System for predicting real estate transaction volume

Abstract

The invention discloses a system for predicting the real estate transaction volume, and relates to the technical field of real estate transaction prediction systems. Comprising a data acquisition module for collecting multi-source data related to the real estate transaction volume; the data preprocessing module is used for cleaning, sorting and standardizing the collected original data; and the correlation analysis module is used for calculating the correlation strength between each factor and the real estate transaction volume and screening out variables closely related to the real estate transaction volume. According to the method, investment market emotion indexes are introduced and are subjected to fusion analysis with traditional influence factors, so that the influence of investment market emotion changes on the real estate transaction volume can be captured in time, and the sensitivity and adaptability of a prediction model to market dynamics are further improved; a weighted fusion method is adopted to carry out comprehensive processing on multi-source features, and the advantages of the features are brought into full play; meanwhile, by applying a combined prediction model based on a machine learning algorithm, the advantages of various algorithms are integrated, and the precision and stability of real estate transaction volume prediction are improved.

Description

Technical Field

[0001] This invention relates to the field of real estate transaction prediction systems, and more particularly to a system for predicting real estate transaction volume. Background Technology

[0002] Accurate forecasting of real estate transaction volume is of great significance to real estate companies, investors, and government regulators. Traditional forecasting methods often rely solely on historical real estate transaction data, failing to fully consider other relevant factors and their complex relationships, resulting in limited forecast accuracy.

[0003] With the advent of the big data era, the types and quantities of available data are constantly increasing. Effectively integrating multi-source data and deeply exploring the influencing factors behind real estate transaction volumes has become crucial for improving forecast accuracy. Meanwhile, investment market sentiment is increasingly impacting the real estate market, but existing forecasting models rarely take this into account.

[0004] A search revealed Chinese patent application CN202210310100.9, which discloses a method for predicting housing prices, belonging to the field of real estate value prediction technology. This method includes the following steps: first, a test set is collected through a first data collection module; then, the received test set is processed by a first data processing module to obtain processed test set data; next, the processed test set data is input into a housing price prediction model through a first data input module; then, the housing price prediction model predicts the corresponding housing price based on the processed test set data to obtain the housing price prediction result; finally, the housing price prediction result is output to the client for reference and research through a first data output module. The housing price prediction method in the aforementioned patent has the following shortcomings: it relies on a single influencing factor and cannot comprehensively judge based on multiple factors, such as correlation factors, causal factors, and investment market sentiment; it also lacks the ability to predict real estate transaction volume. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a system for predicting real estate transaction volume.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A system for predicting real estate transaction volume includes: The data acquisition module collects multi-source data related to real estate transaction volume; The data preprocessing module cleans, organizes, and standardizes the collected raw data. The correlation analysis module calculates the correlation strength between each factor and the volume of real estate transactions, and filters out variables that are closely related to the volume of real estate transactions. The causal analysis module uncovers causal relationships between variables, identifying variables that have direct and indirect impacts on real estate transaction volume and their causal pathways. The investment market sentiment analysis module extracts investment market sentiment indicators and incorporates them as factors affecting real estate transaction volume into the prediction model. The feature fusion module integrates variables related to real estate transaction volume, which have been screened through correlation and causality analysis, with investment market sentiment indicators to form a feature set, which serves as the input to the prediction model. The feature fusion adopts a weighted fusion method, assigning corresponding weights to each feature based on the correlation strength, causal influence, and importance of each variable with real estate transaction volume and investment market sentiment indicators. The results output module outputs the predicted results of real estate transaction volume obtained from the prediction model.

[0007] Preferably, the data acquisition module collects multi-source data related to real estate transaction volume, including: real estate transaction data, macroeconomic data, population and social data, policy and regulatory data, and investment market data.

[0008] Preferably, the correlation analysis module filters variables related to real estate transaction volume, specifically in the following ways: Pearson correlation coefficient calculation: For numerical variables, calculate the Pearson correlation coefficient between them and real estate transaction volume; the formula is as follows: in, and Let x and y represent the i-th observations of variables x and y, respectively. and y represents the mean of variables x and y, respectively, and n is the sample size. The Pearson correlation coefficient between each variable and the real estate transaction volume is calculated, and variables with absolute values ​​greater than a set threshold are selected as variables that have a linear correlation with the real estate transaction volume.

[0009] Preferably, the correlation analysis module filters variables related to real estate transaction volume, and the specific methods further include: Spearman rank correlation coefficient calculation: For variables with obvious non-linear relationships, calculate their Spearman rank correlation coefficient with real estate transaction volume; the Spearman rank correlation coefficient is calculated based on the rank of the variable, and the formula is: in, The difference in rank between variables x and y is represented by n, where n is the sample size. By calculating the Spearman rank correlation coefficient between each variable and the volume of real estate transactions, variables that are correlated with the volume of real estate transactions are selected.

[0010] Preferably, the correlation analysis module filters variables related to real estate transaction volume, and the specific methods further include: Mutual information method calculation: The mutual information method is used to measure the correlation between variables and real estate transaction volume. The calculation formula is as follows: Where X and Y represent the sets of values ​​for variables x and y, respectively. Let x and y be the joint probability density function. and Let x and y be the marginal probability density functions, respectively; by calculating the mutual information value, variables with mutual information with real estate transaction volume are selected.

[0011] Preferably, the causality analysis module employs the Granger causality test method based on a vector autoregression model, with the following specific steps: Establish a VAR model: For the selected set of variables closely related to real estate transaction volume, construct a VAR model; assuming m variables... , ,…, Its VAR model expression is: in, Let be an m×1 dimensional vector representing the observed value at time t; , ,…, is an m×m dimensional coefficient matrix, representing the coefficients of the lag period; p is the lag order; Let be an m×1 dimensional error vector, satisfying a multivariate normal distribution with a mean of 0 and a constant covariance matrix; Granger causality test: For any two variables in a VAR model and Perform a Granger causality test to determine... Is it Granger causation; the null hypothesis is no The Granger causality, given past information about other variables, Past values ​​for prediction No help; construct the F-statistic for testing, as follows: in, and These are the sums of squared residuals for the constrained model and the unconstrained model, respectively. for The number of lagged terms; T is the sample size; m is the number of variables in the model; based on the calculated F-statistic and the corresponding critical value, determine whether to reject the null hypothesis; if the null hypothesis is rejected, it indicates that... yes Granger's reason, namely right There is a causal relationship; Causal path analysis: After determining the causal relationship between variables through Granger causality test, the causal path is analyzed.

[0012] Preferably, the investment market sentiment analysis module constructs the sentiment dictionary as follows: Based on the characteristics of the financial field, an investment market sentiment dictionary is constructed. The sentiment dictionary includes positive sentiment words, negative sentiment words, and neutral words, and assigns a corresponding sentiment score to each word. When calculating sentiment indicators, the specific steps are as follows: Using sentiment analysis algorithms and a constructed sentiment dictionary, we calculate the sentiment score in the text data; using a word frequency-based sentiment analysis method, we count the frequency of positive and negative sentiment words in the text, and calculate the overall sentiment score of the text based on the sentiment score weights in the sentiment dictionary. When aggregating sentiment indicators, the specific steps are as follows: By aggregating the sentiment scores of text data from different sources and at different times, a comprehensive index of investment market sentiment is obtained; by averaging or weighted averaging the sentiment scores of text data over daily, weekly, and monthly time periods, an investment market sentiment index for the corresponding time period is obtained.

[0013] Preferably, the feature fusion module determines the weights in the following way: The correlation coefficients obtained from the correlation analysis, the degree of causal influence determined from the causal analysis, and the importance assessment results of the investment market sentiment indicators are normalized so that their values ​​are between 0 and 1. Then, weight ratios are assigned to the correlation strength, causal influence degree, and importance of sentiment indicators through manual or pre-set rules. Finally, the overall weight of each feature is calculated using the following formula: in, The comprehensive weight of the i-th feature; Let be the correlation coefficient between the i-th feature and the volume of real estate transactions; The degree of causal influence of the i-th feature on the volume of real estate transactions; The importance of the investment market sentiment indicator in the i-th feature; , , These are the weighting coefficients for correlation strength, causal influence, and the importance of sentiment indicators, respectively. ; Based on the calculated comprehensive weights, each feature is weighted to obtain the weighted feature values. The weighted feature values ​​are combined to form a feature vector, which is then used as the input to the prediction model.

[0014] Preferably, the prediction model construction module specifically constructs the model as follows: Model selection: Support vector regression, random forest regression, and long short-term memory network were selected to construct the prediction model; Model training: The preprocessed dataset is divided into training and test sets according to a set ratio; the hyperparameters of each model are optimized by grid search or cross-validation, and then the model is trained on the training set to obtain the optimal model parameters; Model combination: The prediction results of the three models are combined using a weighted average method.

[0015] Preferably, the model combination of the prediction model construction module is as follows: First, calculate the mean squared error of each of the three models on the test set: in, This represents the actual volume of real estate transactions. These are the model's predicted values. This represents the number of samples in the test set. Then, the weights of each model are calculated based on their prediction errors. The weight calculation formula is as follows: in, For the first The weights of each model, For the first The prediction errors of each model are calculated; finally, the prediction results of each model are weighted and averaged to obtain the combined prediction result. in, For combined prediction results, For the first The prediction results of each model.

[0016] The beneficial effects of this invention are as follows: This invention fully integrates multi-source information such as real estate transaction data, macroeconomic data, population and social data, policy and regulatory data, and investment market data, comprehensively considering various factors affecting real estate transaction volume, thus improving the accuracy and reliability of predictions. Through correlation and causality analysis, it not only identifies variables closely related to real estate transaction volume but also delves into the causal relationships between variables, avoiding the problem of traditional prediction methods that only consider correlation while ignoring causal logic, making the prediction results more reasonable and interpretable.

[0017] This invention introduces an investment market sentiment indicator and integrates it with traditional influencing factors for analysis. This allows for timely capture of the impact of changes in investment market sentiment on real estate transaction volume, further improving the sensitivity and adaptability of the forecasting model to market dynamics.

[0018] This invention employs a weighted fusion method to comprehensively process multi-source features, fully leveraging the advantages of each feature. Simultaneously, it utilizes a combined prediction model based on machine learning algorithms, integrating the advantages of multiple algorithms to effectively improve the accuracy and stability of real estate transaction volume prediction. Attached Figure Description

[0019] Figure 1 This is a framework diagram of a system for predicting real estate transaction volume proposed in this invention; Figure 2 This is a flowchart of a Granger causality test method for predicting real estate transaction volume proposed in this invention; Figure 3 This is a flowchart of a system comprehensive weight calculation method for predicting real estate transaction volume, as proposed in this invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0021] Example 1: A system for predicting real estate transaction volume includes: The data acquisition module collects multi-source data related to real estate transaction volume; The data preprocessing module cleans, organizes, and standardizes the collected raw data. The correlation analysis module calculates the correlation strength between each factor and the volume of real estate transactions, and filters out variables that are closely related to the volume of real estate transactions. The causal analysis module uncovers causal relationships between variables, identifying variables that have direct and indirect impacts on real estate transaction volume and their causal pathways. The investment market sentiment analysis module extracts investment market sentiment indicators and incorporates them as factors affecting real estate transaction volume into the prediction model. The feature fusion module integrates variables related to real estate transaction volume, which have been screened through correlation and causality analysis, with investment market sentiment indicators to form a feature set, which serves as the input to the prediction model. The feature fusion adopts a weighted fusion method, assigning corresponding weights to each feature based on the correlation strength, causal influence, and importance of each variable with real estate transaction volume and investment market sentiment indicators. The results output module outputs the predicted results of real estate transaction volume obtained from the prediction model.

[0022] The data acquisition module collects multi-source data related to real estate transaction volume, including: Real estate transaction data: such as transaction price, transaction area, transaction time, and property location (including information such as area and location) and other historical transaction records; Macroeconomic data, such as GDP growth rate, inflation rate, interest rate, and money supply, reflect the impact of the overall economic environment on the real estate market. Population and social data: such as total population, population growth rate, age structure, household income level, urbanization process, etc. Policy and regulatory data: This includes information on the release and changes of real estate regulation policies, tax policies, land policies, etc. Policy changes often have a direct or indirect impact on the real estate market. Investment market data, such as stock market indices, bond market yields, and fund net asset values, are used to analyze investment market sentiment.

[0023] The correlation analysis module filters variables related to real estate transaction volume, specifically in the following ways: ① Pearson correlation coefficient calculation: For numerical variables, calculate their Pearson correlation coefficient with real estate transaction volume; the formula is as follows: in, and Let x and y represent the i-th observations of variables x and y, respectively. and y represents the mean of variables x and y, respectively, and n is the sample size. The Pearson correlation coefficient between each variable and the real estate transaction volume is calculated, and variables with an absolute value greater than a set threshold (such as 0.3) are selected as variables with a strong linear correlation with the real estate transaction volume. ② Spearman Rank Correlation Coefficient Calculation: For variables with a significant non-linear relationship, calculate their Spearman rank correlation coefficient with real estate transaction volume. The Spearman rank correlation coefficient is calculated based on the rank of the variable, and the formula is: in, The difference in rank between variables x and y is represented by n, which is the sample size. By calculating the Spearman rank correlation coefficient between each variable and the volume of real estate transactions, variables with a strong correlation with the volume of real estate transactions are further screened out, especially those variables with no obvious linear relationship but which may have a non-linear correlation. ③ Mutual Information Method: Considering the potentially complex nonlinear relationships between variables, the mutual information method is used to measure the correlation between variables and real estate transaction volume. The calculation formula is as follows: Where X and Y represent the sets of values ​​for variables x and y, respectively. Let x and y be the joint probability density function. and Let x and y be the marginal probability density functions, respectively; by calculating the mutual information value, variables with high mutual information with real estate transaction volume are selected.

[0024] The causality analysis module employs the Granger causality test method based on the Vector Autoregression (VAR) model, with the following specific steps: ① Establish a VAR model: For the selected set of variables closely related to real estate transaction volume, construct a VAR model; assume there are m variables. , ,…, Its VAR model expression is: in, Let be an m×1 dimensional vector representing the observed value at time t; , ,…, is an m×m dimensional coefficient matrix, representing the coefficients of the lag period; p is the lag order; Let be an m×1 dimensional error vector, satisfying a multivariate normal distribution with a mean of 0 and a constant covariance matrix; ② Granger causality test: For any two variables in the VAR model and Perform a Granger causality test to determine... Is it Granger causation; the null hypothesis is no The Granger cause, that is, given past information about other variables, Past values ​​for prediction No help; construct the F-statistic for testing, as follows: in, and Restricted models (excluding) (lagged terms) and unrestricted models (including) The sum of squared residuals (lagging terms); for The number of lagged terms; T is the sample size; m is the number of variables in the model; based on the calculated F-statistic and the corresponding critical value, determine whether to reject the null hypothesis; if the null hypothesis is rejected, it indicates that... yes Granger's reason, namely right There is a causal relationship; ③ Causal path analysis: After determining the causal relationship between variables through Granger causality test, the causal path is analyzed; For example, if variable A is a Granger cause of variable B, and variable B is a Granger cause of variable C, then it can be inferred that variable A indirectly affects variable C through variable B. By constructing a causal graph, the causal path relationships between variables can be visually displayed, providing a basis for subsequent feature fusion and predictive model construction.

[0025] The investment market sentiment analysis module includes: Using a data acquisition module, text data related to the investment market is collected from financial news websites, social media platforms, financial forums, and other channels. The data preprocessing module is used to perform preprocessing operations such as cleaning, word segmentation, and stop word removal on the collected text data; for example, removing punctuation marks, numbers, irrelevant words, etc. from the text, and decomposing the text into words or word sequences. Sentiment Dictionary Construction: Based on the characteristics of the financial field, an investment market sentiment dictionary is constructed. The sentiment dictionary includes positive sentiment words, negative sentiment words, and neutral words, and assigns a corresponding sentiment score to each word. For example, words such as "rise" and "profit" can be classified as positive sentiment words and assigned a positive sentiment score; words such as "fall" and "loss" can be classified as negative sentiment words and assigned a negative sentiment score; words such as "stable" and "fluctuation" can be classified as neutral words and assigned a sentiment score of 0. The construction of the sentiment dictionary can refer to the knowledge of financial experts, existing sentiment dictionaries, and be continuously optimized and improved through statistical analysis of a large amount of text data. Sentiment score calculation: Using sentiment analysis algorithms and a constructed sentiment lexicon, the sentiment score in the text data is calculated. For example, a word frequency-based sentiment analysis method can be used to count the frequency of positive and negative sentiment words in the text, and the overall sentiment score of the text can be calculated according to the sentiment score weights in the sentiment lexicon. In addition, factors such as the semantic information and context of the text can be considered, and more advanced sentiment analysis models, such as deep learning-based recurrent neural networks (RNN), long short-term memory networks (LSTM), or Transformer models, can be used to improve the accuracy of sentiment analysis. Sentiment index aggregation: The sentiment scores of text data from different sources and at different times are aggregated to obtain a comprehensive investment market sentiment index. For example, the sentiment scores of text data can be averaged or weighted by time periods such as daily, weekly, or monthly to obtain the investment market sentiment index for the corresponding time period. This index reflects the overall sentiment state of the investment market in a specific time period, such as optimistic, pessimistic, or neutral.

[0026] Specifically, the feature fusion module determines the weights in the following way: The correlation coefficients obtained from correlation analysis, the degree of causal influence determined from causality analysis, and the importance assessment results of investment market sentiment indicators are normalized to a value between 0 and 1. Then, weight proportions are assigned to the correlation strength, causal influence degree, and importance of sentiment indicators using manual or pre-set rules; for example, correlation strength accounts for 40%, causal influence degree accounts for 40%, and sentiment indicator importance accounts for 20%. Finally, the comprehensive weight of each feature is calculated using the following formula: in, The comprehensive weight of the i-th feature; Let be the correlation coefficient between the i-th feature and the volume of real estate transactions; The degree of causal influence of the i-th feature on real estate transaction volume (which can be quantified by relevant indicators in causal path analysis); The importance of investment market sentiment indicators in the i-th feature (for non-sentiment indicators, ); , , These are the weighting coefficients for correlation strength, causal influence, and the importance of sentiment indicators, respectively. ; Based on the calculated comprehensive weights, each feature is weighted to obtain weighted feature values. These weighted feature values ​​are then combined to form a feature vector, which serves as the input to the prediction model. For example, for the feature vector... The weighted feature vector is .

[0027] The prediction model construction module specifically constructs the model as follows: Model selection: Support Vector Regression (SVR), Random Forest Regression (RFR), and Long Short-Term Memory (LSTM) network were selected to construct the prediction model. The SVR algorithm is good at handling nonlinear problems and has good generalization ability; the RFR algorithm can handle high-dimensional data and is not prone to overfitting; the LSTM algorithm is suitable for processing time series data and can capture long-term dependencies in the data.

[0028] Model Training: The preprocessed dataset is divided into training and test sets according to a certain ratio (e.g., 70%). For the SVR model, hyperparameters such as kernel type (e.g., radial basis function kernel), penalty parameter C, and kernel parameter γ need to be determined; for the RFR model, hyperparameters such as the number of decision trees, maximum depth, and minimum number of sample splits need to be determined; for the LSTM model, hyperparameters such as the number of network layers, number of hidden units, and learning rate need to be determined. The hyperparameters of each model are optimized using methods such as grid search and cross-validation, and then the model is trained on the training set to obtain the optimal model parameters.

[0029] Model Combination: The prediction results of the three models are combined using a weighted average method. First, the prediction errors of the three models are calculated separately on the test set, such as the mean squared error (MSE). in, This represents the actual volume of real estate transactions. These are the model's predicted values. The number of samples in the test set is given; then, the weights of each model are calculated based on their prediction errors, using the following formula: in, For the first The weights of each model, For the first The prediction errors of each model are calculated; finally, the prediction results of each model are weighted and averaged to obtain the combined prediction result. in, For combined prediction results, For the first The prediction results of each model.

[0030] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A system for predicting real estate transaction volume, characterized in that, include: The data acquisition module collects multi-source data related to real estate transaction volume; The data preprocessing module cleans, organizes, and standardizes the collected raw data. The correlation analysis module calculates the correlation strength between each factor and the volume of real estate transactions, and filters out variables that are closely related to the volume of real estate transactions. The causal analysis module uncovers causal relationships between variables, identifying variables that have direct and indirect impacts on real estate transaction volume and their causal pathways. The investment market sentiment analysis module extracts investment market sentiment indicators and incorporates them as factors affecting real estate transaction volume into the prediction model. The feature fusion module integrates variables related to real estate transaction volume, which have been screened through correlation and causality analysis, with investment market sentiment indicators to form a feature set, which serves as the input to the prediction model. The feature fusion adopts a weighted fusion method, assigning corresponding weights to each feature based on the correlation strength, causal influence, and importance of each variable with real estate transaction volume and investment market sentiment indicators. The results output module outputs the predicted results of real estate transaction volume obtained from the prediction model.

2. The system for predicting real estate transaction volume according to claim 1, characterized in that, The data acquisition module collects multi-source data related to real estate transaction volume, including: real estate transaction data, macroeconomic data, population and social data, policy and regulatory data, and investment market data.

3. The system for predicting real estate transaction volume according to claim 1, characterized in that, The correlation analysis module filters variables related to real estate transaction volume, specifically in the following ways: Pearson correlation coefficient calculation: For numerical variables, calculate the Pearson correlation coefficient between them and real estate transaction volume; the formula is as follows: in, and Let x and y represent the i-th observations of variables x and y, respectively. and y represents the mean of variables x and y, respectively, and n is the sample size. The Pearson correlation coefficient between each variable and the real estate transaction volume is calculated, and variables with absolute values ​​greater than a set threshold are selected as variables that have a linear correlation with the real estate transaction volume.

4. The system for predicting real estate transaction volume according to claim 3, characterized in that, The correlation analysis module filters variables related to real estate transaction volume, and the specific methods include: Spearman rank correlation coefficient calculation: For variables with obvious non-linear relationships, calculate their Spearman rank correlation coefficient with real estate transaction volume; the Spearman rank correlation coefficient is calculated based on the rank of the variable, and the formula is: in, The difference in rank between variables x and y is represented by n, where n is the sample size. By calculating the Spearman rank correlation coefficient between each variable and the volume of real estate transactions, variables that are correlated with the volume of real estate transactions are selected.

5. A system for predicting real estate transaction volume according to claim 4, characterized in that, The correlation analysis module filters variables related to real estate transaction volume, and the specific methods include: Mutual information method calculation: The mutual information method is used to measure the correlation between variables and real estate transaction volume. The calculation formula is as follows: Where X and Y represent the sets of values ​​for variables x and y, respectively. Let x and y be the joint probability density function. and Let x and y be the marginal probability density functions, respectively; by calculating the mutual information value, variables with mutual information with real estate transaction volume are selected.

6. A system for predicting real estate transaction volume according to claim 1, characterized in that, The causality analysis module employs the Granger causality test method based on a vector autoregression model, with the following specific steps: Establish a VAR model: For the selected set of variables closely related to real estate transaction volume, construct a VAR model; assuming m variables... , ,…, Its VAR model expression is: in, Let be an m×1 dimensional vector representing the observed value at time t; , ,…, is an m×m dimensional coefficient matrix, representing the coefficients of the lag period; p is the lag order; Let be an m×1 dimensional error vector, satisfying a multivariate normal distribution with a mean of 0 and a constant covariance matrix; Granger causality test: For any two variables in a VAR model and Perform a Granger causality test to determine... Is it Granger causation; the null hypothesis is no The Granger causality, given past information about other variables, Past values ​​for prediction No help; construct the F-statistic for testing, as follows: in, and These are the sums of squared residuals for the constrained model and the unconstrained model, respectively. for The number of lagged terms; T is the sample size; m is the number of variables in the model; based on the calculated F-statistic and the corresponding critical value, determine whether to reject the null hypothesis; if the null hypothesis is rejected, it indicates that... yes Granger's reason, namely right There is a causal relationship; Causal path analysis: After determining the causal relationship between variables through Granger causality test, the causal path is analyzed.

7. A system for predicting real estate transaction volume according to claim 1, characterized in that, The investment market sentiment analysis module constructs the sentiment dictionary as follows: Based on the characteristics of the financial field, an investment market sentiment dictionary is constructed. The sentiment dictionary includes positive sentiment words, negative sentiment words, and neutral words, and assigns a corresponding sentiment score to each word. When calculating sentiment indicators, the specific steps are as follows: Using sentiment analysis algorithms and a constructed sentiment dictionary, we calculate the sentiment score in the text data; using a word frequency-based sentiment analysis method, we count the frequency of positive and negative sentiment words in the text, and calculate the overall sentiment score of the text based on the sentiment score weights in the sentiment dictionary. When aggregating sentiment indicators, the specific steps are as follows: By aggregating the sentiment scores of text data from different sources and at different times, a comprehensive index of investment market sentiment is obtained; by averaging or weighted averaging the sentiment scores of text data over daily, weekly, and monthly time periods, an investment market sentiment index for the corresponding time period is obtained.

8. A system for predicting real estate transaction volume according to claim 1, characterized in that, The feature fusion module determines the weights in the following specific way: The correlation coefficients obtained from the correlation analysis, the degree of causal influence determined from the causal analysis, and the importance assessment results of the investment market sentiment indicators are normalized so that their values ​​are between 0 and 1. Then, weight ratios are assigned to the correlation strength, causal influence degree, and importance of sentiment indicators through manual or pre-set rules. Finally, the overall weight of each feature is calculated using the following formula: in, The comprehensive weight of the i-th feature; Let be the correlation coefficient between the i-th feature and the volume of real estate transactions; The degree of causal influence of the i-th feature on the volume of real estate transactions; The importance of the investment market sentiment indicator in the i-th feature; , , These are the weighting coefficients for correlation strength, causal influence, and the importance of sentiment indicators, respectively. ; Based on the calculated comprehensive weights, each feature is weighted to obtain the weighted feature values. The weighted feature values ​​are combined to form a feature vector, which is then used as the input to the prediction model.

9. A system for predicting real estate transaction volume according to claim 8, characterized in that, The prediction model construction module specifically constructs the model as follows: Model selection: Support vector regression, random forest regression, and long short-term memory network were selected to construct the prediction model; Model training: The preprocessed dataset is divided into training and test sets according to a set ratio; the hyperparameters of each model are optimized by grid search or cross-validation, and then the model is trained on the training set to obtain the optimal model parameters; Model combination: The prediction results of the three models are combined using a weighted average method.

10. A system for predicting real estate transaction volume according to claim 9, characterized in that, The model combination of the prediction model construction module is as follows: First, calculate the mean squared error of each of the three models on the test set: in, This represents the actual volume of real estate transactions. These are the model's predicted values. This represents the number of samples in the test set. Then, the weights of each model are calculated based on their prediction errors. The weight calculation formula is as follows: in, For the first The weights of each model, For the first The prediction errors of each model are calculated; finally, the prediction results of each model are weighted and averaged to obtain the combined prediction result. in, For combined prediction results, For the first The prediction results of each model.

Images