, the number of alarms is scored as 10 points; when the total number of alarms is 2, the number of alarms is scored as 6 points; when the total number of alarms is 1, the number of alarms is scored as 0 points. When the alarm time interval is less than 60 minutes, the interval is scored as 10 points; when the alarm time interval is between 60 and 180 minutes, the interval is scored as 7 points; and when the alarm time interval is greater than 180 minutes, the interval is scored as 4 points. Among them, the repeat alarm factor R = number of times × interval.
[0073] Determine the duration of dissemination and the impact score of each historical related public opinion event, and take the maximum value of the product of the duration of dissemination and the impact score as the public opinion impact factor.
[0074] Specifically, if the duration is ≥2 hours, the dissemination duration is scored as 10 points; if the duration is >30 minutes, the dissemination duration is scored as 6 points; and if the duration is <30 minutes, the dissemination duration is scored as 1 point. When historically relevant public opinion appears on national-level platforms (such as CCTV News, Weibo trending topics, and Douyin hot lists), the influence score is 10 points; when historically relevant public opinion appears on local-level platforms (such as provincial news clients and local Weibo influencers), the influence score is 7 points; and when historically relevant public opinion appears on niche platforms (such as local forums, small WeChat groups, and general information on Douyin and Weibo), the influence score is 1 point.
[0075] Among them, the public opinion influence factor Y = max (spreading duration score × influence score).
[0076] Determine the number of calls and the call type label score for each historically related social situation, calculate the product of the number of calls and the call type label score, and use the sum of the products as the social situation influence factor.
[0077] Specifically, when the call type is a prominent, sensitive, and unsatisfactory call, the call type score is 10 points; when the call type is a key call concerning public security issues, the call type score is 7 points; when the call type is a report on clues related to pornography, gambling, or drugs, the call type score is 5 points; when the call type is a general dispute call, the call type score is 2 points; and when the call type is other, the call type score is 0 points.
[0078] Among them, the social condition influence factor S= In the formula, M represents the number of historically related social situations.
[0079] Determine the cluster size score and location risk score for each historical correlation stability situation, and use the maximum value of the product of the cluster size score and location risk score as the stability situation influence factor.
[0080] Specifically, when the number of people gathered is ≥30, the score is 10 points; when 30 > the number of people gathered is ≥10, the score is 7 points; when 10 > the number of people gathered, the score is 4 points; and when the number of people gathered is >5, the score is 2 points. When the location risk level is high, the location risk score is 10 points; when the location risk level is medium, the location risk score is 6 points; and when the location risk level is low, the location risk score is 2 points.
[0081] Among them, the stability impact factor W = max (gathering number score × location risk score).
[0082] In some embodiments of the present invention, such as Figure 5 As shown, step S103 includes: S501. Numericalize the multi-dimensional attribute features to obtain multi-dimensional numerical features.
[0083] Among them, keyword scores, contextual tone scores, and external correlation urgency factors are numerical features and do not require numerical processing. However, the nature and type of the incident, the location of the incident, the items involved, and the personnel involved are non-numerical features and require numerical processing.
[0084] Specifically, the numerical processing of the nature and type of the incident, the location of the incident, the items involved, and the personnel involved is as follows: based on the score mapping relationship constructed by experts, the nature and type of the incident, the location of the incident, the items involved, and the personnel involved are mapped into numerical features.
[0085] S502. Construct cross-combination features between numerical features of different dimensions to obtain multi-dimensional feature vectors.
[0086] In a specific embodiment of the present invention, the cross-combination features include first-order interactions and second-order interactions. Specifically, numerical features of different dimensions are used as basic feature terms, and cross-combination features are designed based on interaction features. The specific interaction feature design is shown in Table 1. Table 1. Interaction feature design for each numerical feature
[0087] This invention, through the construction of cross-combination features between numerical features of different dimensions, combines potentially interactive attribute features to form a new feature with higher information density, which greatly reduces the learning difficulty of the emergency urgency scoring model and improves the efficiency of emergency urgency scoring.
[0088] To achieve dynamic risk perception of the nature and category of police incidents, in some embodiments of the present invention, such asFigure 6 As shown, the nature of the police incident is categorized numerically, including: S601. Based on the pre-built mapping relationship, map the nature category of the alarm to the nature category value; S602. Obtain the first frequency of occurrence of the nature category of the alarm within a first set time range and the second frequency of occurrence within a second set time range, wherein the first set time range is shorter than the second set time range; S603. Based on the first frequency and the second frequency, feature enhancement is performed on the property category values to obtain enhanced property category values.
[0089] Specifically, if the first time frame is set to one month and the second time frame is set to one year, then the formula for calculating the enhanced property category value is as follows: Enhanced score = Base score × [(1 - weight) + (Monthly average / Annual average × weight)] Mean = Number of incidents of this type / Total number of incidents The weights are pre-set: the monthly average is the ratio of the first frequency to the total number of police incidents in one month, and the annual average is the ratio of the second frequency to the total number of police incidents in one year.
[0090] The embodiments of the present invention take into account that the urgency of an emergency also depends on its background. Therefore, the embodiments of the present invention dynamically adjust the static nature category values based on the first frequency and the second frequency. When the frequency of a certain type of risk is perceived to increase, its value can be automatically increased, so that the obtained emergency urgency score is more in line with the real-time situation, that is, further improve the accuracy of the determined emergency urgency score.
[0091] In some embodiments of the present invention, step S401 includes: S701. Convert the alarm data to be evaluated and multiple historical data texts into semantic vectors to be evaluated and multiple historical semantic vectors, respectively.
[0092] Specifically, Sentence-BERT is used to convert the alarm data to be evaluated and multiple historical data texts into semantic vectors to be evaluated and multiple historical semantic vectors, respectively.
[0093] S702. Determine the cosine similarity between the semantic vector to be evaluated and each historical semantic vector, and determine whether there is historical related data based on the cosine similarity.
[0094] Specifically, when the cosine similarity is greater than or equal to 0.75, the historical data text is determined to be historical related data.
[0095] Relying solely on text similarity can lead to the following misjudgments: Scenario 1: The same person reports multiple incidents to the police within a short period, but the descriptions differ significantly (e.g., the incident escalates from an argument to a fight). If only cosine similarity is used, the text variations may prevent correlation, resulting in an underestimation of duplicate alarms. Scenario 2: Different people report similar but independent events. For example, two fights in different locations with similar descriptions. If only text similarity is used, they might be associated as duplicate reports of the same event, leading to an overestimation of duplicate alarms.
[0096] To address the aforementioned technical problems, in some embodiments of the present invention, when the historical correlation data includes historical correlation police reports and historical correlation social information, the invention further includes: Obtain the current caller ID of the alarm data to be evaluated, as well as the historical caller ID of each historical data text.
[0097] Specifically, when the historical data text is related to historical police incidents, the caller is the emergency caller; when the historical data text is related to historical social events, the caller is the caller.
[0098] When the current caller and the historical caller are the same, the historical data text will be used as historical supplementary associated data. The existence of historical related data is then determined based on cosine similarity, including: When the cosine similarity is greater than the similarity threshold, it is determined that there is historical associated data. The historical data text with a cosine similarity greater than the similarity threshold is taken as historical candidate associated data, and the intersection of the historical supplementary associated data and the historical candidate associated data is taken as historical associated data.
[0099] This invention, through the introduction of the caller to supplement and verify historical correlation data, uses the intersection of historical supplementary correlation data and historical candidate correlation data as historical correlation data, thereby ensuring the accuracy of correlation between police reports and social situations and improving the accuracy of urgency scoring.
[0100] In summary, the intelligent emergency urgency scoring method proposed in this invention has the following advantages: 1. It integrates multiple models: Deepseek-v3 for attribute extraction, BERT for capturing latent signals, and LightGBM for regression scoring, thus solving the limitations of a single model and improving scoring accuracy. 2. Based on a large language model and machine learning model, it automatically completes the entire process of attribute extraction, deduplication, and scoring, improving processing speed by more than 95% compared to manual methods. This avoids frontline police officers from selecting numerous catch-all labels, meeting the need for rapid classification of massive amounts of emergency calls. Simultaneously, it unifies the labeling system and scoring algorithm, replacing subjective manual scoring and achieving consistent emergency urgency classification standards. 3. Developed based on a mature technology stack (Spring Boot, Vue, LightGBM, etc.), it can be quickly integrated into existing public security alarm receiving systems. The labeling system supports flexible iteration and is easily implemented and expanded. In other words, this invention constructs an intelligent and automated emergency urgency scoring system, providing strong technical support for modern policing work.
[0101] On the other hand, embodiments of the present invention also provide an intelligent scoring system for the urgency of police incidents, such as... Figure 8 As shown, the emergency urgency intelligent scoring system 800 includes: The police incident data acquisition unit 801 is used to acquire police incident data to be evaluated. The multi-dimensional attribute extraction unit 802 is used to extract multi-dimensional attributes from the police data to be evaluated, and obtain multi-dimensional attribute features. The multi-dimensional attribute features include the nature and category of the police incident, the location of the incident, the items involved, the personnel involved, the keyword score, the context and tone score, and the external correlation urgency factor obtained based on the analysis of external historical situation information related to the police data to be evaluated. Feature engineering unit 803 is used to perform feature engineering on multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition and interaction relationships between different attribute features. The urgency scoring unit 804 is used to input multidimensional feature vectors into a trained emergency urgency scoring model to obtain the urgency score of the emergency data to be evaluated; the emergency urgency scoring model is a machine learning model that represents the mapping relationship between multidimensional feature vectors and urgency scores.
[0102] The emergency urgency intelligent scoring system 800 provided in the above embodiments can implement the technical solutions described in the above emergency urgency intelligent scoring method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above emergency urgency intelligent scoring method embodiments, and will not be repeated here.
[0103] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0104] The above provides a detailed description of the intelligent scoring method and system for emergency situations provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. An intelligent scoring method for emergency degree of police cases, characterized in that, The method comprises the following steps: obtaining to-be-evaluated police case data; extracting multi-dimensional attributes of the to-be-evaluated police case data to obtain multi-dimensional attribute features, the multi-dimensional attribute features including a police case nature category, a place of occurrence, involved articles, involved personnel, a keyword score, a context tone score, and an external correlation emergency degree factor obtained based on external historical situation information analysis related to the to-be-evaluated police case data; performing feature engineering processing on the multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition interaction relationships between different attribute features; inputting the multi-dimensional feature vector into a trained police case emergency degree scoring model to obtain an emergency degree score of the to-be-evaluated police case data; the police case emergency degree scoring model is a machine learning model representing a mapping relationship between the multi-dimensional feature vector and the emergency degree score.
2. The method of claim 1, wherein, The extraction process of the police case nature category, the place of occurrence, the involved articles, and the involved personnel includes: constructing a label knowledge base, the label knowledge base including a police case nature category label, a place of occurrence label, an involved article label, and an involved personnel label; inputting the to-be-evaluated police case data into a large language model, guiding the large language model to extract entity description texts of the involved articles, the involved personnel, and the place of occurrence from the to-be-evaluated police case data based on a first prompt word template; inputting the entity description texts, the to-be-evaluated police case data, and the label knowledge base into the large language model, and guiding the large language model to output attribute labels conforming to the label knowledge base specifications based on a second prompt word template, the attribute labels including the police case nature category, the place of occurrence, the involved articles, and the involved personnel.
3. The method of claim 1, wherein, The extraction process of the keyword score includes: constructing a sensitive keyword set containing a basic score, and performing word segmentation processing on the to-be-evaluated police case data to obtain a plurality of to-be-evaluated words; the basic score is determined by expert scoring; calculating the similarity between each to-be-evaluated word and each sensitive keyword in the sensitive keyword set; taking to-be-evaluated words with a similarity greater than a similarity threshold as initial keywords; eliminating keywords in the initial keywords that are semantically repetitive with the police case nature category, the place of occurrence, the involved articles, and the involved personnel to obtain target keywords; determining the basic score of the target keywords, and determining the keyword score based on the basic score, the similarity, and the occurrence frequency of the target keywords in the to-be-evaluated police case data.
4. The method of claim 1, wherein, The external correlation emergency degree factor includes a repeated alarm factor, an opinion influence factor, a social atmosphere influence factor, and a stable atmosphere influence factor; the extraction process of the external correlation emergency degree factor includes: determining whether there is historical correlation data corresponding to the to-be-evaluated police case data within a preset time range, the historical correlation data including at least one of a historical correlation police case, a historical correlation opinion, a historical correlation social atmosphere, and a historical correlation stable atmosphere; if not, setting the external correlation emergency degree factor to a preset value; if so, determining the repeated alarm factor, the opinion influence factor, the social atmosphere influence factor, and the stable atmosphere influence factor based on the historical correlation police case, the historical correlation opinion, the historical correlation social atmosphere, and the historical correlation stable atmosphere, respectively; The weights of the repeated alarm factor, the public opinion influence factor, the social emotion influence factor and the stable emotion influence factor are determined based on a fuzzy analytic hierarchy process, and the repeated alarm factor, the public opinion influence factor, the social emotion influence factor and the stable emotion influence factor are weighted and summed based on the weights to obtain the external correlation emergency degree factor.
5. The method of claim 4, wherein, The repeated alarm factor, the public opinion influence factor, the social emotion influence factor and the stable emotion influence factor are respectively determined based on the historical correlation police case, the historical correlation public opinion, the historical correlation social emotion and the historical correlation stable emotion, and the method comprises the following steps of: determining the total number of alarms of the historical correlation police case and the first alarm in the historical correlation police case, and determining the repeated alarm factor based on the alarm time interval between the first alarm and the receiving time of the to-be-evaluated police case data and the total number of alarms; determining the propagation time length and the influence degree score of each historical correlation public opinion, and taking the maximum value of the product of the propagation time length and the influence degree score as the public opinion influence factor; respectively calculating the product of the number of incoming calls and the incoming call type label score of each historical correlation social emotion, and accumulating the products corresponding to all historical correlation social emotions, and taking the accumulation result as the social emotion influence factor; determining the gathering number score and the place risk score of each historical correlation stable emotion, and taking the maximum value of the product of the gathering number score and the place risk score as the stable emotion influence factor.
6. The method of claim 1, wherein, The method further comprises the following steps of: performing numerical value processing on the multi-dimensional attribute features to obtain multi-dimensional numerical value features; constructing cross-combination features between different dimensional numerical value features to obtain the multi-dimensional feature vector.
7. The method of claim 6, wherein, The method further comprises the following steps of: performing numerical value processing on the police case property category, comprising: mapping the police case property category into a property category numerical value based on a pre-constructed mapping relationship; obtaining a first frequency of the police case property category occurring within a first set time range and a second frequency of the police case property category occurring within a second set time range, the first set time range being smaller than the second set time range; 8. The method of claim 4, wherein, performing feature enhancement on the property category numerical value based on the first frequency and the second frequency to obtain an enhanced property category numerical value. The method further comprises the following steps of: performing numerical value processing on the police case property category, comprising:
9. The method of claim 8, wherein, mapping the police case property category into a property category numerical value based on a pre-constructed mapping relationship; obtaining a first frequency of the police case property category occurring within a first set time range and a second frequency of the police case property category occurring within a second set time range, the first set time range being smaller than the second set time range; performing feature enhancement on the property category numerical value based on the first frequency and the second frequency to obtain an enhanced property category numerical value. The method further comprises the following steps of: obtaining the current incoming call subject of the to-be-evaluated police case data and the historical incoming call subject of each historical data text; when the current incoming call subject and the historical incoming call subject are consistent, taking the historical data text as historical supplementary correlation data; the method further comprises the following steps of: obtaining the current incoming call subject of the to-be-evaluated police case data and the historical incoming call subject of each historical data text; when the current incoming call subject and the historical incoming call subject are consistent, taking the historical data text as historical supplementary correlation data; When the cosine similarity is greater than the similarity threshold, it is determined that there is historical correlation data, the historical data text with the cosine similarity greater than the similarity threshold is taken as historical candidate correlation data, and an intersection of the historical supplementary correlation data and the historical candidate correlation data is taken as the historical correlation data.
10. An intelligent scoring system for emergency severity of police cases, characterized in that, Comprise: A police case data acquisition unit configured to acquire to-be-evaluated police case data; A multi-dimensional attribute extraction unit configured to perform multi-dimensional attribute extraction on the to-be-evaluated police case data to obtain multi-dimensional attribute features, the multi-dimensional attribute features including a police case nature category, an occurrence place, an involved article, an involved person, a keyword score, a context tone score, and an external correlation emergency degree factor obtained based on external historical situation information analysis related to the to-be-evaluated police case data; A feature engineering unit configured to perform feature engineering processing on the multi-dimensional attribute features to generate a multi-dimensional feature vector containing risk superposition interaction relationships among different attribute features; An emergency degree scoring unit configured to input the multi-dimensional feature vector into a trained police case emergency degree scoring model to obtain an emergency degree score of the to-be-evaluated police case data, the police case emergency degree scoring model being a machine learning model representing a mapping relationship between the multi-dimensional feature vector and the emergency degree score.