Real estate transaction monitoring and prediction analysis method and device

Abstract

The invention relates to the technical field of big data analysis and prediction, and particularly provides a real estate transaction monitoring and prediction analysis method and device, and the method comprises the following steps: S1, data collection and preprocessing; s2, constructing and training a model; and S3, prediction and result output. Compared with the prior art, the prediction error can be obviously reduced, and the method can be applied to the fields of regulation and control, real estate department decision-making, house enterprise pushing and financial institution risk control.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis and prediction, and specifically provides a method and device for monitoring and predicting real estate transactions and analyzing them. Background Art

[0002] Accurate prediction of real estate transaction volume is the key basis for regulating the market and enterprises to formulate strategies. Although the existing technologies adopt advanced technologies and algorithms, due to the fact that most existing models are based on historical transaction data (such as ARIMA (Autoregressive Integrated Moving Average Model) time series analysis) or single economic indicators (such as housing price index, GDP growth rate), there are the following deficiencies and limitations:

[0003] Traditional models do not deeply integrate sudden variables such as deed tax adjustment and purchase restriction relaxation, and it is difficult to capture transaction fluctuations during the window period.

[0004] Only relying on annual / quarterly cycles, ignoring short-cycle rules such as working days (such as signing efficiency in the middle of the week) and holidays.

[0005] Most models do not incorporate unstructured data (such as social media sentiment, land auction public opinion), resulting in prediction lagging behind market sentiment changes.

[0006] Statistical models (such as linear regression) cannot effectively fit the complex interaction between policies, economy, and social factors.

[0007] Black box models (such as deep neural networks) cannot clearly show the weights of key influencing factors.

[0008] Therefore, in order to improve the accuracy of monitoring and prediction, it is urgent to develop a new technical method to systematically improve the existing prediction models. In addition to historical transaction data, it is also necessary to consider the impact of deed tax on real estate transaction volume, and at the same time introduce multi-source unstructured data and structured data to construct a robust and reliable influencing factor matrix, and enhance the adaptability of the prediction model in different regions. Summary of the Invention

[0009] The present invention aims at the above-mentioned deficiencies of the existing technologies and provides a practical method for monitoring and predicting real estate transactions and analyzing them.

[0010] A further technical task of the present invention is to provide a device for monitoring and predicting real estate transactions and analyzing them, which is reasonably designed, safe and applicable.

[0011] The technical solution adopted by the present invention to solve its technical problems is:

[0012] A method for monitoring and predicting real estate transactions and analyzing them has the following steps:

[0013] S1, Data collection and preprocessing;

[0014] S2, Model construction and training;

[0015] S3, Prediction and result output.

[0016] Furthermore, in step S1, it includes:

[0017] S1-1, When performing text scraping, real-time scrape the measure text of the website through network technology, provide structured fields, then obtain economic indicators, dynamically monitor the market, and perform sentiment polarity classification analysis on media public opinion based on the BERT model using natural language processing;

[0018] S1-2, Perform data preprocessing.

[0019] Furthermore, in step S1-2, it includes missing value filling, normalization processing, and feature extraction;

[0020] For the missing value filling, forward filling is used for measure fields, and linear interpolation is used for economic data;

[0021] For the normalization processing, Min-Max normalization is performed on data with dimensional differences;

[0022] The feature extraction is to extract the measure strength coefficient from the measure text and the housing purchase confidence index from the social media text.

[0023] Furthermore, in step S2, it includes:

[0024] S2-1, Feature fusion and cycle decomposition;

[0025] S2-2, Hybrid model design;

[0026] S2-3, Model training and optimization.

[0027] Furthermore, in step S2-1, it includes static features, dynamic time series features, long cycles, medium cycles, and short cycles;

[0028] The static features include the deed tax measure strength coefficient, loan interest rate, and inventory clearance cycle;

[0029] The long cycles include the annual CPI trend and the net inflow of population;

[0030] The medium cycles include the quarterly land auction premium rate;

[0031] The short cycles include the difference in trading volume between weekdays and weekends.

[0032] Further, in step S2-2, it includes an XGBoost module and an LSTM module.

[0033] The input of the XGBoost module is the measure intensity coefficient and the credit threshold value, and the output is the short-term impact probability of each measure on the trading volume.

[0034] The LSTM module captures dynamic time series features and learns long-term dependencies through a gating mechanism. The input layer is a time window, the number of neurons in the hidden layer is set to 128, and an attenuation function is designed for measure type features:

[0035] w(t) = w0·e -λt

[0036] where w0 is the initial weight, λ is the attenuation coefficient.

[0037] Further, in step S2-3, the Huber loss function is adopted to balance the robustness of MAE and MSE. The Huber loss function is:

[0038]

[0039] where δ is set to 10% of the actual trading volume fluctuation range. Bayesian optimization is used to search for the tree depth of XGBoost and the learning rate of LSTM for hyperparameter tuning. The measure weights output by XGBoost and the LSTM time series prediction results are weighted and summed, and the weights are dynamically adjusted by the validation set.

[0040] Further, in step S3, it includes real-time prediction and a visualization decision interface;

[0041] The real-time prediction includes:

[0042] (1) Input real-time data;

[0043] (2) Generate static feature vectors and dynamic time series after preprocessing;

[0044] (3) Input into the XGBoost and LSTM models respectively to obtain the policy impact sub-item S p and the time series trend sub-item S t ;

[0045] (4) Fuse and output the final predicted value:

[0046] V pred = α·S p +(1 - α)·S t ;

[0047] where α is the measure weight, which is dynamically adjusted according to the measure window period, and the default value is 0.6.

[0048] Furthermore, the visual decision-making interface includes heat map display, measure simulator, and multi-dimensional comparison;

[0049] In the heat map display, the predicted trading volume values for the next 3 months are shown by region, and the color depth represents the growth or decline amplitude.

[0050] The measure simulator allows users to manually adjust measure parameters, and the model calculates the change in the predicted value in real time.

[0051] The multi-dimensional comparison conducts historical prediction error analysis and factor contribution ranking.

[0052] A real estate transaction monitoring and prediction analysis device includes: at least one memory and at least one processor;

[0053] The at least one memory is used to store machine-readable programs;

[0054] The at least one processor is used to call the machine-readable program to execute a real estate transaction monitoring and prediction analysis method.

[0055] Compared with the prior art, a real estate transaction monitoring and prediction analysis method and device of the present invention have the following outstanding beneficial effects:

[0056] (1) It breaks through the limitation of traditional models relying on a single historical data, effectively captures the impact of sudden measures and market sentiment, and the prediction error is significantly reduced compared with the traditional ARIMA model. XGBoost accurately quantifies static measure features, and LSTM captures complex time series patterns. The dynamic weight fusion of the two enables the prediction result to have both measure sensitivity and trend continuity.

[0057] (2) Through the exponential decay model w(t) = w0·e -λt dynamically adjusts the measure weight to avoid the defect of "one-size-fits-all" in the impact of measures in traditional methods. Users can manually adjust measure parameters, and the model immediately feedbacks the change in the prediction result.

[0058] (3) Synchronously extracts long-term cycles (annual economic trends), medium-term cycles (quarterly supply and demand fluctuations), and short-term cycles (weekday-weekend differences), solving the problem that traditional models only rely on a single cycle.

[0059] (4) Outputs the weights of key influencing factors to help users understand the prediction logic and avoid the trust crisis of black-box models.

[0060] (5) Accurately predicts the effect of measure regulation, reasonably allocates the personnel windows of the real estate registration department, optimizes the opening time and pricing strategy, reduces the risk of inventory backlog; evaluates the probability of mortgage default, and dynamically adjusts the credit limit. Description of the Drawings

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0062] The attached Figure 1 is a schematic flowchart of a method for monitoring and predicting real estate transactions and analysis. Specific embodiments

[0063] To enable those skilled in the art of this technology to better understand the solutions of the present invention, the following will further elaborate on the present invention in conjunction with specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0064] The following presents a best embodiment:

[0065] As Figure 1 shown, a method for monitoring and predicting real estate transactions and analysis in this embodiment has the following steps:

[0066] S1. Data collection and preprocessing;

[0067] When performing text scraping, real-time scrape the measure text of the website through network technology and provide structured fields, and then obtain economic indicators, interface with the Bureau of Statistics API (Application Programming Interface), and obtain data such as monthly CPI (Consumer Price Index), PPI (Industrial producer price index), loan interest rate, and land auction premium rate.

[0068] Dynamically monitor the market, collect the listing volume of new / second-hand houses, inventory digestion cycle, and transaction average price from real estate trading platforms (such as Lianjia and Beike). Use natural language processing to perform sentiment polarity classification analysis (-1 to +1) on media public opinion based on the BERT model.

[0069] Missing value filling: For measure fields, use forward filling (continue to use the previous value when the measure is not adjusted), and for economic data, use linear interpolation;

[0070] Standardization processing: Perform Min-Max normalization on data with different dimensions (such as the interest rate unit is %, and the inventory quantity is sets);

[0071] Feature extraction:

[0072] Extract the "measure intensity coefficient" (ranging from 0 to 1, manually marked according to the tightness of the measure terms) from the measure text;

[0073] Extract the "housing purchase confidence index" (output based on the sentiment analysis model) from the social media text.

[0074] S2. Model construction and training;

[0075] Including:

[0076] S2-1. Feature fusion and cycle decomposition;

[0077] Including static features, dynamic time series features, long cycles, medium cycles, and short cycles;

[0078] Static features include the deed tax measure intensity coefficient, loan interest rate, and inventory clearance cycle;

[0079] Long cycles include the annual CPI trend and net population inflow;

[0080] Medium cycles include the quarterly land auction premium rate;

[0081] Short cycles include the difference in trading volume between weekdays and weekends (extracting the periodic component through Fourier transform).

[0082] S2-2. Hybrid model design;

[0083] Including the XGBoost module and the LSTM module,

[0084] The input of the XGBoost module is the measure intensity coefficient and the credit threshold value, and the output is the short-term impact probability of each measure on the trading volume;

[0085] The LSTM module captures dynamic time series features, learns long-term dependencies through the gating mechanism, the input layer is a time window (such as data for the past 6 months), the number of neurons in the hidden layer is set to 128, and a decay function is designed for measure type features:

[0086] w(t) = w0·e -λt

[0087] where w0 is the initial weight (set to 0.7 for example when the purchase restriction is lifted), λ is the decay coefficient such as λ = 0.05 for credit policies, and t is the number of months after the policy takes effect.

[0088] S2-3. Model training and optimization;

[0089] Loss function: The Huber loss function is adopted (to balance the robustness of MAE (Mean Absolute Error) and MSE (Mean Squared Error)):

[0090]

[0091] where δ is set to 10% of the actual trading volume fluctuation range.

[0092] Hyperparameter tuning: Use Bayesian optimization to search for parameters such as the tree depth (3 - 8) of XGBoost and the learning rate (0.001 - 0.1) of LSTM.

[0093] Model fusion: Weighted sum the policy weights output by XGBoost and the LSTM time series prediction results, and the weights are dynamically adjusted by the validation set.

[0094] S3, Prediction and result output;

[0095] Including real-time prediction and a visual decision-making interface;

[0096] Among them, real-time prediction includes:

[0097] (1) Input real-time data (such as the current month's loan interest rate, measure text, social media sentiment)

[0098] (2) Generate static feature vectors and dynamic time series after preprocessing;

[0099] (3) Input into the XGBoost and LSTM models respectively to obtain the policy impact sub-item S p and the time series trend sub-item S t ;

[0100] (4) Fusion output the final predicted value:

[0101] V pred = α·S p +(1 - α)·S t

[0102] where α is the measure weight (dynamically adjusted according to the measure window period, default 0.6).

[0103] In the visual decision-making interface:

[0104] Heat map display: Show the predicted values of trading volume for the next 3 months by region, and the color depth indicates the growth or decline rate;

[0105] Measure simulator: Users can manually adjust measure parameters (such as reducing the down payment ratio from 30% to 20%), and the model calculates the change in the predicted value in real time;

[0106] Multi-dimensional comparison: Support historical prediction error analysis (MAPE, RMSE), factor contribution ranking (such as showing "+15% contribution from the reduction of loan interest rate").

[0107] Take the lifting of the housing purchase restriction policy in a second-tier city in June 2024 as an example:

[0108] 1. Data input: The measure intensity coefficient suddenly increases from 0.3 to 0.8, the loan interest rate drops by 0.5%, and the social media confidence index rises from 0.2 to 0.5;

[0109] 2. Model prediction: XGBoost outputs a measure contribution value of +25%, LSTM predicts natural growth of +8% based on historical trends, and the total predicted value after fusion increases by 33%;

[0110] 3. Actual verification: The actual transaction volume in that month increased by 35%, and the prediction error was 2%, significantly better than the traditional ARIMA model (error of 12%).

[0111] This solution has been verified through simulation data sets and actual business scenarios and can be deployed on the government cloud platform or the commercial real estate data analysis system.

[0112] Based on the above method, in this embodiment, a real estate transaction monitoring and prediction analysis device includes: at least one memory and at least one processor;

[0113] The at least one memory is used to store machine-readable programs;

[0114] The at least one processor is used to call the machine-readable program and execute a real estate transaction monitoring and prediction analysis method.

[0115] The above specific implementation manners are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above specific implementation manners. Any technical solution that conforms to the above specific implementation manners of the present invention and any appropriate changes or substitutions made by those of ordinary skill in the art shall fall within the patent protection scope of the present invention.

[0116] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and predicting analysis of real estate transactions, characterized in that, It has the following steps: S1. Data collection and preprocessing; S2. Model construction and training; S3. Prediction and result output.

2. The real estate transaction monitoring and prediction analysis method according to claim 1, characterized in that In step S1, it includes: S1-1. When performing text scraping, use network technology to scrape the measure text of the website in real time, provide structured fields, then obtain economic indicators, dynamically monitor the market, and use natural language processing to perform sentiment polarity classification analysis on media public opinion based on the BERT model; S1-2. Perform data preprocessing.

3. The method for monitoring and predicting analysis of real estate transactions according to claim 2, characterized in that, In step S1-2, it includes missing value filling, normalization processing, and feature extraction; For the missing value filling, forward filling is used for the measure fields, and linear interpolation is used for economic data; For the normalization processing, Min-Max normalization is performed on the data with dimensional differences; The feature extraction is to extract the measure strength coefficient from the measure text and extract the housing purchase confidence index from the social media text.

4. A real estate transaction monitoring and prediction analysis method according to claim 3, characterized in that, In step S2, it includes: S2-1. Feature fusion and cycle decomposition; S2-2. Hybrid model design; S2-3. Model training and optimization.

5. The real estate transaction monitoring and prediction analysis method according to claim 4, characterized in that In step S2-1, it includes static features, dynamic time series features, long cycles, medium cycles, and short cycles; The static features include the deed tax measure strength coefficient, loan interest rate, and inventory clearance cycle; The long cycles include the annual CPI trend and the net inflow of population; The medium cycles include the quarterly land auction premium rate; The short cycles include the difference in trading volume between weekdays and weekends.

6. The real estate transaction monitoring and prediction analysis method according to claim 5, wherein In step S2-2, it includes the XGBoost module and the LSTM module, The input of the XGBoost module is the measure strength coefficient and the credit threshold value, and the output is the short-term impact probability of each measure on the trading volume; The LSTM module captures dynamic time series features, learns long-term dependencies through the gating mechanism, the input layer is the time window, the number of neurons in the hidden layer is set to 128, and an attenuation function is designed for the measure features: w(t) = w0·e -λt where w0 is the initial weight, λ and is the attenuation coefficient.

7. A method for monitoring and predicting analysis of real estate transactions according to claim 6, characterized in that, In step S2-3, the Huber loss function is used to balance the robustness of MAE and MSE. The Huber loss function is: where δ is set to 10% of the actual trading volume fluctuation range, Bayesian optimization is used to search for the tree depth of XGBoost and the learning rate of LSTM for hyperparameter tuning, and the measure weights output by XGBoost and the LSTM time series prediction results are weighted and summed, and the weights are dynamically adjusted by the validation set.

8. A method for monitoring and predicting analysis of real estate transactions according to claim 7, characterized in that, In step S3, it includes real-time prediction and a visual decision-making interface; The real-time prediction includes: (1). Input real-time data; (2) Generate static feature vectors and dynamic time series after preprocessing; (3) Input the XGBoost and LSTM models respectively to obtain the sub-item S of policy impact p and the sub-item S of time series trend t ; (4) Fusion output of the final predicted value: V pred = α·S p + (1 - α)·S t ; where α is the measure weight, which is dynamically adjusted according to the measure window period, and the default value is 0.

6.

9. The real estate transaction monitoring and prediction analysis method according to claim 8, characterized in that In the visual decision-making interface, it includes heat map display, measure simulator, and multi-dimensional comparison; In the heat map display, the predicted values of the trading volume for the next 3 months are displayed by region, and the color depth indicates the growth or decline amplitude The measure simulator allows users to manually adjust the measure parameters, and the model calculates the change in the predicted value in real time; The multi-dimensional comparison performs historical prediction error analysis and factor contribution degree ranking.

10. A real estate transaction monitoring and prediction analysis device, characterized in that, It includes: At least one memory and at least one processor; The at least one memory for storing a machine-readable program; The at least one processor for invoking the machine-readable program to execute the method according to any one of claims 1 to 9.

Images