CN-122020037-A - Intelligent evaluation system for crime risk of young and young people

CN122020037ACN 122020037 ACN122020037 ACN 122020037ACN-122020037-A

Abstract

The invention relates to the technical field of crime risk assessment, in particular to an intelligent evaluation system for the crime risk of young people, which comprises a data acquisition module, a dynamic modeling module, a stage prediction module, an intervention recommendation module and a privacy protection module, wherein the data acquisition module is used for acquiring multi-source heterogeneous data of individual teenagers, the dynamic modeling module is used for mapping a multi-dimensional feature vector into a time-varying risk state vector by using a continuous time neural network, the stage prediction module is used for dividing an implementation stage of the crime risk of the young people based on the current risk state vector, the intervention recommendation module is used for automatically selecting and pushing intervention measures suitable for the current implementation stage based on probability and time points, and the privacy protection module is used for encrypting and anonymously processing personal privacy. The invention can completely capture the progressive process of risk accumulation, motivation formation and behavior implementation, is not limited to static analysis of a single time node, accurately calculates the occurrence probability of each stage and the time point of the next stage, and provides scientific basis for the selection of intervention opportunities.

Inventors

ZHONG WEIFANG
MO LEI
GAO RUIXIANG
GUO YONGXING
HUANG XIAOTONG

Assignees

广东司法警官职业学院

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (9)

1. The intelligent evaluation system for the crime risk of the young and young people is characterized by comprising: The data acquisition module is used for acquiring multi-source heterogeneous data of teenager individuals, wherein the multi-source heterogeneous data comprises learning records, social contents, body signals and campus activity tracks, and desensitizing the multi-source heterogeneous data to generate multi-dimensional feature vectors Wherein In order to learn to record the score, For the score of the social content, For the body signal to be scored, Scoring the activity track; a dynamic modeling module for using a continuous time neural network to model the multidimensional feature vector Mapping to time-varying risk state vector, outputting current risk state vector ; The stage prediction module is used for dividing the implementation stages of the crime risks of the teenagers based on the current risk state vector, wherein the implementation stages comprise a front motivation stage, a motivation stage and a behavior implementation stage Wherein For the probability of being in the forward stage, For the probability of being in the motivational phase, For the probability of being in the behavior implementation stage, simultaneously calculating the time point of the next stage; The intervention recommendation module is used for automatically selecting and pushing intervention measures suitable for the current implementation stage based on the probability and the time point; the privacy protection module is used for encrypting and anonymously processing the personal privacy in the whole process of data collection, modeling, prediction and intervention.
2. The intelligent evaluation system for the risk of a young offender crime of claim 1, wherein the desensitization process comprises the steps of: S101, identifying a sensitive field, marking the name, the school number, the identity card number, the mobile phone number and the GPS coordinate of teenagers as direct identifiers, and marking the combination of birth year, class and sex as standard identifiers; s102, performing global hash transformation on the direct identifier, performing date blurring or interval binning on the alignment identifier, and applying differential privacy noise to the logarithmic field; S103, finally generating the desensitized multidimensional feature vector 。
3. The intelligent assessment system for crime risk of young people according to claim 1, wherein learning records in the data acquisition module score Obtained by the following steps: s201, collecting the learning record, wherein the learning record comprises the original score of m classes Rate of completion of job Duty failure rate Teacher evaluation text Wherein ; S202, evaluating the text for the teacher Extraction of emotion polarity using pre-trained emotion model Extracting the key words to obtain education key word density The keywords include late arrival, progress and disturbance; s203, constructing five-dimensional vectors of each course Will be Input educational performance scoring model Outputting the class level The education performance scoring model is a three-layer full-connection network and sequentially comprises a first full-connection layer, a second full-connection layer and an output full-connection layer; S204, scoring the learning record Taking all course fractions Is calculated as: ; Wherein, the For the total number of take courses, Grading weights for the course; The Cheng Jifen number weights The calculation formula is as follows: ; Wherein, the Penalty coefficients for absences.
4. The intelligent assessment system for crime risk of young people according to claim 1, wherein the social content score Obtained by the following steps: S301, acquiring the social content, wherein the social content comprises text, voice and pictures of a social platform extracted from a teenager mobile phone end, and uniformly coding the text, voice and pictures to obtain text word vectors Meier spectrum of speech And picture visual vector ; S302, constructing a multi-mode emotion recognition network The multi-modal emotion recognition network is formed by sequentially connecting a text channel, a voice channel, an image channel, a cross-modal attention fusion layer, a gating circulation unit layer and a random inactivation layer, and is used for carrying out text word vector Meier spectrum of speech And picture visual vector Inputting the multi-modal emotion recognition network, and outputting three intermediate indexes including negative emotion intensity Aggressive vocabulary duty cycle And population interaction density ; S303, weighting and fusing the three intermediate indexes to obtain social content scores The calculation formula is as follows: ; Wherein, the 、 And Is a fusion weight 。
5. The intelligent assessment system for crime risk of young people according to claim 1, wherein the body signal score Obtained by the following steps: s401, acquiring the body signal, wherein the body signal comprises dermatomes acquired from a wearable device at a 30S fixed window Heart rate And blood oxygen ; S402, calculating the dermatome in each window Heart rate And blood oxygen Mean, variance, and peak value of (a); S403, constructing a physiological pressure state model, wherein the physiological pressure state model is formed by sequentially connecting a long-short-period memory network layer LSTM, a linear transformation layer and a Sigmoid activation layer, inputting the mean value, the variance and the peak value into the physiological pressure state model, outputting a body signal score, and the formula is as follows: ; Wherein, the As a function of the Sigmoid, In order to output the weight matrix, For the 64-dimensional hidden vector that LSTM outputs in the last time step, To output a bias term.
6. The intelligent assessment system for crime risk of young people according to claim 1, wherein the activity trajectory score Obtained by the following steps: s501, acquiring the campus activity track, wherein the campus activity track comprises a coordinate sequence obtained from a campus Wi-Fi log Wherein n is the total number of coordinate sequences, and the coordinate sequences are subjected to stay point detection through a clustering algorithm DBSCAN to obtain a stay point set ; S502, setting a campus high risk area, and acquiring a center coordinate of the high risk area Calculating a dwell point The shortest distance to the center of all high risk areas is given by: ; Wherein, the Is the stay point The shortest distance to the center of all high risk areas, For the kth dwell point in the dwell point set P, the coordinates are , Is the coordinates of the center of the first high risk area, Is the euclidean norm; According to Acquiring dwell points The number of times of falling in the 30m buffer area of any high risk area, and the frequency of the visit is obtained And obtaining the night detention time length according to the accumulated detention minutes in the high risk area ; S503, calculating an activity track score The formula is: ; Where n is the total number of recorded dwell points, For each stay point Risk density of (a) The calculation formula of (2) is as follows: ; Wherein, the 、 And As the weight coefficient of the light-emitting diode, As a parameter of the distance decay, Is the total duration of the night.
7. The intelligent evaluation system for the risk of a young crime according to claim 1, wherein the continuous-time neural network in the dynamic modeling module adopts a neural differential equation structure to define a risk state vector Wherein d is the risk state vector dimension, and the specific evolution equation is: ; Wherein, the Is as the parameter of The three-layer perceptron network of (3) has the following specific structure: The first layer, wherein the input dimension is d+4, the output dimension is 64, and the activation function is ReLU; The second layer, the input dimension is 64, the output dimension is 32, and the activation function is ReLU; The third layer is that the input dimension is 32, the output dimension is d, and the activation function is Tanh; Solving the differential equation by using a numerical integration method to obtain each time step Risk state vector of (a) At the current time step Outputting the risk state vector As an output of the dynamic modeling module.
8. The intelligent assessment system of the risk of a young year crime according to claim 1, wherein the phase prediction module divides the implementation phase of the risk of a young year crime based on the current risk status by: s601, inputting the risk state vector output by the dynamic modeling module The dimension is d; S602, classifying network using pre-training Vector risk status Mapping into probability distribution of three implementation phases Determining the current stage according to the probability distribution, if Highest, the current is in the forward engine stage, if Highest, the current is in the motivation stage, if Highest, then currently in the behavior implementation phase; s603, setting a dimension threshold of the risk state vector If a risk state vector is detected Is the derivative of (2) At a certain moment A dimension corresponding to 0 and a derivative of 0 is greater than a dimension threshold When it is, the time is The potential triggering time is marked, and an early warning mechanism is triggered; s604, assuming that the residual time obeys the exponential distribution, the rate parameter thereof Obtained from the historical sample maximum likelihood estimation, the time point estimation of the next stage is: ; Wherein, the Is the point in time at which the next phase occurs.
9. The intelligent evaluation system for crime risk of young people according to claim 1, wherein the intervention recommendation module, when selecting an intervention measure, depends on a probability distribution Point in time when the next phase occurs Building a priority score, wherein the formula is as follows: ; Wherein, the The priority score of which is given, And In order to set the weight of the weight in the preset, For a preset constant, the system matches the intervention templates in the database in descending order of priority scores S.

Description

Intelligent evaluation system for crime risk of young and young people Technical Field The invention relates to the technical field of crime risk assessment, in particular to an intelligent assessment system for crime risk of young and young people. Background In the field of crime risk prediction of young and young people, the prior art has obvious static limitation, most models are often focused on risk factors of a single time node, only specific aspects such as psychological pains or illegal behaviors are concerned, but dynamic development rules in the process from potential states to manifestation of risks cannot be fully considered, the progressive process of 'risk accumulation-motivation formation-behavior implementation' is difficult to be completely reflected by the static view angle, so that unilaterality exists in capturing of risk evolution, and the change track of risks along with time cannot be accurately tracked. The existing prediction tools have obvious defects in factor integration, which are used for analyzing various factors such as environment, personality, emotion and the like in many isolation ways and lack of systematic integration of risk chains, and particularly, the remote vulnerable background such as long-term home environment cannot be organically connected with elements such as near-end triggering events such as negative life events and immediate psychological reactions such as vulnerable mood in series to form a complete risk analysis framework, and the fragmentation analysis mode leads to insufficient comprehensive risk identification and is difficult to grasp the inherent association and interaction among the factors. The prior art is fuzzy in stage division and has insufficient consideration of adjustment factors, key stages of crime risk development, such as essential differences before and after motivation formation, are not clearly distinguished, so that intervention measures are lack of pertinence, the situation that the same intervention strategy is adopted for individuals without obvious motivation and individuals with motivation is frequently caused, intervention time lag or resource mismatch is easy to occur, and meanwhile, protective resources such as active psychological capital and social support, and negative adjustment factors such as anti-social cognition are insufficient in importance in risk conversion, the probability of risk upgrading is difficult to accurately judge, and effectiveness and scientificity of an intervention effect are influenced. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides an intelligent evaluation system for the risk of crimes of young people, which can effectively solve the problems in the prior art. In order to achieve the above purpose, the invention is realized by the following technical scheme: the invention provides an intelligent evaluation system for crime risks of young and young people, which comprises the following components: The data acquisition module is used for acquiring multi-source heterogeneous data of teenager individuals, wherein the multi-source heterogeneous data comprises learning records, social contents, body signals and campus activity tracks, and desensitizing the multi-source heterogeneous data to generate multi-dimensional feature vectors WhereinIn order to learn to record the score,For the score of the social content,For the body signal to be scored,Scoring the activity track; a dynamic modeling module for using a continuous time neural network to model the multidimensional feature vector Mapping to time-varying risk state vector, outputting current risk state vector; The stage prediction module is used for dividing the implementation stages of the crime risks of the teenagers based on the current risk state vector, wherein the implementation stages comprise a front motivation stage, a motivation stage and a behavior implementation stageWhereinFor the probability of being in the forward stage,For the probability of being in the motivational phase,For the probability of being in the behavior implementation stage, simultaneously calculating the time point of the next stage; The intervention recommendation module is used for automatically selecting and pushing intervention measures suitable for the current implementation stage based on the probability and the time point; the privacy protection module is used for encrypting and anonymously processing the personal privacy in the whole process of data collection, modeling, prediction and intervention. Further, the desensitization treatment comprises the following steps: S101, identifying a sensitive field, marking the name, the school number, the identity card number, the mobile phone number and the GPS coordinate of teenagers as direct identifiers, and marking the combination of birth year, class and sex as standard identifiers; s102, performing global hash transformation on the direct identifier, performing date