A method and device for early warning of rental risks for urban villages

Abstract

The application discloses a kind of urban village-oriented leasing risk intelligent early warning method and device.In the method, the multi-modal data of urban village house leasing is collected;The multi-modal data is preprocessed and feature engineering, and an optimized feature set is generated;Based on the optimized feature set, rent correlation analysis is carried out to obtain rent derivative features;Based on the optimized feature set and the rent derivative features, a multi-task learning model is used to simultaneously predict multiple leasing risk probabilities;Establish the corresponding relationship between each risk type and the rent derivative features and determine the enhancement condition;When the probability of a certain risk is in the critical interval and the corresponding rent derivative features meet the enhancement condition, the risk is corrected;According to the correction result, calculate the comprehensive score and delimit grade, trigger early warning and generate intervention suggestions when the grade exceeds the standard.The application realizes fine identification and dynamic correction of critical risk state through the synergistic enhancement mechanism of rent derivative features and multi-task learning model, improves the early warning accuracy and sensitivity.

Description

Technical Field

[0001] This invention relates to the fields of smart city and big data and machine learning technologies, and in particular to a rental risk early warning system for urban villages. Background Technology

[0002] As a crucial component of urban housing supply, the rental market in urban villages has long faced challenges such as high rental risks, information asymmetry, and difficulties in government regulation. Currently, mainstream rental platforms and applications rely heavily on manual screening or single-dimensional static rules for risk identification. Technically, they cannot effectively integrate multi-source, heterogeneous information such as the physical condition of the property, environmental sensor data, and tenant behavior profiles, resulting in low accuracy and poor real-time performance of risk warnings. Furthermore, the lack of quantitative analysis of the correlation between dynamic market factors such as rent and risk makes it difficult for existing systems to achieve early risk perception and precise risk localization. For the unique challenges of urban villages, such as building safety hazards, fire risks, high population mobility, and illegal subletting, existing platforms lack effective automated identification and early warning mechanisms.

[0003] The main technical shortcomings of existing technologies can be summarized as follows: First, there is a lack of intelligent models that can effectively integrate multi-source heterogeneous data and quantitatively assess various potential rental risks; second, the risk assessment mechanism fails to fully utilize market-related information such as rent dynamics, resulting in insufficient foresight in early warning; third, in existing solutions, rent analysis results are only used as independent price prediction values, failing to establish a deep-level linkage triggering mechanism with various rental risks, and unable to use abnormal rent signals to finely identify and dynamically correct risks in a critical state; fourth, the data dimension is singular and the processing method is simple, failing to effectively integrate and deeply mine the value of multi-source heterogeneous data; finally, the system has poor adaptability and is difficult to self-learn and optimize the model according to market changes and new risk patterns.

[0004] Therefore, how to construct an intelligent technology solution that can integrate multi-dimensional information, accurately quantify risks, and achieve proactive early warning, so as to improve the standardization, safety, and intelligent management level of the urban village rental market, is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a rental risk early warning method and device for urban villages, which can improve the standardization, security and intelligent management level of the rental market in urban villages.

[0006] The first aspect of this invention provides a method for early warning of rental risks in urban villages, comprising: Collect multimodal data related to housing rental in urban villages, wherein the multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data; The multimodal data is preprocessed and feature-engineered to generate an optimized feature set for risk identification. The preprocessing and feature engineering include at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature selection and dimensionality reduction. Based on the optimized feature set, a deep neural network model is used to analyze the rent correlation value to obtain rent-derived features related to rental risk. The rent-derived features include at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend features. Based on the optimized feature set and the rent-derived features, a multi-task learning model is used to simultaneously predict the probability of occurrence of various rental risks. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, vacancy or early termination risk. Based on the types of various rental risks, establish a correspondence between the risk type and at least one of the aforementioned rental-derived characteristics, and determine the risk enhancement conditions for that type of risk based on the correspondence; Based on the occurrence probability of each type of rental risk, the corresponding early warning threshold, and the corresponding rental derivative characteristics, a task-level risk enhancement judgment is performed: when the occurrence probability of a certain type of rental risk is in a preset critical range, and the rental derivative characteristics corresponding to that risk type meet the risk enhancement conditions, risk correction processing is performed on that type of rental risk. A comprehensive risk score is calculated based on the revised results of various lease risks, and the comprehensive risk level is determined according to the preset scoring threshold. When the overall risk level reaches the preset warning threshold, an early warning is triggered and targeted intervention suggestions are generated.

[0007] Optionally, the data cleaning and completion includes: For missing continuous numerical features, the statistical mean of the continuous numerical features is used to fill in the missing features; For missing categorical features, the category with the highest frequency of occurrence of the categorical feature is used to fill in the missing features; For missing features with spatial correlation, a spatial interpolation method based on the K-nearest neighbor algorithm is used to fill in the missing features.

[0008] Optionally, the feature selection and dimensionality reduction include: The mutual information method is used to quantify the correlation between each feature and the prediction target; By combining L1 regularization with sparsity constraints on model weights, feature selection can be automatically achieved.

[0009] Optionally, the deep neural network model is trained by minimizing a weighted mean squared error loss function; the loss function is a weighted average of the squares of the prediction errors of each sample, wherein the weight of each sample is dynamically determined according to its confidence level or market timeliness.

[0010] Optionally, the step of simultaneously predicting the probability of occurrence of multiple rental risks using a multi-task learning model based on the optimized feature set, and calculating a comprehensive risk score based on the weighted sum of the probabilities of each risk, thereby classifying the risk level, includes: Construct the multi-task learning model, wherein the multi-task learning model includes multiple prediction sub-tasks corresponding to the number of risk types; The multi-task learning model is trained by minimizing a comprehensive loss function, which is a weighted sum of the prediction losses of each subtask and includes a regularization term to prevent overfitting. A comprehensive risk score is calculated based on the probability, risk level, or risk weight of various risks after risk correction. The comprehensive risk score is compared with a preset threshold range to determine the low, medium, and high risk levels.

[0011] Optionally, the intervention recommendations may include at least one of the following, depending on the type of risk: Send tenant risk information and preventative measures to landlords; Send high-risk alerts to housing management and public security departments; Automatic rectification suggestions are generated for houses with potential safety hazards.

[0012] Optionally, the correspondence between the risk type and the rental derivative characteristics, and the risk correction processing, include at least one of the following: When the risk type is illegal subletting or fake listings, the corresponding rent-derived feature is rent deviation. The risk enhancement condition is that the rent deviation rate exceeds a first preset threshold. The risk correction process is to increase the risk level of the risk type or decrease its warning trigger threshold. When the risk type is rent arrears risk, the corresponding rent-derived characteristic is rent payment pressure, and the risk enhancement condition is that the rent payment pressure exceeds a second preset threshold; the risk correction process is to increase the weight of this type of risk in the comprehensive risk score. When the risk type is housing safety hazard risk, the corresponding rent derivative feature is rent safety matching degree. The risk enhancement condition is that the rent safety matching degree exceeds a third preset threshold and the building's construction year exceeds a preset number of years. The risk correction process is to increase the risk level of this risk type. When the risk type is vacancy or early termination risk, the corresponding rent derivative feature is the rent change trend feature, and the risk enhancement condition is that the rent change trend feature is lower than the fourth preset threshold; the risk correction process is to increase the risk level of this risk type.

[0013] A second aspect of the present invention provides a rental risk early warning device for urban villages, comprising: The data acquisition module is used to collect multimodal data related to housing rental in urban villages. The multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data. The feature processing module is used to preprocess and feature engineer the multimodal data to generate an optimized feature set for risk identification. The preprocessing and feature engineering include at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature filtering and dimensionality reduction. The rent correlation analysis module is used to perform rent correlation value analysis based on the optimized feature set using a deep neural network model to obtain rent-derived features related to rental risk. The rent-derived features include at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend features. The risk prediction module is used to simultaneously predict the probability of occurrence of various rental risks based on the optimized feature set and the rent-derived features using a multi-task learning model. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, vacancy or early termination risk. The risk enhancement and correction module establishes a correspondence between various types of rental risks and at least one of the aforementioned rent-derived features, and determines the risk enhancement conditions for each type of risk based on the correspondence. Based on the occurrence probability of each type of rental risk, the corresponding warning threshold, and the corresponding rent-derived feature, it performs task-level risk enhancement judgment: when the occurrence probability of a certain type of rental risk is within a preset critical range, and the rent-derived feature corresponding to that risk type satisfies the risk enhancement conditions, risk correction processing is performed on that type of rental risk. The early warning output module is used to calculate a comprehensive risk score based on the corrected results of various rental risks, and to define a comprehensive risk level according to a preset scoring threshold. When the comprehensive risk level reaches the preset early warning threshold, an early warning is triggered and targeted intervention suggestions are generated.

[0014] A third aspect of the present invention provides an electronic device, comprising: One or more processors; A memory on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the rental risk warning method for urban villages as described in any of the above.

[0015] A fourth aspect of the present invention provides a computer storage medium for storing a program, which, when executed, is used to implement the rental risk warning method for urban villages as described in any of the preceding claims.

[0016] Compared with the prior art, the present invention has the following significant advantages: A precise and proactive risk early warning technology system was established: By integrating multi-source data through a multi-task learning model, simultaneous quantitative assessment of multiple risk factors was achieved. Rent correlation analysis was innovatively introduced as a key feature of risk scoring. Deep neural networks were used to uncover the complex relationships between rent and housing attributes, environment, and market conditions, serving as auxiliary indicators for risk situation perception and significantly improving the accuracy and foresight of the risk early warning model.

[0017] This invention innovatively constructs a "task-level risk enhancement mechanism": It breaks through traditional independent price forecasting by reconstructing rent correlation analysis results into task-level enhancement signals oriented towards various rental risks. By establishing a refined correspondence between rent-derived characteristics and different risk types, and setting up critical interval detection and conditionally triggered risk correction processing, it achieves secondary, accurate identification of rental behaviors at the critical point of risk assessment. This mechanism effectively solves the technical problem of existing technologies where rent information and risk warnings are disconnected and cannot be linked for triggering, significantly improving the sensitivity of risk warnings and early identification capabilities.

[0018] Enhancing data processing efficiency and model interpretability: By integrating feature engineering techniques such as mean / mode imputation, K-nearest neighbor spatial interpolation, mutual information filtering, and L1 regularization, not only are missing, noise, and redundancy issues in multi-source heterogeneous data efficiently addressed, but the model is also provided with high-quality, highly relevant input features, thereby improving model performance and feature interpretability.

[0019] Technology Integration and Integrated Innovation: Rent correlation analysis and multi-risk prediction are seamlessly integrated into the correlation analysis task, forming a technological closed loop with the early warning and response module. This not only saves system resources but also automates the entire process from risk perception, analysis, rating to intervention recommendation generation through data and model sharing and collaboration, improving system response speed and reliability.

[0020] It possesses strong adaptability and scalability: The dynamic weight loss function, regularized training, and implicit periodic update mechanism used in the model enable the system to adapt to market changes and new risk patterns, giving it good continuous learning and scalability capabilities and a long technology lifecycle. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a rental risk early warning method for urban villages provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a rental risk early warning device for urban villages provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0023] This invention provides a rental risk early warning method and device for urban villages, which can improve the standardization, security and intelligent management level of the rental market in urban villages.

[0024] See Figure 1 This figure is a flowchart illustrating a rental risk warning method for urban villages provided by an embodiment of the present invention. The rental risk warning method for urban villages provided by this embodiment of the present invention can be implemented, for example, through the following steps S101-106.

[0025] S101: Collect multimodal data related to housing rentals in urban villages.

[0026] In this embodiment of the invention, the multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data.

[0027] Specifically, multimodal data related to urban village housing rentals will be collected, including housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data. The collection methods include: Building data ( ): Collects information such as house area, house type, orientation, decoration standards, year of construction, furniture and appliances, and uses image recognition algorithms to automatically inventory and assess the condition of the facilities.

[0028] Spatial geographic data ( ): Use map APIs to obtain latitude and longitude, transportation accessibility, distance to business districts, and quantity and quality indicators of living area facilities.

[0029] Community environmental data ( ): Environmental parameters such as air quality index (AQI), noise, greening rate, and building density are obtained through IoT sensors and remote sensing images.

[0030] Market transaction data ( ): This involves using web scraping technology to integrate dynamic market information such as listing prices, transaction prices, and vacancy rates from various rental platforms.

[0031] Tenant profile data ( ): Under the premise of legal authorization, integrate credit scores, rental history, credit data, behavioral characteristics, complaint records, etc.

[0032] S102: Preprocess and feature-engineer the multimodal data to generate an optimized feature set for risk identification.

[0033] In this embodiment of the invention, preprocessing and feature engineering include at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature selection and dimensionality reduction. Data cleaning and completion includes filling in missing continuous numerical features using the statistical mean of the continuous numerical features; filling in missing categorical features using the category with the highest frequency of occurrence; and filling in missing features with spatial correlation using a spatial interpolation method based on the K-nearest neighbors algorithm. Feature selection and dimensionality reduction include quantifying the correlation between each feature and the prediction target using mutual information; and applying sparsity constraints to the model weights using L1 regularization to automatically select features.

[0034] Specifically, the feature set F undergoes data cleaning, standardization, normalization, feature encoding, feature expansion, and filtering to obtain the optimized feature set. Specifically, it includes the following: 1. Data cleaning and completion The system detects and corrects missing values, outliers, and redundant fields in the original dataset. Let the original dataset be:

[0035] in, Representing the sample Dimensional features, For corresponding labels, such as transaction rent or risk level.

[0036] For missing values, an adaptive imputation strategy based on feature type is used:

[0037] That is, when features When the variable is a continuous variable (such as numerical features like area, distance, or price), the mean of that feature is used. This is done to fill in the gaps and ensure the stability of the overall data distribution; When the feature is a discrete or categorical variable (such as decoration level, orientation, school district category, etc.), use the mode ( To fill in the gaps and maintain consistency in the category distribution; When features have spatial correlation (such as geographical location, environmental monitoring values, etc.), the K-nearest neighbor interpolation function is used. The formula for predictive imputation can be expressed as: ; in, , This represents the spatial distance from the target sample.

[0038] 2. Data Standardization and Normalization To eliminate the influence of dimensions, all continuous features are standardized so that the feature distribution has a mean of 0 and a variance of 1. The standardization formula is as follows: ; in, , denoted as the mean and standard deviation of the j-th feature, respectively.

[0039] For proportional or distance-based data (such as the distance between a building and a subway station, or the greening rate), Min-Max normalization is used to scale the feature values ​​to the [0,1] interval. The formula for Min-Max normalization is as follows: .

[0040] 3. Feature encoding Semantic encoding is used for categorical variables (such as decoration standards, apartment types, and tenant occupations). A hybrid encoding strategy is adopted: One-Hot encoding is used for variables with low cardinality (small number of categories); Embedding encoding is used for variables with high cardinality (such as occupation, school name, business district ID, etc.).

[0041] in, For a trainable embedding matrix, For the embedded dimension.

[0042] 4. Feature Expansion and Crossover To enhance the model's feature representation capabilities, cross and nonlinear features are introduced:

[0043] in, Represents the feature intersection term, calculated by two features. and The product of these factors generates higher-order interactive features, such as "house area × distance from subway," which can reflect the combined impact of location and size on rent. This represents the extended term of a polynomial feature. By squaring or cubicating the original feature, a nonlinear relationship is introduced. For example, "greening rate²" can reflect the marginal effect of environmental indicators. By combining features such as house area, geographical location, and surrounding amenities, a spatial-attribute cross-feature set is constructed. :

[0044] 5. Feature selection and dimensionality reduction Highly relevant features were screened using mutual information (MI) and feature importance evaluation methods: ; Used to measure a certain characteristic With target variable The correlation between (such as rent or risk level) when A larger value indicates that the feature contributes significantly to the target variable.

[0045] Feature sparsity selection using L1 regularization (LASSO):

[0046] in, This indicates the model's prediction error; This is an L1 norm penalty term used to force some weights to approach zero, thereby automatically removing redundant features; parameters The sparsity is controlled by the value; the larger the value, the fewer features are selected, but the model is simpler.

[0047] Output the final feature set:

[0048] S103: Based on the optimized feature set, a deep neural network model is used to analyze the rental correlation value and the rental derivative features related to rental risk.

[0049] In this embodiment of the application, the rent correlation value analysis results include at least one rent-derived feature constructed based on the rent forecast value. The rent-derived feature includes at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend characteristics. Let the rent correlation value of the property be... The actual listed price or transaction price is The average monthly disposable income of tenants is The building was built in [year]. The fire protection facilities are complete. The structural safety rating is Based on the above variables, rent-related risk characteristics are constructed, thereby transforming rent forecasting results into technical input variables for risk identification. Specifically: First, to characterize the degree of deviation between the actual housing price and the predicted reasonable rent, a rent deviation rate index is constructed. The calculation formula is as follows: ; in, This indicates the actual listing price or transaction price of the property. This represents the rent association value obtained based on multi-source feature learning. When the deviation exceeds a preset threshold (e.g., 30%), it indicates a significant deviation between the property price and its reasonable rent. This deviation can serve as a quantitative technical signal for abnormal pricing, triggering early warnings of risks such as fake listings, illegal subletting, or information concealment. This invention achieves automated detection of price anomalies through the construction of a deviation function, rather than relying on subjective judgment for evaluation.

[0050] Meanwhile, a rent-to-income ratio indicator was constructed to assess tenants' ability to fulfill their obligations and their level of financial pressure. Its definition is: ; in, This refers to the average monthly income of tenants obtained under legal authorization. This indicator normalizes rent-related values ​​to create a dimensionless risk ratio, thereby eliminating the influence of dimensions across different income levels. When the rent exceeds a preset threshold (e.g., 40% or 50%), the prediction weights of rent arrears risk and contract default risk are automatically increased and used as important input variables in the default probability model to achieve a technical assessment of the tenant's ability to perform its obligations.

[0051] Furthermore, to identify a reasonable relationship between rent levels and the physical safety attributes of buildings, a benchmark rent function based on building safety characteristics is constructed: ; in, This represents the reasonable rent level a property should command under normal and safe conditions. The matching error between the predicted rent and the benchmark rent is then calculated. ; when Exceeding a preset threshold (e.g., 25%), and the age of the buildings exceeding a certain number of years (e.g.) If the rent is not matched with the safety attributes of the property in a given year, an anomaly is determined, triggering a warning for potential structural safety hazards or missing fire safety facilities. This indicator identifies "price-masked safety risks" by constructing a consistency detection function between price and physical characteristics, thus improving the ability to detect hidden structural risks.

[0052] Furthermore, to capture regional market fluctuations and lease termination risks, time series modeling is performed on the predicted rent. Let the correlation value of the current rent be... , One month ago Then the rent change rate is defined as: ; when Below a preset threshold (e.g.) A 15% decrease indicates a significant short-term decline in rents. By combining data on changes in the community environment, such as industrial relocation or adjustments to transportation infrastructure, it is possible to identify trends of declining rental demand in advance and trigger warnings of vacancy risks or early tenant termination. This invention achieves dynamic perception of structural market changes by introducing time trend characteristics.

[0053] In one implementation of this invention, the deep neural network model is trained by minimizing a weighted mean squared error loss function; the loss function is a weighted average of the squares of the prediction errors of each sample, wherein the weight of each sample is dynamically determined according to its confidence level or market timeliness.

[0054] Specifically, in the feature set Based on this, a deep neural network (DNN) model is constructed for rent correlation analysis. The DNN model analyzes the correlation between features and property rental value, outputting a rent correlation value. This value reflects the expected value of the property under the current multidimensional features and serves as a feature input for subsequent comprehensive risk scoring. The model structure is as follows:

[0055] in, For the first Layer weights and biases; The activation function (LeakyReLU) is used to introduce nonlinearity; This is the result of the rent correlation analysis.

[0056] The loss function uses weighted mean squared error (WMSE):

[0057] in, This represents the actual transaction rent. The results of the rent correlation analysis; These are sample weights, determined by sample confidence or market timeliness, used to improve the model's sensitivity to the latest data.

[0058] The model uses the Adam optimization algorithm as follows:

[0059] In one implementation of this invention, in addition to using a deep neural network (DNN), a convolutional neural network (CNN) can also be used to process house images and CAD plan data, or a graph neural network (GNN) can be used to model spatial location and community relationships; a Transformer structure can also be used to achieve the fusion of high-dimensional sparse features, or an ensemble learning algorithm (such as XGBoost, LightGBM) can be combined with a neural network to improve valuation accuracy.

[0060] S104: Based on the optimized feature set and rental-derived features, a multi-task learning model is used to simultaneously predict the probability of occurrence of various rental risks.

[0061] In this embodiment of the invention, a multi-task learning model is constructed, comprising multiple prediction sub-tasks corresponding to the number of risk types. The multi-task learning model is trained by minimizing a comprehensive loss function, which is a weighted sum of the prediction losses of each sub-task and includes a regularization term to prevent overfitting. A comprehensive risk score is calculated based on the probabilities, risk levels, or risk weights of various risks after risk correction processing. The comprehensive risk score is compared with a preset threshold range to determine low, medium, and high risk levels. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, and vacancy or early termination risk.

[0062] Specifically, for the risk label set Establish a multi-task learning model:

[0063] Each task predicts one risk (such as default, security, or illegal activities).

[0064] The total loss function is:

[0065] in, For binary cross-entropy loss; For regularization terms; These are the weighting coefficients.

[0066] In one implementation of this invention, a multi-factor risk prediction system is constructed using a multi-task learning model, a Bayesian network, or a hybrid model combining neural networks and decision trees, to balance model interpretability and accuracy. Regarding feature engineering, an automatic feature generation algorithm (AutoFeature) can be used to automatically select feature combinations, or market dynamic features can be extracted through embedding crossovers and time-series windows. In terms of data augmentation, spatial interpolation, image enhancement, and natural language semantic analysis can be performed on geographic, image, and text data to improve the model's generalization ability.

[0067] In one implementation of this invention, model training can be completed in the cloud or a distributed cluster, or collaborative training with data privacy protection can be achieved through federated learning. Model deployment can be encapsulated as a cloud API service or embedded in mobile terminals or government regulatory platforms to achieve real-time invocation and risk linkage. The risk warning mechanism can adopt fixed or dynamic threshold strategies according to the risk type and combine spatial clustering to achieve regional risk warning. Intervention strategies can also be automatically optimized based on historical feedback through reinforcement learning algorithms to form an adaptive risk prevention and control system.

[0068] In one implementation of this invention, a periodic retraining mechanism is used to automatically optimize model parameters based on the latest transaction and risk data, thereby achieving adaptive market changes.

[0069] In one implementation of this invention, the correspondence between risk type and rental derivative characteristics can be established in advance through historical data statistical analysis or expert knowledge base. The risk correction process includes at least one of the following: When the risk type is illegal subletting or fake listing risk, the corresponding rent-derived feature is rent deviation rate, and the risk enhancement condition is that the rent deviation rate exceeds a first preset threshold; the risk correction process is to increase the risk level of this risk type or decrease its warning trigger threshold; when the risk type is rent arrears risk, the corresponding rent-derived feature is rent payment pressure, and the risk enhancement condition is that the rent payment pressure exceeds a second preset threshold; the risk correction process is to increase the weight of this type of risk in the comprehensive risk score; when the risk type is housing safety hazard risk, the corresponding rent-derived feature is rent safety matching degree, and the risk enhancement condition is that the rent safety matching degree exceeds a third preset threshold and the building's construction year exceeds a preset number of years; the risk correction process is to increase the risk level of this risk type; when the risk type is vacancy or early termination risk, the corresponding rent-derived feature is rent change trend feature, and the risk enhancement condition is that the rent change trend feature is below a fourth preset threshold; the risk correction process is to increase the risk level of this risk type.

[0070] Taking a rental dataset from an urban village as an example, the strength of the association between each rent-derived feature and each risk label can be quantified by calculating the mutual information value or correlation coefficient between them, and the feature-risk pair with the strongest association can be identified as the corresponding relationship. For example, the following correspondence matrix can be set:

[0071] S105: Based on the type of various rental risks, establish a correspondence between the risk type and at least one rental-derived feature, and determine the risk enhancement conditions for that type of risk based on the correspondence.

[0072] In this embodiment of the invention, the rental-derived features are not output as independent rental price prediction results, but rather as task-level risk enhancement signals for subsequent risk warning models.

[0073] Specifically, based on the mechanisms of various rental risks, a technical correlation is established between risk types and rent-related characteristics. For example, the risk of illegal subletting or fake listings is highly correlated with the rent deviation rate; a significant deviation of rent from a reasonable level is often an important indicator of abnormal listing information. The risk of rent arrears is directly related to the rent-to-income ratio; an excessively high rent-to-income ratio indicates that tenants lack the ability to fulfill their obligations. The risk of housing safety hazards is closely related to the rent-safety matching degree; a discrepancy between rent levels and housing safety conditions may mean that safety hazards are being concealed with low prices. The risk of vacancy or early termination is closely related to the rent change rate; a continuous decline in rent may indicate a contraction in regional rental demand.

[0074] The correspondence between risk types and rent-derived characteristics can be established in advance through historical data statistical analysis or expert knowledge bases. Taking a rental dataset from an urban village as an example, the strength of the association between each rent-derived characteristic and each risk label can be quantified by calculating the mutual information value or correlation coefficient, and the feature-risk pair with the strongest association can be identified as the corresponding relationship. For example, the following correspondence can be set: When the risk type is illegal subletting or fake listings, it is highly correlated with rent deviation. A significant deviation of rent from a reasonable level is often an important indicator of abnormal listing information, and the corresponding rent-derived characteristic is rent deviation. ; When the risk type is rent arrears risk, it is directly related to rent payment pressure. Excessive rent payment pressure indicates that the tenant's ability to fulfill their obligations is insufficient, and the corresponding rent-related characteristic is rent payment pressure. ; When the risk type is housing safety hazard risk, it is closely related to the matching degree between rent and housing safety features. The discrepancy between rent level and housing safety condition may mean that safety hazards are being concealed by low prices. The corresponding rent-derived feature is the matching degree between rent and housing safety features. ; When the risk type is vacancy or early termination risk, it is closely related to the trend of rent changes. A continuous decline in rent may indicate a contraction in regional rental demand, and the corresponding rent-derived characteristic is the trend of rent changes. .

[0075] S106: Based on the probability of occurrence of each type of lease risk, the corresponding early warning threshold, and the corresponding rental derivative characteristics, perform task-level risk enhancement judgment: when the probability of occurrence of a certain type of lease risk is in the preset critical range, and the rental derivative characteristics corresponding to that risk type meet the risk enhancement conditions, perform risk correction processing on that type of lease risk.

[0076] In this embodiment of the invention, for the first For risk-related situations, the early warning triggering depends not only on the output probability of the multi-task learning model. It also incorporates the corresponding rental derivative characteristics for comprehensive judgment. (Setting the first...) The warning threshold for this type of risk is The critical interval is The risk enhancement condition is that the corresponding rental derivative feature exceeds the preset abnormal threshold.

[0077] When the model outputs probability When this occurs, a high-risk warning for that risk type is directly triggered; when In the critical range If the corresponding rental derivative characteristics meet the risk enhancement conditions, then risk correction processing is performed on the type of rental risk. Risk correction processing includes, but is not limited to, raising the risk level of the type of rental risk, increasing the weight of the type of rental risk in the comprehensive risk score, or lowering the warning trigger threshold of the type of rental risk. If the conditions are not met, then the original risk judgment result is maintained.

[0078] For example, in a specific embodiment of the present invention, the task-level risk enhancement judgment is implemented through the following logic: Let the first The output probability of the multi-task learning model for risk class is The corresponding warning threshold is The critical interval parameter is The corresponding rental derivative features are The abnormal threshold is .

[0079] Step S1061: Preliminary risk assessment. Comparison and The size. If If so, this type of risk is directly classified as high risk, without triggering the enhancement mechanism; if If the risk is low, no warning will be triggered.

[0080] Step S1062: Critical interval detection and enhancement condition judgment. If If so, proceed to the risk enhancement assessment process. Calculate the corresponding rental-derived characteristic value. and compare it with the abnormal threshold Compare them.

[0081] Step S1063: Perform risk correction. If (Or, based on the corresponding greater than / less than characteristic type) if the risk enhancement condition is met, then the risk will be corrected.

[0082] For example, when A high-risk warning is triggered when the model probability is in the critical range and simultaneously meets the following conditions: If the risk level is not high, the risk enhancement mechanism will be used to upgrade the risk level from "medium risk" to "high risk", or to revise the risk score from 65 points to 85 points.

[0083] Similarly, when When a default risk warning is triggered; if If the model output falls within the medium-risk range, the default risk level will be increased, and the weight of this type of risk in the overall risk score will be raised.

[0084] Regarding housing safety risks, when A warning will be triggered at any time; if And the year the house was built This serves as a signal of heightened security anomalies, raising the risk level.

[0085] Regarding the risks of vacancy or early termination of lease, when A warning will be triggered at any time; if This serves as a signal of increased market volatility, prompting risk correction.

[0086] Among them, threshold Critical interval parameters and abnormal threshold It can be determined based on the statistical distribution of historical samples or through adaptive optimization during the model training process.

[0087] Through the aforementioned task-level risk enhancement mechanism, this invention enables refined identification of leasing behaviors in a risk-critical state, effectively improving the sensitivity and accuracy of risk warnings, and is particularly suitable for leasing scenarios such as urban villages where information is opaque and risk patterns are complex.

[0088] S107: Calculate a comprehensive risk score based on the revised results of various rental risks, and determine the comprehensive risk level according to the preset scoring threshold; when the comprehensive risk level reaches the preset warning threshold, trigger an warning and generate targeted intervention suggestions. In this embodiment of the invention, a comprehensive risk score is used. The weighted calculation is performed based on the adjusted results for each type of risk: ; in, This refers to the risk probability or rating score after risk correction processing.

[0089] The final risk level is defined by a piecewise function: .

[0090] When the risk level reaches a preset warning threshold, an alert is triggered and targeted intervention suggestions are generated. These intervention suggestions, depending on the type of risk, include at least one of the following: sending tenant risk information and preventative measures to landlords; sending high-risk alerts to housing management and public security departments; and generating automatic rectification suggestions for properties with safety hazards.

[0091] When the risk level is medium or high, the system triggers an early warning and outputs intervention suggestions, pushing tenant risk information and preventive measures to landlords; sending high-risk alerts to housing management and public security departments; and generating automatic rectification suggestions for houses with safety hazards. In one implementation of this invention, the invention can also be extended to scenarios such as urban renewal areas, affordable housing, long-term rental apartments, rural homestays, and commercial real estate, and can achieve API-level data connection with public security, housing management, fire protection and other departments for dynamic supervision and risk visualization of the rental market.

[0092] Beneficial Effects: This invention addresses the problems of low accuracy, poor data timeliness, and lack of risk warning mechanisms in existing urban village rental market correlation analysis models. It proposes a rental risk warning method and system for urban villages. Unlike traditional analysis methods that rely solely on manual experience or single regression models, this invention employs a four-layer architecture: multi-source data acquisition, intelligent feature extraction, deep neural network analysis, and multi-dimensional risk warning, achieving intelligent processing from data input to risk output. A missing value repair algorithm combining spatial correlation and K-nearest neighbor interpolation is introduced to ensure the continuity and comparability of housing data geographically. A feature engineering mechanism combining LASSO feature sparsity and mutual information filtering significantly improves model training efficiency and feature interpretability. At the model level, deep neural networks (DNNs) are used for rent correlation analysis, automatically capturing the nonlinear relationship between rent and multi-dimensional factors. In terms of risk management, a weighted probability model of multiple risk factors is constructed to dynamically assess and provide graded warnings for risks such as default, safety, and illegal subletting in the rental market.

[0093] Compared with existing technologies, this invention forms a scalable and self-learning model system. Therefore, this invention can significantly improve the risk controllability of the urban village rental market and has high practical value and promotion significance.

[0094] Based on the methods provided in the above embodiments, this invention also provides a rental risk warning device for urban villages. The rental risk warning device for urban villages is described below with reference to the accompanying drawings.

[0095] See Figure 2 The figure is a schematic diagram of a rental risk early warning device for urban villages provided by an embodiment of the present invention.

[0096] The rental risk early warning device 200 for urban villages provided in this embodiment of the invention includes: a data acquisition module 201, a feature processing module 202, a rent correlation analysis module 203, a risk prediction module 204, a risk enhancement and correction module 205, and an early warning output module 206.

[0097] The data acquisition module 201 is used to collect multimodal data related to housing rental in urban villages. The multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data. The feature processing module 202 is used to preprocess and feature engineer multimodal data to generate an optimized feature set for risk identification. The preprocessing and feature engineering includes at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature selection and dimensionality reduction. The rent correlation analysis module 203 is used to perform rent correlation value analysis based on the optimized feature set and using a deep neural network model to obtain rent-derived features related to rental risk. Among them, the rent-derived features include at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend features. The risk prediction module 204 is used to simultaneously predict the probability of occurrence of various rental risks based on the optimized feature set and rent-derived features using a multi-task learning model. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, vacancy or early termination risk. The risk enhancement and correction module 205 establishes a correspondence between various types of lease risks and at least one rental derivative feature, and determines the risk enhancement conditions for such risks based on the correspondence. Based on the occurrence probability, corresponding warning threshold, and corresponding rental derivative feature of each type of lease risk, it performs task-level risk enhancement judgment: when the occurrence probability of a certain type of lease risk is within a preset critical range, and the rental derivative feature corresponding to that risk type meets the risk enhancement conditions, risk correction processing is performed on that type of lease risk. The early warning output module 206 is used to calculate a comprehensive risk score based on the corrected results of various rental risks, and to define the comprehensive risk level according to the preset scoring threshold. When the comprehensive risk level reaches the preset early warning threshold, an early warning is triggered and targeted intervention suggestions are generated.

[0098] In one possible implementation, the feature processing module 202 is specifically used for: For missing continuous numerical features, the statistical mean of the continuous numerical features is used to fill in the gaps; For missing categorical features, the category with the highest frequency of occurrence is used to fill in the missing features; For missing features with spatial correlation, a spatial interpolation method based on the K-nearest neighbor algorithm is used to fill in the missing features.

[0099] In one possible implementation, the feature processing module 202 is specifically used for: The mutual information method is used to quantify the correlation between each feature and the prediction target; By combining L1 regularization with sparsity constraints on model weights, feature selection can be automatically achieved.

[0100] In one possible implementation, the rental risk early warning device 200 for urban villages also includes a model training module, specifically used for: The deep neural network model is trained by minimizing the weighted mean squared error loss function; the loss function is the weighted average of the squares of the prediction errors of each sample, where the weight of each sample is dynamically determined based on its confidence level or market timeliness.

[0101] In one possible implementation, the risk prediction module 204 has the following functions: Construct a multi-task learning model, which includes multiple prediction sub-tasks corresponding to the number of risk types; Multi-task learning models are trained by minimizing a comprehensive loss function, which is a weighted sum of the prediction losses of each subtask and includes a regularization term to prevent overfitting. A comprehensive risk score is calculated based on the probability, risk level, or risk weight of various risks after risk correction. The comprehensive risk score is compared with a preset threshold range to determine the low, medium, and high risk levels.

[0102] In one possible implementation, the intervention recommendation includes any one or a combination of the following: Send tenant risk information and preventative measures to landlords; Send high-risk alerts to housing management and public security departments; Automatic rectification suggestions are generated for houses with potential safety hazards.

[0103] In one possible implementation, the risk enhancement and correction module 205 includes at least one of the following: the correspondence between risk type and rental derivative characteristics, and the risk correction processing. When the risk type is illegal subletting or fake listings, the corresponding rent-derived characteristic is rent deviation. The risk enhancement condition is that the rent deviation rate exceeds the first preset threshold. The risk correction process is to increase the risk level of this risk type or lower its warning trigger threshold. When the risk type is rent arrears risk, the corresponding rent-derived characteristic is rent payment pressure, and the risk enhancement condition is that the rent payment pressure exceeds the second preset threshold; the risk correction process is to increase the weight of this type of risk in the comprehensive risk score. When the risk type is housing safety hazard risk, the corresponding rent derivative feature is rent safety matching degree. The risk enhancement condition is that the rent safety matching degree exceeds the third preset threshold and the building's construction year exceeds the preset number of years. The risk correction process is to upgrade the risk level of this risk type. When the risk type is vacancy or early termination risk, the corresponding rent derivative feature is the rent change trend feature, and the risk enhancement condition is that the rent change trend feature is lower than the fourth preset threshold; the risk correction process is to upgrade the risk level of this risk type.

[0104] Since the rental risk warning device 200 for urban villages is a device corresponding to the rental risk warning method for urban villages provided in the above method embodiments, the specific implementation of each module of the rental risk warning device 200 for urban villages is based on the same concept as that in the above method embodiments. Therefore, for the specific implementation of each module of the rental risk warning device 200 for urban villages, please refer to the description of the rental risk warning method for urban villages in the above method embodiments, and it will not be repeated here.

[0105] This invention also provides a rental risk early warning device for urban villages, the device comprising: a processor and a memory; The memory is used to store instructions; The processor is used to execute the instructions in the memory to perform the rental risk warning method for urban villages mentioned in the above embodiments.

[0106] It should be noted that the hardware structure of the rental risk early warning device for urban villages provided in the embodiments of the present invention can be as follows: Figure 3 The structure shown, Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention.

[0107] Please see Figure 3 As shown, device 300 includes: a processor 310, a communication interface 320, and a memory 330. The number of processors 310 in device 300 can be one or more. Figure 3 Taking a processor as an example, in this embodiment of the invention, the processor 310, communication interface 320, and memory 330 can be connected via a bus system or other means. Figure 3 Taking the connection between China and Israel via bus system 340 as an example.

[0108] Processor 310 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. Processor 310 may further include hardware chips. These hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0109] The memory 330 may include volatile memory, such as random-access memory (RAM); the memory 330 may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 330 may also include a combination of the above types of memory.

[0110] Optionally, the memory 330 stores an operating system and programs, executable modules, or data structures, or subsets thereof, or extended sets thereof. The programs may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic business processes and handling hardware-based tasks. The processor 310 can read the programs from the memory 330 to implement the rental risk warning method for urban villages provided in this embodiment of the invention.

[0111] The bus system 340 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus system 340 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0112] This invention also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to execute the rental risk warning method for urban villages mentioned in the above embodiments.

[0113] This invention also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the rental risk warning method for urban villages mentioned in the above embodiments.

[0114] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for early warning of rental risks in urban villages, characterized in that, include: Collect multimodal data related to housing rental in urban villages, wherein the multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data; The multimodal data is preprocessed and feature-engineered to generate an optimized feature set for risk identification. The preprocessing and feature engineering include at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature selection and dimensionality reduction. Based on the optimized feature set, a deep neural network model is used to analyze the rent correlation value to obtain rent-derived features related to rental risk. The rent-derived features include at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend features. Based on the optimized feature set and the rent-derived features, a multi-task learning model is used to simultaneously predict the probability of occurrence of various rental risks. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, vacancy or early termination risk. Based on the types of various rental risks, establish a correspondence between the risk type and at least one of the aforementioned rental-derived characteristics, and determine the risk enhancement conditions for that type of risk based on the correspondence; Based on the occurrence probability of each type of rental risk, the corresponding early warning threshold, and the corresponding rental derivative characteristics, a task-level risk enhancement judgment is performed: when the occurrence probability of a certain type of rental risk is in a preset critical range, and the rental derivative characteristics corresponding to that risk type meet the risk enhancement conditions, risk correction processing is performed on that type of rental risk. A comprehensive risk score is calculated based on the revised results of various lease risks, and the comprehensive risk level is determined according to the preset scoring threshold. When the overall risk level reaches the preset warning threshold, an early warning is triggered and targeted intervention suggestions are generated.

2. The method according to claim 1, characterized in that, The data cleaning and completion includes: For missing continuous numerical features, the statistical mean of the continuous numerical features is used to fill in the missing features; For missing categorical features, the category with the highest frequency of occurrence of the categorical feature is used to fill in the missing features; For missing features with spatial correlation, a spatial interpolation method based on the K-nearest neighbor algorithm is used to fill in the missing features.

3. The method according to claim 1, characterized in that, The feature selection and dimensionality reduction include: The mutual information method is used to quantify the correlation between each feature and the prediction target; By combining L1 regularization with sparsity constraints on model weights, feature selection can be automatically achieved.

4. The method according to claim 1, characterized in that, The deep neural network model is trained by minimizing a weighted mean squared error loss function; the loss function is a weighted average of the squares of the prediction errors of each sample, wherein the weight of each sample is dynamically determined according to its confidence level or market timeliness.

5. The method according to claim 1, characterized in that, Based on the optimized feature set, a multi-task learning model is used to simultaneously predict the probability of occurrence of various rental risks, and a comprehensive risk score is calculated based on the weighted sum of the probabilities of each risk, thereby classifying the risk level, including: Construct the multi-task learning model, wherein the multi-task learning model includes multiple prediction sub-tasks corresponding to the number of risk types; The multi-task learning model is trained by minimizing a comprehensive loss function, which is a weighted sum of the prediction losses of each subtask and includes a regularization term to prevent overfitting. A comprehensive risk score is calculated based on the probability, risk level, or risk weight of various risks after risk correction. The comprehensive risk score is compared with a preset threshold range to determine the low, medium, and high risk levels.

6. The method according to claim 1, characterized in that, The intervention recommendations, depending on the type of risk, include at least one of the following: Send tenant risk information and preventative measures to landlords; Send high-risk alerts to housing management and public security departments; Automatic rectification suggestions are generated for houses with potential safety hazards.

7. The method according to claim 1, characterized in that, The correspondence between the risk type and the rental derivative characteristics, and the risk correction process, include at least one of the following: When the risk type is illegal subletting or fake listings, the corresponding rent-derived feature is rent deviation. The risk enhancement condition is that the rent deviation rate exceeds a first preset threshold. The risk correction process is to increase the risk level of the risk type or decrease its warning trigger threshold. When the risk type is rent arrears risk, the corresponding rent-derived characteristic is rent payment pressure, and the risk enhancement condition is that the rent payment pressure exceeds a second preset threshold; the risk correction process is to increase the weight of this type of risk in the comprehensive risk score. When the risk type is housing safety hazard risk, the corresponding rent derivative feature is rent safety matching degree. The risk enhancement condition is that the rent safety matching degree exceeds the third preset threshold and the building age exceeds the preset number of years. The risk correction process involves raising the risk level of this risk type. When the risk type is vacancy or early termination risk, the corresponding rent derivative feature is the rent change trend feature, and the risk enhancement condition is that the rent change trend feature is lower than the fourth preset threshold; the risk correction process is to increase the risk level of this risk type.

8. A rental risk early warning device for urban villages, characterized in that, include: The data acquisition module is used to collect multimodal data related to housing rental in urban villages. The multimodal data includes at least housing data, spatial geographic data, community environmental data, market transaction data, and tenant profile data. The feature processing module is used to preprocess and feature engineer the multimodal data to generate an optimized feature set for risk identification. The preprocessing and feature engineering include at least data cleaning and completion, data standardization and normalization, feature encoding, feature expansion and cross-referencing, and feature filtering and dimensionality reduction. The rent correlation analysis module is used to perform rent correlation value analysis based on the optimized feature set using a deep neural network model to obtain rent-derived features related to rental risk. The rent-derived features include at least rent deviation, rent payment pressure, rent safety matching degree, and rent change trend features. The risk prediction module is used to simultaneously predict the probability of occurrence of various rental risks based on the optimized feature set and the rent-derived features using a multi-task learning model. The various rental risks include at least one or more of the following: rent arrears or default risk, illegal subletting or fake listing risk, housing safety hazard risk, vacancy or early termination risk. The risk enhancement and correction module establishes a correspondence between various types of rental risks and at least one of the aforementioned rent-derived features, and determines the risk enhancement conditions for each type of risk based on the correspondence. Based on the occurrence probability of each type of rental risk, the corresponding warning threshold, and the corresponding rent-derived feature, it performs task-level risk enhancement judgment: when the occurrence probability of a certain type of rental risk is within a preset critical range, and the rent-derived feature corresponding to that risk type satisfies the risk enhancement conditions, risk correction processing is performed on that type of rental risk. The early warning output module is used to calculate a comprehensive risk score based on the corrected results of various rental risks, and to define a comprehensive risk level according to a preset scoring threshold. When the comprehensive risk level reaches the preset early warning threshold, an early warning is triggered and targeted intervention suggestions are generated.

9. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store instructions; The processor is configured to execute the instructions in the memory to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Including instructions that, when run on a computer, cause the computer to perform the method described in any one of claims 1-7 above.

Images