Real estate asset operation management method and system based on big data analysis

Abstract

The invention discloses a real estate asset operation management method and system based on big data analysis, and relates to the field of real estate operation management, and the method comprises the steps: generating the prices of daily granularity, weekly granularity and monthly granularity according to collected real estate data through employing a time sequence analysis algorithm; respectively adopting an exponential smoothing algorithm to obtain a daily house price change curve, a weekly house price change curve and a monthly house price change curve; according to the section data, adopting a DBSCAN algorithm to cluster the real estate; dividing sub-clusters according to room types and floors; calculating the average price and the median price of each real estate cluster and each real estate sub-cluster to obtain price mean value distribution and price median distribution of different sections, house types and floor combinations; and constructing a real estate transaction probability prediction model through a logistic regression algorithm, and carrying out real estate asset operation management. In view of insufficient decision basis in real estate asset operation management in the prior art, the precision of real estate asset operation management is improved.

Description

Technical Field

[0001] This application relates to the field of real estate operation and management, and particularly to a method and system for real estate asset operation and management based on big data analysis. Background Art

[0002] Currently, there is a common problem of insufficient decision-making basis in real estate asset operation and management. Traditional operation and management mainly rely on expert experience and subjective judgment, lacking objective and quantitative data support, and it is difficult to adapt to the increasingly complex and changeable market environment. The specific manifestations are as follows:

[0003] Firstly, in the real estate planning and development stage, there is a lack of accurate analysis of the regional market and demand prediction, resulting in frequent problems such as blind investment and imbalance between supply and demand. Secondly, in the real estate pricing and marketing links, there is insufficient understanding of the factors affecting real estate prices, the pricing strategy lacks flexibility and pertinence, the marketing means are single, and it is difficult to effectively match customer needs. Thirdly, in the process of real estate rental, sale and property management, there is a lack of dynamic monitoring of asset value and income, insufficient exploration of asset appreciation potential, and uneven management and service levels.

[0004] The root cause of these problems lies in the lack of scientific and efficient data analysis and decision-making support means in real estate asset operation and management. A large amount of real estate data is deposited in various business systems and departments, with data fragmentation, low data quality, and insufficient exploration of data value, and it fails to be transformed into the driving force for management optimization.

[0005] In order to solve this problem, it is urgent to introduce advanced big data analysis concepts and technologies, fully explore and utilize the value of real estate data, and build a data-driven intelligent asset operation and management system. Through comprehensive data collection, multi-dimensional integration and in-depth mining, describe the internal value law of real estate assets, insight into the changing trend of market supply and demand, identify risks and opportunities, optimize resource allocation, and improve operation efficiency and management level. Summary of the Invention

[0006] Aiming at the problem of insufficient decision-making basis in real estate asset operation and management in the prior art, this application provides a method and system for real estate asset operation and management based on big data analysis. By multi-dimensionally depicting the spatio-temporal distribution law of real estate prices and constructing a real estate transaction probability prediction model based on machine learning algorithms, a data-driven real estate management plan is formed, improving the accuracy of real estate asset operation and management.

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] One aspect of the present application provides a method for real estate asset operation management based on big data analysis, including: collecting real estate data, where the real estate data includes house prices, locations, house types, and floors; according to the collected real estate data, using a time series analysis algorithm to aggregate house prices in the time dimension to generate prices at daily granularity, weekly granularity, and monthly granularity; respectively using an exponential smoothing algorithm to smooth the daily granularity price, weekly granularity price, and monthly granularity price to obtain a daily house price change curve, a weekly house price change curve, and a monthly house price change curve; according to the location data in the collected real estate data, using the DBSCAN algorithm to cluster the real estate to obtain real estate clusters in different locations; within each real estate cluster, dividing sub-clusters by house type and floor respectively; calculating the average price and median price of each real estate cluster and sub-cluster to obtain the price mean distribution and price median distribution of different combinations of location, house type, and floor; according to the daily house price change curve, the weekly house price change curve, the monthly house price change curve, as well as the price mean distribution and price median distribution, constructing a real estate transaction probability prediction model through a logistic regression algorithm, where the real estate transaction probability prediction model takes real estate data as input and the transaction probability of the real estate as output, and performing real estate asset operation management according to the predicted transaction probability.

[0009] Furthermore, generate prices at daily, weekly, and monthly granularities, including: extracting the timestamps of each real estate data, sorting the real estate data in ascending order according to the timestamps to form a time-ordered housing price sequence; traversing each housing price data in the time-ordered housing price sequence to obtain the timestamp of the corresponding housing price data; using regular expressions to extract the year, month, and day data from the timestamp and concatenating them into a date string; using the hash table data structure, with the date string as the key, adding the housing prices corresponding to the date to the value list of the hash table; traversing each key-value pair in the hash table, calculating the arithmetic mean of the housing price lists in the value list to obtain the daily granularity price for the corresponding date; traversing each housing price data in the time-ordered housing price sequence to obtain the timestamp of the corresponding housing price data; using regular expressions to extract the year, month, and day data from the timestamp, calculating the corresponding week number based on the date as the key for weekly granularity grouping; using the hash table data structure, with the weekly granularity grouping as the key, adding the housing prices within the corresponding week to the value list of the hash table; traversing each key-value pair in the hash table, calculating the arithmetic mean of the housing price lists in the value list to obtain the weekly granularity price for the corresponding week; traversing each housing price data in the time-ordered housing price sequence to obtain the timestamp of the corresponding housing price data; using regular expressions to extract the year and month data from the timestamp and concatenating them into a month string; using the hash table data structure, with the month string as the key, adding the housing prices corresponding to the month to the value list of the hash table; traversing each key-value pair in the hash table, calculating the arithmetic mean of the housing price lists in the value list to obtain the monthly granularity price for the corresponding month.

[0010] Among them, a Hash Table is a data structure that stores data based on key-value pairs. It uses a hash function to map a key to a specific location in the table to achieve fast data insertion, deletion, and lookup operations. The advantage of a hash table is that it can achieve a constant-level time complexity on average and is suitable for scenarios that require frequent data lookup and insertion. In this application, the hash table is used to group and aggregate housing price data by date, week number, or month. By using the date, week number, or month as the key and storing the corresponding housing price data in the value of the hash table, the housing price data can be quickly grouped and statistically analyzed. In a hash table, the key is the unique identifier used to index and lookup data. Each key corresponds to a value, and the corresponding value can be quickly located and accessed through the key. In this application, the date string: the date string formed by concatenating the year, month, and day data, such as "2023-05-18", is used as the key of the hash table to group the housing price data by date. Weekly granularity grouping: Calculate the corresponding week number based on the date and use it as the key of the hash table to group the housing price data by week. Month string: The month string formed by concatenating the year and month data, such as "2023-05", is used as the key of the hash table to group the housing price data by month. By using time information at different granularities as keys, it is convenient to perform aggregations and analyses at different levels on the housing price data.

[0011] In a hash table, the value list is a set of values corresponding to the key. Each key can correspond to multiple values, and these values are stored in the hash table in the form of a list. In this application, the value list is used to store housing price data belonging to the same date, week, or month. Specifically: For the calculation of daily granularity prices, use the date string as the key and store the housing price data for the same date in the corresponding value list. For the calculation of weekly granularity prices, use the week number as the key and store the housing price data for the same week in the corresponding value list. For the calculation of monthly granularity prices, use the month string as the key and store the housing price data for the same month in the corresponding value list. By storing the housing price data in the value list, it is convenient to perform aggregations and statistical analyses on the housing price data within the same time granularity, such as calculating the average value, median, etc.

[0012] Further, obtain the daily housing price change curve, weekly housing price change curve, and monthly housing price change curve, including: setting a smoothing coefficient a; specifically, in this application, the value of the smoothing coefficient a is dynamically adjusted according to the data characteristics in the recent period. A fixed-size sliding window can be set, the statistical indicators of the data within the window are calculated, and the value of a is adjusted according to the size of the indicators. More specifically, determine that the size of the sliding window is n, indicating that the data of the most recent n time points is considered. The size of the sliding window can be determined according to factors such as the frequency and change speed of the data. For example, for daily-granularity price data, the size of the sliding window can be set to 7, indicating that the data of the most recent week is considered; for weekly-granularity price data, the size of the sliding window can be set to 4, indicating that the data of the most recent month is considered. At each time point, obtain the data within the sliding window to form a data subset of size n. Calculate the statistical indicators for the data subset within the sliding window. Commonly used indicators include: Variance: Measures the degree of dispersion of the data and reflects the volatility of the data. The larger the variance, the greater the data volatility; the smaller the variance, the more stable the data. Range: The difference between the maximum and minimum values of the data within the window, reflecting the range of the data. The larger the range, the greater the data volatility; the smaller the range, the more stable the data. Moving average: The arithmetic average of the data within the window, reflecting the overall level of the data. The change trend of the moving average can reflect the rising and falling trend of the data. Moving median: The median of the data within the window, which is more robust to outliers and can reflect the central position of the data. Set the thresholds for the statistical indicators, and divide the volatility of the data into different levels according to the thresholds. For example, the variance can be divided into three levels: low volatility, medium volatility, and high volatility. According to the level of data volatility, dynamically adjust the value of the smoothing coefficient a. Generally speaking: When the data fluctuates greatly (such as when the variance or range exceeds the high volatility threshold), increase the value of a so that the model pays more attention to the recent observations and quickly responds to the changes in the data. The value of a can be adjusted to a larger value, such as 0.7 or 0.8. When the data is relatively stable (such as when the variance or range is below the low volatility threshold), decrease the value of a so that the model pays more attention to the historical observations and obtains a smoother curve. The value of a can be adjusted to a smaller value, such as 0.2 or 0.3. When the data volatility is at a medium level, the value of a can be maintained unchanged, or the value of a can be appropriately adjusted to balance short-term fluctuations and long-term trends. By setting a sliding window, calculating the statistical indicators of the data within the window, and dynamically adjusting the value of the smoothing coefficient a according to the size of the indicators, the exponential smoothing model can adapt to the dynamic changes of the data, quickly respond to changes when the data fluctuates greatly, obtain a smoother curve when the data is relatively stable, and improve the prediction accuracy and adaptability of the housing price change curve.

[0013] Taking the first element of the daily-granularity price, weekly-granularity price, or monthly-granularity price as the initial smoothing value, using the smoothing value at the previous moment and the current actual value, recursively calculate the first-order exponential smoothing value at each moment to obtain the first-order exponential smoothing sequence; taking the first element of the first-order exponential smoothing sequence as the initial smoothing value, using the second-order smoothing value at the previous moment and the current first-order smoothing value, recursively calculate the second-order exponential smoothing value at each moment to obtain the second-order exponential smoothing sequence; according to the first-order exponential smoothing sequence and the second-order exponential smoothing sequence, calculate the smoothed predicted value of the daily-granularity price, weekly-granularity price, or monthly-granularity price in a future period of time to obtain the daily housing price change curve, weekly housing price change curve, or monthly housing price change curve.

[0014] Furthermore, use the DBSCAN algorithm to cluster real estates to obtain real estate clusters in different locations, including: obtaining real estate data, extracting the location data in the real estate data to form a location feature vector; preprocessing the location feature vector;

[0015] Set the density threshold parameter ε; and set the minimum number of points threshold MinPts according to the location data; select any unvisited location feature point p, calculate the number of points within the ε-neighborhood of p; if the number of points within the ε-neighborhood of p is greater than or equal to MinPts, mark p as a core point, determine a new cluster starting from p, and classify all unclassified points within the ε-neighborhood of p into the corresponding cluster; if the number of points within the ε-neighborhood of p is less than MinPts, mark p as a noise point; for the location feature points marked as noise points, if there is any core point within the corresponding ε-neighborhood, classify the core point into the corresponding cluster, otherwise mark the core point as an outlier; output all clusters in the location feature space by the DBSCAN algorithm, each cluster corresponds to a real estate location cluster, and the real estates within the location cluster are geographically aggregated;

[0016] Furthermore, set the density threshold parameter ε; and set the minimum number of points threshold MinPts according to the location data, including: calculating the Euclidean distance between each location feature vector and other location feature vectors to construct a distance matrix of location feature points; sorting each row of the distance matrix in ascending order to obtain the K-nearest neighbor distance list of each location feature point; selecting the Kth nearest neighbor distance of each location feature point as the KNN distance to form a set of KNN distances of location feature points; sorting the set of KNN distances of location feature points in ascending order, plotting the KNN distance curve graph, with the horizontal axis being the serial number of the location feature point and the vertical axis being the corresponding KNN distance value; on the KNN distance curve graph, calculate the slope value of each point, obtain the first inflection point with a slope value greater than the preset threshold, and use the KNN distance value corresponding to the first inflection point as the density threshold parameter ε; count the total number N of location feature vectors, and set the minimum number of points threshold MinPts according to N.

[0017] Among them, the K-nearest neighbor distance list refers to the distance list obtained by calculating the Euclidean distance between each lot feature point and other lot feature points and sorting them in ascending order of the distance values. The list contains the distance values between each lot feature point and other lot feature points, arranged in ascending order. In this application, for each lot feature point, first calculate the Euclidean distance between it and all other lot feature points to form a distance vector. Then, sort this distance vector in ascending order to obtain the K-nearest neighbor distance list of this lot feature point. The first element in the list represents the distance value of the lot feature point closest to this lot feature point, the second element represents the distance value of the lot feature point with the second closest distance to this lot feature point, and so on. The K-nearest neighbor distance list reflects the distance relationship between each lot feature point and other lot feature points in the feature space and can be used to analyze the similarity and clustering characteristics between lots.

[0018] The KNN distance refers to selecting the Kth nearest neighbor distance in the K-nearest neighbor distance list as the KNN distance of this lot feature point. The KNN distance reflects the distance value between each lot feature point and its Kth nearest neighbor lot feature point. In this application, for the K-nearest neighbor distance list of each lot feature point, select the Kth element in the list as the KNN distance of this lot feature point. For example, if K = 5, then select the 5th element in the list as the KNN distance of this lot feature point. The KNN distance can be used to measure the local density of lot feature points in the feature space. A lot feature point with a smaller KNN distance indicates that there are more neighboring points around it and a higher local density; while a lot feature point with a larger KNN distance indicates that there are fewer neighboring points around it and a lower local density.

[0019] The KNN distance curve graph is a visualization tool used to display the distribution of the KNN distances of lot feature points. In the KNN distance curve graph, the horizontal axis represents the serial number of the lot feature points, and the vertical axis represents the corresponding KNN distance value. In this application, first sort the KNN distances of all lot feature points in ascending order to obtain a KNN distance set. Then, use the serial number of the lot feature points as the abscissa and the corresponding KNN distance value as the ordinate to draw the KNN distance curve graph. The position of each point in the graph represents the KNN distance value of this lot feature point. By observing the shape and trend of the KNN distance curve graph, the distribution characteristics of the KNN distance values can be found. The positions where obvious inflection points or steep changes occur on the curve graph usually indicate significant changes in the distribution of the KNN distance values and can be used as a reference for determining the density threshold parameter ε.

[0020] Furthermore, within each real estate cluster, sub-clusters are divided according to housing type and floor, including: extracting the corresponding housing type attributes and floor attributes based on the real estate data within each real estate plot cluster; where the housing type attributes include the number of rooms and the number of bathrooms, and the floor attributes include the floor and the total number of floors; converting the housing type attributes into a numerical housing type feature vector, calculating the Euclidean distance matrix of housing types for the real estate plot cluster according to the housing type feature vector; using the hierarchical clustering algorithm to cluster the Euclidean distance matrix of housing types to obtain K1 housing type sub-clusters; according to the K1 housing type sub-clusters, converting the floor attributes into a numerical floor feature vector, calculating the Euclidean distance matrix of floors for the sub-clusters according to the floor feature vector; using the K-means clustering algorithm to cluster the Euclidean distance matrix of floors to obtain K2 housing type sub-clusters; according to the K1 housing type sub-clusters and the K2 housing type sub-clusters, dividing each real estate plot cluster into K1*K2 housing type-floor sub-clusters;

[0021] Furthermore, calculate the average price and median price of each real estate cluster and sub-cluster to obtain the price mean distribution and price median distribution of different plot, housing type, and floor combinations, including: extracting the transaction price list price_list of the real estate plot cluster according to each real estate plot cluster; traversing the real estate plot cluster and adding the transaction price of each real estate to price_list; calculating the arithmetic mean of price_list to obtain the average price cluster_avg_price of the real estate plot cluster; sorting price_list in ascending order and selecting the element in the middle position as the median price cluster_median_price of the real estate plot cluster; according to each housing type-floor sub-cluster, extracting the transaction price list sub_price_list of all real estates within the housing type-floor sub-cluster; calculating the arithmetic mean of sub_price_list to obtain the average price sub_cluster_avg_price of the housing type-floor sub-cluster; sorting sub_price_list in ascending order and selecting the element in the middle position as the median price sub_cluster_median_price of the housing type-floor sub-cluster; where a dictionary is a data structure composed of key-value pairs, and it indexes and accesses the corresponding value through the key (key). Each key is unique in the dictionary and can be used to identify and retrieve the associated value. In this application, the plot average price dictionary: uses the plot information as the key and stores the corresponding average price as the value. The price mean of the plot can be quickly found through the plot information.

[0022] Lot median price dictionary: Using lot information as the key, store the corresponding median price as the value. Through the lot information, the price median of that lot can be quickly found. Housing type - floor average price dictionary: Using the combination of lot, housing type, and floor information as the key, store the corresponding average price as the value. Through the combination of lot, housing type, and floor, the average price of this combination can be quickly found. Housing type - floor median price dictionary: Using the combination of lot, housing type, and floor information as the key, store the corresponding median price as the value. Through the combination of lot, housing type, and floor, the price median of this combination can be quickly found. When outputting the average price distribution and price median distribution of different combinations of lot, housing type, and floor, the corresponding average price and median price information can be directly obtained through the keys of the dictionary, which is convenient and fast.

[0023] Furthermore, calculate the average price and median price of each real estate cluster and sub - cluster to obtain the average price distribution and price median distribution of different combinations of lot, housing type, and floor. It also includes: initializing the lot average price dictionary region_avg_dict, the lot median price dictionary region_median_dict, the housing type - floor average price dictionary type_floor_avg_dict, and the housing type - floor median price dictionary type_floor_median_dict; traversing each real estate lot cluster, taking the lot information region of the real estate lot cluster as the key; adding the average price cluster_avg_price of the real estate lot cluster as the value to region_avg_dict[region]; adding the median price cluster_median_price of the real estate lot cluster as the value to region_median_dict[region]; traversing each housing type - floor sub - cluster, concatenating the lot information region, housing type information type, and floor information floor to form region_type_floor as the key; adding the average price sub_cluster_avg_price of the housing type - floor sub - cluster as the average price value to

[0024] type_floor_avg_dict[region_type_floor]; Use the median price sub_cluster_median_price of the house type - floor sub - cluster as the median price value and add it to type_floor_median_dict[region_type_floor]; Output four dictionaries: region_avg_dict, region_median_dict, type_floor_avg_dict, and type_floor_median_dict. region_avg_dict represents the average price distribution of different locations; region_median_dict represents the median price distribution of different locations; type_floor_avg_dict represents the average price distribution of different combinations of location, house type, and floor; type_floor_median_dict represents the median price distribution of different combinations of location, house type, and floor.

[0025] Furthermore, construct a real - estate transaction probability prediction model through a logistic regression algorithm, including: Extract the feature vector of each real - estate data, and the feature vector includes location, house type, and floor; Extract the timestamp of the real - estate data, and obtain the predicted prices corresponding to the corresponding time granularity from the daily house price change curve, weekly house price change curve, and monthly house price change curve as the price dimension of the feature vector; Extract the location of the real - estate data, and obtain the corresponding average price and median price from the average price distribution of different locations and the median price distribution of different locations as the location price dimension of the feature vector; Extract the location, house type, and floor of the real - estate data, and extract the corresponding average price and median price from the average price distribution of different combinations of location and house type - floor, and the median price distribution of different combinations of location and house type - floor as the house type - floor price dimension of the feature vector; Construct a multi - dimensional feature vector based on the price dimension of the feature vector, the location price dimension of the feature vector, and the house type - floor price dimension of the feature vector, and mark positive and negative class labels according to whether the real - estate is transacted to form a training sample set; For each sample in the training sample set, extract its multi - dimensional feature vector x=(x1,x2,......,x n ), including attribute features, price features, location price features, and house type - floor price features; Normalize the multi - dimensional feature vector, scale each dimension feature value to the interval [0, 1] to eliminate the influence of the dimension of different features; Introduce a weight vector w=(w1,w2,......,w n ) and a bias term b, where w i$w_i$ represents the weight coefficient of the $i$-th feature, and $b$ represents the bias term, which is used to control the position of the decision boundary; the probability that the multi-dimensional feature vector $x$ belongs to the positive class (transaction) is calculated through the weighted Softmax function: where $w$ T $x$ represents the inner product of the weight vector $w$ and the feature vector $x$, $b$ represents the bias term, and $\exp$ represents the natural exponential function; an L2 regularization term is introduced to control the complexity of the model and avoid overfitting. According to the true label $y$ of the training samples ($0$ represents no transaction, and $1$ represents transaction), the regularized Softmax regression loss function is defined:

[0026] where $\sum$ represents the sum over all samples in the training sample set, $\lambda$ represents the regularization coefficient, and $\|w\|$ 2 represents the L2 norm of the weight vector $w$; the stochastic gradient descent algorithm is used to optimize and solve the loss function $J(w, b)$: where $\alpha$ represents the learning rate, represents the gradient of the loss function with respect to the weight vector $w$, represents the gradient of the loss function with respect to the bias term $b$; the stochastic gradient descent is repeated iteratively using Mini-batch until the loss function converges or reaches the preset number of iterations, and the optimized Softmax regression model parameters $w$ and $b$ are obtained; the trained Softmax regression model is used to predict the transaction probability of new real estate data:

[0027] The feature vector $x$ of the input real estate data is input, $P(y = 1|x; w, b)$ is calculated, and the predicted transaction probability value is output.

[0028] In this application, the weight vector $w$ and the bias term $b$ are introduced into the Softmax regression model, which enhances the expression ability of the model. The complexity of the model is controlled by the L2 regularization term, effectively avoiding the overfitting problem. The stochastic gradient descent algorithm is used to accelerate the model training, improving the training efficiency and generalization ability. At the same time, the introduction of Mini-batch can balance the computational efficiency and convergence speed, enabling the model to be trained quickly on large-scale data sets.

[0029] Another aspect of this application provides a real estate asset operation and management system based on big data analysis, which is used to execute a real estate asset operation and management method based on big data analysis.

[0030] Compared with the prior art, the advantages of this application are:

[0031] Smoothing the time series using the first-order and second-order exponential smoothing algorithms can filter out random noise while adaptively capturing the local and long-term trends of price changes. The first-order exponential smoothing emphasizes the impact of recent price changes by weighted averaging historical data, depicting the short-term trend of prices; the second-order exponential smoothing further smooths the sequence based on the first-order exponential smoothing, emphasizing the cumulative effect of historical data and depicting the long-term development trend of prices. Combining the first-order and second-order exponential smoothing can model price changes on different time scales, taking into account both the local volatility characteristics and the global change trends of prices, thus improving the dynamic adaptability and stability of price prediction.

[0032] Using the DBSCAN clustering algorithm to perform spatial clustering on real estate data can automatically discover real estate location clusters, revealing the spatial heterogeneity of real estate prices. During the clustering process, by plotting the KNN distance curve and analyzing its change characteristics, the key parameter of density clustering, i.e., the density radius ε, is adaptively determined. The KNN distance reflects the density characteristics of the local neighborhood of sample points, and the inflection point of the KNN distance curve corresponds to the density mutation, representing the inherent scale of the clustering structure. By selecting the first obvious inflection point of the KNN distance curve as ε, it can automatically adapt to the density differences of real estate distributions in different locations, adopting different density thresholds in dense and sparse areas, overcoming the blindness of parameter selection, and improving the rationality and robustness of the clustering results. At the same time, by analyzing the clustering structure of real estate locations, the market segmentation of real estate can be objectively divided, which has important commercial value in the real estate field. On the basis of obtaining the location clusters, further subdivide the sub-clusters within each location cluster according to house types and floors. By comparing the price distribution differences of different house type and floor sub-clusters, the influence of factors such as house types and floors on real estate prices can be quantified, the influencing factors of real estate prices can be mined, and more refined real estate pricing can be achieved, providing data support for formulating differentiated real estate marketing strategies.

[0033] Based on spatio-temporal multi-granularity analysis, calculate the mean and median distributions of prices for different location clusters and house type - floor sub-clusters, quantifying the regional and type differences in real estate prices. Compared with a single mean, the median is more robust to outliers, and the combination of the two indicators can more comprehensively depict the price distribution. The quantitative description of the price distribution provides a direct reference for real estate valuation and pricing decisions.

[0034] By integrating multi-dimensional information such as price changes in the time dimension, lot clustering in the space dimension, and price distribution in the house type - floor dimension, the feature vectors of each real estate are extracted, and a comprehensive and fine-grained real estate feature representation is constructed. In the feature vectors, price prediction values with different time granularities such as daily, weekly, and monthly are incorporated, reflecting the dynamic change trend of real estate prices; the results of lot clustering and the corresponding average and median prices are introduced to depict the overall price level and market positioning of the area where the real estate is located; at the same time, by combining the average and median prices of the house type - floor sub-clusters, the real estate features are further refined, reflecting the price impact of key factors such as house type and floor. The integration of such multi-dimensional heterogeneous features can comprehensively represent the value influencing factors of real estate from different perspectives of macro and micro, dynamic and static, providing all-round feature information. Brief Description of the Drawings

[0035] This application will be further described in the form of exemplary embodiments, which will be described in detail through the drawings. These embodiments are not restrictive. In these embodiments, the same numbers represent the same structures, where:

[0036] Figure 1 is an exemplary flowchart of a real estate asset operation management method based on big data analysis shown in some embodiments of this application;

[0037] Figure 2 is an exemplary flowchart of generating prices with different time granularities shown in some embodiments of this application;

[0038] Figure 3 is an exemplary flowchart of generating price change curves of real estate at different times shown in some embodiments of this application;

[0039] Figure 4 is an exemplary flowchart of generating real estate lot clusters shown in some embodiments of this application;

[0040] Figure 5 is an exemplary flowchart of generating house type - floor sub-clusters shown in some embodiments of this application. Detailed Description of the Embodiments

[0041] The methods and systems provided in the embodiments of this application will be described in detail below with reference to the drawings.

[0042] As Figure 1As shown in the figure, real estate data is collected, which includes housing prices, locations, housing types, and floors. According to the collected real estate data, a time series analysis algorithm is used to aggregate housing prices by time dimension, generating prices at daily, weekly, and monthly granularities. The exponential smoothing algorithm is respectively used to smooth the daily granularity price, weekly granularity price, and monthly granularity price, obtaining the daily housing price change curve, weekly housing price change curve, and monthly housing price change curve. According to the location data in the collected real estate data, the DBSCAN algorithm is used to cluster the real estate, obtaining real estate clusters in different locations. Within each real estate cluster, sub-clusters are respectively divided by housing type and floor. The average price and median price of each real estate cluster and sub-cluster are calculated, obtaining the price mean distribution and price median distribution for different combinations of location, housing type, and floor. According to the daily housing price change curve, weekly housing price change curve, and monthly housing price change curve, as well as the price mean distribution and price median distribution, a real estate transaction probability prediction model is constructed through a logistic regression algorithm. The real estate transaction probability prediction model takes real estate data as input and the transaction probability of real estate as output, and real estate asset operation management is carried out according to the predicted transaction probability.

[0043] Specifically, a crawler program is used to obtain real estate data containing housing price, location, housing type, and floor information from channels such as real estate websites and government public data platforms. The collected data is cleaned to remove noise data such as missing values and outliers to ensure data quality. The cleaned data is stored in a relational database or a distributed file system for subsequent processing and analysis.

[0044] As Figure 2 shown in the figure, according to the collected real estate data, a time series analysis algorithm is used to aggregate housing prices by time dimension, generating prices at daily, weekly, and monthly granularities. The real estate data is read from the database or file system, and the timestamp field of each piece of data is extracted. The timestamp field is converted into a datetime object and used as the sorting key to sort the real estate data in ascending order. The sorted data is arranged in chronological order of housing prices. Aggregate the daily granularity price. Traverse the chronological housing price sequence. For each housing price data, first, we traverse the chronological housing price sequence. For each housing price data, extract its timestamp field. The timestamp is usually stored in the form of a string, such as "2023-05-18 09:30:00". Next, we use a regular expression to extract the year, month, and day information from the timestamp. A regular expression can match and extract the target content in a string according to a specific pattern. For example, using the regular expression "(\d{4})-(\d{2})-(\d{2})" can match a date string in the form of "YYYY-MM-DD" and extract the three parts of the year, month, and day.

[0045] Then, we concatenate the extracted year, month, and day information into a date string in the format of "YYYY-MM-DD", which serves as the key for the hash table. A hash table is an efficient key-value storage structure that allows for quick lookup and access to the corresponding value through the key. Next, we create a hash table with the date string as the key and the list of housing prices as the value. We iterate through the housing price sequence and add each housing price data to the housing price list corresponding to the date. Through this step, we group the housing price data by date, with each date corresponding to a housing price list that stores all the housing price records for that date.

[0046] Finally, we iterate through each key-value pair in the hash table and calculate the arithmetic mean of the housing prices in the value list to obtain the daily-granularity price for that date. The arithmetic mean can be calculated by summing all the prices in the housing price list and then dividing by the list length. Through this step, we aggregate the housing price records for each date into a daily-granularity price, which reflects the average price level for that date.

[0047] Through daily-granularity price aggregation, we can analyze the intraday fluctuation patterns of real estate prices. For example, we can plot a time series graph of the daily-granularity prices to observe the daily price change trends and identify sudden abnormal price changes. This is of great significance for daily price monitoring and market risk prevention. In addition, daily-granularity price aggregation can also provide data support for other analyses. For example, we can calculate statistical metrics of the daily-granularity prices, such as mean, variance, volatility, etc., to evaluate the market stability and risk level. We can also perform time series modeling on the daily-granularity prices to predict future price trends and provide reference for investment decisions.

[0048] To aggregate weekly-granularity prices, iterate through the housing price sequence in chronological order. For each housing price data, extract its timestamp field. Use regular expressions to extract the year, month, and day information from the timestamp. Calculate which week of the year the date belongs to, and use the week number as the key for the hash table. Create a hash table with the week number as the key and the list of housing prices as the value. Iterate through the housing price sequence and add each housing price data to the housing price list corresponding to the week number. Iterate through each key-value pair in the hash table and calculate the arithmetic mean of the housing prices in the value list to obtain the weekly-granularity price for that week. Through weekly-granularity price aggregation, we can analyze the weekly change trends of real estate prices, understand periodic characteristics such as weekend effects and holiday effects, optimize marketing strategies, and reasonably arrange the sales rhythm.

[0049] Aggregate the housing prices at monthly granularity. Traverse the housing price sequence in chronological order. For each housing price data, extract its timestamp field. Use regular expressions to extract the year and month information from the timestamp, and concatenate them into a month string in the form of "YYYY-MM" as the key of the hash table. Create a hash table with the month string as the key and the list of housing prices as the value. Traverse the housing price sequence and add each housing price data to the housing price list corresponding to the month. Traverse each key-value pair in the hash table, calculate the arithmetic mean of the housing prices in the value list to obtain the monthly granularity price for that month. Through the aggregation of monthly granularity prices, the monthly trend of real estate prices can be analyzed, seasonal patterns can be mined, market turning points can be grasped, and data basis can be provided for major decisions such as real estate development planning, pricing strategy formulation, and marketing plan selection.

[0050] In this application, by performing multi-time scale aggregations of daily granularity, weekly granularity, and monthly granularity on the housing price sequence, the time variation law of real estate prices can be comprehensively characterized, and the price fluctuation characteristics can be revealed from different time dimensions. Using technical means such as regular expressions and hash tables, the time granularity conversion and aggregation calculation of housing price data can be efficiently and flexibly implemented. Visualize and analyze the results after granularity aggregation, draw corresponding price trend charts, volatility charts, etc., intuitively display the price dynamics of the real estate market, and provide macro decision-making references for managers. The price analysis based on the time dimension is an important data basis for the operation and management of real estate assets, which can significantly improve the scientificity, accuracy, and forward-looking of decisions, and promote the preservation and appreciation of real estate assets.

[0051] As Figure 3 shown, the exponential smoothing algorithm is used to smooth the daily granularity price, weekly granularity price, and monthly granularity price respectively to obtain the daily housing price change curve, weekly housing price change curve, and monthly housing price change curve, including: First, the key parameter of the exponential smoothing algorithm - the smoothing coefficient a needs to be set. The value of a ranges from 0 to 1, which reflects the relative weights of the model for recent data and historical data. The larger the value of a, the higher the weight given to recent data, and the higher the fitting degree of the smoothed curve to the actual value, but the stronger the random volatility; the smaller the value of a, the higher the weight given to historical data, and the more stable the smoothed curve, but there will be a certain lag in the response to price changes. In practical applications, the value of a can be flexibly set according to the volatility characteristics of real estate prices and decision-making requirements. For example, for the daily granularity price, due to strong volatility, a smaller value of a such as 0.2 can be selected, while for the monthly granularity price, due to weak volatility, a larger value of a such as 0.6 can be selected to balance trend capture and stability.

[0052] Secondly, taking the first element of the daily-granularity price, weekly-granularity price, or monthly-granularity price series as the initial smoothing value, using the smoothing value at the previous moment and the actual value at the current moment, recursively calculate the first-order exponential smoothing value at each moment through weighted average, so as to obtain the first-order exponential smoothing series. Let the actual housing price at the t-th moment be Y t , the first-order smoothing value be S t , then the calculation formula for S t is: S t = a×Y t + (1 - a)×S t-1 , where S t-1 represents the first-order smoothing value at the (t - 1)-th moment. Through first-order exponential smoothing, short-term random fluctuations in the housing price series can be eliminated to a certain extent, and the change trend of the price can be initially shown.

[0053] Next, on the basis of the first-order exponential smoothing, further smooth the first-order exponential smoothing series to obtain the second-order exponential smoothing series. Taking the first element of the first-order exponential smoothing series as the initial value, using the second-order smoothing value at the previous moment and the first-order smoothing value at the current moment, recursively calculate the second-order exponential smoothing value at each moment. Let the first-order smoothing value at the t-th moment be S t , the second-order smoothing value be S t ', then the calculation formula for S t ' is: S t ' = a×S t + (1 - a)S t-1 ', where S t-1 ' represents the second-order smoothing value at the (t - 1)-th moment. Second-order exponential smoothing is equivalent to further removing fluctuations on the basis of first-order exponential smoothing, making the long-term change trend of housing prices smoother and more obvious.

[0054] Finally, in step S34, according to the exponential smoothing model, combining the first-order exponential smoothing series and the second-order exponential smoothing series, smooth prediction of housing prices for a period of time in the future is carried out. Let the first-order smoothing value at the t-th moment be S t , the second-order smoothing value be S t ', then the smoothing prediction value F t+k at the future (t + k)-th moment is calculated as follows:

[0055] Connecting the smoothing prediction values at each moment, the smoothed daily housing price change curve, weekly housing price change curve, or monthly housing price change curve is obtained. Through exponential smoothing prediction, based on the historical change law of housing prices, the price trend for a period of time in the future can be extrapolated, assisting in judging the market trend and optimizing resource allocation and marketing decisions.

[0056] The exponential smoothing algorithm is used to analyze the real estate price series with multiple time granularities, which can reveal the change patterns of housing prices from different scales. First-order exponential smoothing focuses on short-term trends, and second-order exponential smoothing pays attention to long-term trends. The combination of the two can comprehensively depict the operation situation of housing prices. Based on the prediction of exponential smoothing, the future trend of housing prices can be judged prospectively, and potential opportunities and risks can be identified.

[0057] As Figure 4 shown, according to the lot data in the collected real estate data, the DBSCAN algorithm is used to cluster real estate, aiming to discover the geographical space structure in the real estate market, identify lot clusters with different characteristics, and provide regionalized and refined decision-making basis for asset operation and management.

[0058] First, extract the lot information from the collected real estate data to construct a lot feature vector. The lot feature vector can include multi-dimensional attributes such as geographical coordinates (longitude, latitude), regional labels (such as business districts, school districts, etc.), and traffic convenience (such as the distance to the subway station) to comprehensively depict the spatial characteristics of the lot where the real estate is located. Then, preprocess the lot feature vector, such as normalization, standardization, etc., to eliminate the influence of different attribute dimensions and make the features of each dimension comparable. The preprocessed feature vector lays the data foundation for DBSCAN clustering.

[0059] The density threshold ε and the minimum number of points threshold MinPts. These two parameters directly affect the clustering effect, so they need to be adaptively determined according to the characteristics of the data. To adaptively determine the density threshold ε of the DBSCAN algorithm, we first need to calculate the Euclidean distance between each lot feature vector and other vectors. Suppose we have N lot feature vectors, and each vector contains M attribute dimensions, then we can construct an N×N distance matrix D. The element D ij of the distance matrix D represents the Euclidean distance between the i-th lot feature vector and the j-th lot feature vector, and its calculation formula is:

[0060] where, X i and X j represent the i-th and j-th lot feature vectors respectively. The distance matrix D is a symmetric matrix, and the elements on the main diagonal are 0, which reflects the spatial proximity relationship between lot data points.

[0061] Next, sort the distance values of each row (i.e., each lot point) of the distance matrix D in ascending order to obtain the K-nearest neighbor distance list of each lot point. For example, for the i-th lot point, its K-nearest neighbor distance list is: KNN i =[D i1 ,D i2 ,......,D iK , where, Di1 , D i2 ,......, D iK is the distance value between the i-th lot point and other lot points, arranged in ascending order, representing the distance between the i-th lot point and its K nearest points. The K-nearest neighbor distance list reflects the density characteristics of the local area of each lot point. The smaller the distance value, the higher the local density.

[0062] Then, we select the K-th nearest neighbor distance of each lot point as its KNN distance to form a set of KNN distances: KNNDist = [KNN1[K], KNN2[K],....., KNN N [K]], where KNN i [K] represents the K-th nearest neighbor distance of the i-th lot point. Sort the set of KNN distances in ascending order to obtain an ordered sequence of KNN distances: SortedKNNDist = [KNNDist[1], KNNDist[2],......., KNNDist[N]]. By plotting the KNN distance curve, with the serial number of the lot point (from 1 to N) on the horizontal axis and the corresponding KNN distance value on the vertical axis, the density distribution characteristics of the data can be visually displayed. The steepness of the curve reflects the rate of change of density.

[0063] Next, we calculate the slope change of each point on the KNN distance curve. For the i-th point (i > 1), its slope can be calculated by the following formula: Slope i = SortedKNNDist[i] - SortedKNNDist[i - 1]. The slope reflects the steepness of the curve. A sudden change in the slope often means a significant change in density. We can observe the slope change to find the first obvious inflection point on the curve, that is, the position where the slope suddenly increases. Assume that the j-th point is the first point that satisfies the following condition: Slope j > ThresholdSlope, where ThresholdSlope is a preset slope threshold used to determine whether the slope change is significant. We take the KNN distance value corresponding to the j-th point as the density threshold ε: ε = SortedKNNDist[j]. This adaptive parameter selection method fully considers the characteristics of the data distribution, can automatically determine a suitable density threshold, and avoids the subjectivity and blindness of manual setting.

[0064] Finally, based on the total number N of the lot feature vectors, we set the minimum inclusion point threshold MinPts, which is usually set to lnN: MinPts = ln(N). MinPts controls the minimum number of samples in each cluster and ensures the reliability of the clustering result. The value of MinPts needs to balance the clustering granularity and noise tolerance. A too large value may lead to under-clustering, while a too small value may lead to over-clustering.

[0065] After determining the two key parameters ε and MinPts, we use the DBSCAN algorithm to perform density clustering on the lot feature vectors. The DBSCAN algorithm is based on the concepts of density reachability and density connection. It classifies points in the feature space with a density greater than or equal to MinPts and mutually density-reachable into one category, forming a lot cluster. The points within these lot clusters have a high density and are spatially adjacent, representing hotspots in the real estate market. Points with a density less than MinPts are marked as noise points or outliers, which may represent some unique or atypical lots. Through DBSCAN clustering, we can discover the lot agglomeration patterns in the real estate market and reveal the spatial difference characteristics of different regions.

[0066] Finally, we interpret and apply the results of DBSCAN clustering. By visually displaying the spatial distribution of each lot cluster and analyzing the attribute characteristics of each cluster (such as location, transportation, housing price, etc.), we can comprehensively characterize the real estate market characteristics of different regions. We can identify value depressions and growth poles, explore the potential for regional development, and provide a reference for real estate investment decisions. Based on the lot clustering results, we can achieve regional asset evaluation, pricing strategy formulation, marketing plan optimization, etc., improve the accuracy and differentiation level of asset operation, and maximize asset value. At the same time, the clustering results can also provide data support for macro-decisions such as real estate development planning and regional balanced layout, optimize resource allocation, and promote coordinated regional development.

[0067] As Figure 5 shown, within each real estate cluster, sub-clusters are divided according to house type and floor, including: First, based on the real estate data within each real estate lot cluster, we extract the corresponding house type attributes and floor attributes. The house type attributes include the number of rooms (Rooms) and the number of bathrooms (Bathrooms), which reflect the housing type characteristics of the real estate. The floor attributes include the floor where it is located (Floor) and the total number of floors (Total Floors), which reflect the vertical position characteristics of the real estate. These attributes can be obtained from the structured description data of the real estate, for example:

[0068] Real Estate ID Number of Rooms Number of Bathrooms Floor Where Located Total Number of Floors 1 2 1 3 6 2 3 2 5 10

[0069] Next, we convert the housing type attributes into numerical housing type feature vectors. For each real estate property, its housing type feature vector can be represented as (Rooms, Bathrooms). Based on the housing type feature vectors, we calculate the Euclidean distance matrix of housing types within the real estate location cluster. Suppose there are N real estate properties within the cluster, then the dimension of the distance matrix is N×N, where the element in the i-th row and j-th column represents the Euclidean distance of housing types between the i-th real estate property and the j-th real estate property, and the calculation formula is:

[0070] Then, we use the hierarchical clustering algorithm to cluster the Euclidean distance matrix of housing types. Hierarchical clustering is a bottom-up clustering method that generates a dendrogram by iteratively merging the most similar samples or clusters. We can select an appropriate cut-off point according to the structure of the dendrogram and the merging distance to divide the real estate location cluster into K1 housing type sub-clusters. The value of K1 can be determined according to factors such as the shape of the dendrogram and business requirements.

[0071] We convert the floor attributes into numerical floor feature vectors according to the K1 housing type sub-clusters. For each real estate property, its floor feature vector can be represented as (Floor / Total Floors, Total Floors). The first element represents the relative position of the floor where the real estate property is located, with a value range of (0, 1], reflecting the relative height of the real estate property in the vertical direction; the second element represents the total number of floors, reflecting the scale of the building where the real estate property is located. Based on the floor feature vectors, we calculate the Euclidean distance matrix of floors within each housing type sub-cluster. Suppose there are M real estate properties within the sub-cluster, then the dimension of the distance matrix is M×M, where the element in the i-th row and j-th column represents the Euclidean distance of floors between the i-th real estate property and the j-th real estate property, and the calculation formula is: Next, we use the K-means clustering algorithm to cluster the Euclidean distance matrix of floors for each housing type sub-cluster. K-means clustering is a centroid-based clustering method that divides the data into K clusters by iteratively updating the cluster centroids and reassigning samples. We can select an appropriate K value (K2) according to business requirements and data distribution to perform floor clustering on each housing type sub-cluster to obtain K2 floor sub-clusters.

[0072] Finally, based on the K1 housing type sub-clusters and K2 floor sub-clusters, we divide each real estate plot cluster into K1×K2 housing type - floor sub-clusters. Each sub-cluster represents a real estate niche market with a specific combination of housing type and floor. Through this multi-level clustering division, we can more precisely characterize the structure and characteristics of the real estate market, identify the niche demands and price differences of different housing types and floors, and provide more targeted support for decisions such as pricing, marketing, and investment. By further sub-clustering according to housing type and floor within the real estate plot cluster, multi-granularity and multi-dimensional segmentation of the real estate market are achieved. This segmentation method fully utilizes the structured attribute information of real estate, combines hierarchical clustering and K-means clustering algorithms, and automatically discovers the potential grouping patterns of housing types and floors. The generated housing type - floor sub-clusters provide a data basis for refined real estate management and differential strategy formulation, and help improve the accuracy and efficiency of real estate asset operation.

[0073] We calculate the average price and median price of each real estate plot cluster and housing type - floor sub-cluster to obtain the price distribution characteristics of different plot, housing type, and floor combinations, providing a reference for real estate price analysis and decision-making. First, for each real estate plot cluster, we extract the transaction prices of all real estates within the cluster to form a price list price_list. By traversing each real estate within the cluster and adding its transaction price to price_list, we obtain the complete price data of the plot cluster. Then, we calculate the arithmetic mean of price_list, that is, sum all the prices in price_list and divide by the number of prices, to obtain the average price cluster_avg_price of the real estate plot cluster. The average price reflects the average level of real estate prices in this plot. Next, we sort price_list in ascending order and select the element in the middle position as the median price cluster_median_price of the real estate plot cluster. If the length of price_list is odd, the median price is the price in the middle position; if the length of price_list is even, the median price is the average of the two middle prices. The median price reflects the central position of real estate prices in this plot and is more robust than the average price, being less affected by extreme values.

[0074] Similarly, for each housing type - floor sub-cluster, we extract the transaction prices of all real estates within the sub-cluster to form a price list sub_price_list. Calculate the arithmetic mean of sub_price_list to obtain the average price sub_cluster_avg_price of the housing type - floor sub-cluster; sort sub_price_list in ascending order and select the element in the middle position as the median price sub_cluster_median_price of the housing type - floor sub-cluster.

[0075] Next, we initialize four dictionaries: the average price dictionary of regions region_avg_dict, the median price dictionary of regions region_median_dict, the average price dictionary of housing types and floors type_floor_avg_dict, and the median price dictionary of housing types and floors type_floor_median_dict, to store price distribution information at different granularities.

[0076] Traverse each real estate region cluster, extract the region information region of the cluster, use region as the key, and use the average price cluster_avg_price of the cluster as the value, and add it to region_avg_dict; use the median price cluster_median_price of the cluster as the value and add it to region_median_dict. In this way, region_avg_dict and region_median_dict respectively represent the average price distribution and median price distribution of different regions.

[0077] Traverse each housing type - floor sub - cluster, extract the region information region, housing type information type, and floor information floor of the sub - cluster, concatenate them into a string region_type_floor as the key, and use the average price

[0078] sub_cluster_avg_price of the sub - cluster as the value, and add it to type_floor_avg_dict; use the median price sub_cluster_median_price of the sub - cluster as the value and add it to type_floor_median_dict. In this way, type_floor_avg_dict and type_floor_median_dict respectively represent the average price distribution and median price distribution of different combinations of regions, housing types, and floors.

[0079] Finally, we output four dictionaries: region_avg_dict, region_median_dict, type_floor_avg_dict, and type_floor_median_dict, which characterize the distribution features of real estate prices from different granularities and dimensions. region_avg_dict shows the average price distribution of different locations, and region_median_dict shows the median price distribution of different locations, reflecting the impact of location factors on housing prices; type_floor_avg_dict shows the average price distribution of different combinations of location, housing type, and floor, and type_floor_median_dict shows the median price distribution of different combinations of location, housing type, and floor, reflecting the further impact of housing type and floor factors on housing prices based on location factors.

[0080] Through the analysis of the average and median prices of real estate location clusters and housing type - floor sub - clusters, we can comprehensively characterize the price distribution pattern of the real estate market, identify price characteristics under different factor combinations, and provide data support for applications such as real estate pricing, investment decision - making, and housing source recommendation. The average price reflects the average level of prices, and the median price reflects the central position of prices. The combination of the two can more robustly describe the price distribution. Organizing price distribution information in the form of a dictionary facilitates flexible querying and comparison according to dimensions such as location, housing type, and floor, providing a data basis for the refined management of real estate.

[0081] We construct a real estate transaction probability prediction model through a logistic regression algorithm based on the time - series characteristics (daily housing price change curve, weekly housing price change curve, and monthly housing price change curve) and spatial distribution characteristics (average price distribution and median price distribution) of the real estate market. This model takes real estate data as input and the transaction probability of real estate as output, realizing the prediction of the possibility of real estate transactions.

[0082] First, we extract the feature vectors of each real estate data. The feature vectors contain basic attribute information such as location, housing type, and floor, which are used to characterize the static features of real estate. Then, we extract the timestamps of real estate data and obtain the predicted prices corresponding to the corresponding time granularities from the daily housing price change curve, weekly housing price change curve, and monthly housing price change curve according to the timestamps, as the price dimension of the feature vectors. This step introduces the time - dynamic characteristics of the real estate market, reflecting the impact of price trends at different time scales on the transaction probability.

[0083] Then, we extract the lot information of the real estate data. From the average price distribution and median price distribution of different lots, we obtain the corresponding average price and median price respectively, which serve as the lot price dimension of the feature vector. This step introduces the spatial difference characteristics of the real estate market, reflecting the impact of price levels in different lots on the transaction probability.

[0084] We further extract the housing type and floor information of the real estate data. From the average price distribution and median price distribution of different lots and housing type - floor combinations, we extract the corresponding average price and median price respectively, which serve as the housing type - floor price dimension of the feature vector. This step introduces more fine - grained spatial difference characteristics, reflecting the impact of housing type and floor factors on the transaction probability based on the lot factor.

[0085] We construct a multi - dimensional feature vector based on the price dimension, lot price dimension, and housing type - floor price dimension of the feature vector, and mark positive and negative class labels according to whether the real estate is transacted, thus forming a training sample set. The multi - dimensional feature vector synthesizes the static attributes, time - dynamic characteristics, and spatial difference characteristics of the real estate, providing rich information for the prediction model.

[0086] Next, we perform logistic regression modeling on the training sample set: For each sample, we extract its multi - dimensional feature vector x=(x1,x2,......,x n ), including attribute features, price features, lot price features, and housing type - floor price features. We normalize the multi - dimensional feature vector, scaling each dimensional feature value to the interval [0, 1] to eliminate the influence of the dimension of different features. We introduce a weight vector w=(w1,w2,......,w n ) and a bias term b, where w i represents the weight coefficient of the i - th feature, and b represents the bias term, which is used to control the position of the decision boundary.

[0087] We calculate the probability that the multi - dimensional feature vector x belongs to the positive class (transaction) through the weighted Softmax function: where, w T x represents the inner product of the weight vector w and the feature vector x, b represents the bias term, and exp represents the natural exponential function. We introduce an L2 regularization term to control the complexity of the model and avoid overfitting. According to the true label y of the training sample (0 represents no transaction, 1 represents transaction), we define the regularized Softmax regression loss function:

[0088] where, Σ represents the sum over all samples in the training sample set, λ represents the regularization coefficient, and ||w||^2 represents the L2 norm of the weight vector w.

[0089] Optimize and solve the loss function J(w, b) using the stochastic gradient descent algorithm:

[0090] where α represents the learning rate, represents the gradient of the loss function with respect to the weight vector w, represents the gradient of the loss function with respect to the bias term b.

[0091] Repeatedly iterate and use Mini - batch for stochastic gradient descent until the loss function converges or reaches the preset number of iterations to obtain the optimized Softmax regression model parameters w and b. Finally, we use the trained Softmax regression model to predict the transaction probability for new real estate data: input the feature vector x of the real estate data, calculate P(y = 1|x; w, b), and output the predicted transaction probability value.

[0092] The real estate transaction probability prediction model constructed through the logistic regression algorithm fully utilizes the spatio - temporal characteristics and price distribution information of the real estate market, and learns the influence rules of different factors on the transaction probability. This model can predict the possibility of a transaction based on the property characteristics, time nodes, location segments, and house type - floor combinations of the real estate, providing quantitative references for decision - making such as formulating real estate marketing strategies and optimizing resource allocation. At the same time, by introducing the L2 regularization term, while fitting the training data, the model also controls the complexity and improves the generalization ability, making it better applicable to new real estate data.

[0093] This application uses the logistic regression algorithm, integrates multi - dimensional characteristics of the real estate market, constructs a transaction probability prediction model, and provides data - driven decision - making support for precise real estate marketing and asset optimization.

Claims

1. A real estate asset operation management method based on big data analysis, characterized in that, Including: Collecting real estate data, where the real estate data includes house prices, locations, house types, and floors; According to the collected real estate data, using the time series analysis algorithm, aggregating house prices by time dimension to generate prices at daily, weekly, and monthly granularities; Respectively using the exponential smoothing algorithm to smooth the daily-granularity price, weekly-granularity price, and monthly-granularity price to obtain the daily house price change curve, weekly house price change curve, and monthly house price change curve; According to the location data in the collected real estate data, using the DBSCAN algorithm to cluster the real estate to obtain real estate clusters in different locations; Within each real estate cluster, sub-clusters are divided according to house type and floor respectively; Calculating the average price and median price of each real estate cluster and sub-cluster to obtain the price mean distribution and price median distribution of different combinations of location, house type, and floor; According to the daily house price change curve, weekly house price change curve, and monthly house price change curve, as well as the price mean distribution and price median distribution, constructing a real estate transaction probability prediction model through the logistic regression algorithm. The real estate transaction probability prediction model takes real estate data as input and the transaction probability of real estate as output, and conducts real estate asset operation management according to the predicted transaction probability.

2. The real estate asset operation management method based on big data analysis according to claim 1, characterized in that: Generating prices at daily, weekly, and monthly granularities includes: Extracting the timestamp of each real estate data, sorting the real estate data in ascending order according to the timestamp to form a house price sequence in time order; Traversing each house price data in the house price sequence in time order to obtain the timestamp of the corresponding house price data; Using regular expressions to extract year, month, and day data from the timestamp and concatenating them into a date string; Using the hash table data structure, with the date string as the key, adding the house prices corresponding to the date to the value list of the hash table; Traversing each key-value pair in the hash table, calculating the arithmetic mean of the house price list in the value list to obtain the daily-granularity price of the corresponding date; Traversing each house price data in the house price sequence in time order to obtain the timestamp of the corresponding house price data; Using regular expressions to extract year, month, and day data from the timestamp, calculating the corresponding week number according to the date as the key for weekly-granularity grouping; Using the hash table data structure, with the weekly-granularity grouping as the key, adding the house prices within the corresponding week to the value list of the hash table; Traversing each key-value pair in the hash table, calculating the arithmetic mean of the house price list in the value list to obtain the weekly-granularity price of the corresponding week; Traversing each house price data in the house price sequence in time order to obtain the timestamp of the corresponding house price data; Using regular expressions to extract year and month data from the timestamp and concatenating them into a month string; Using the hash table data structure, with the month string as the key, adding the house prices corresponding to the month to the value list of the hash table; Traversing each key-value pair in the hash table, calculating the arithmetic mean of the house price list in the value list to obtain the monthly-granularity price of the corresponding month.

3. The real estate asset operation management method based on big data analysis according to claim 2, characterized in that: Obtain the daily house price change curve, weekly house price change curve and monthly house price change curve, including: Set the smoothing coefficient a; Use the first element of the daily-granularity price, weekly-granularity price or monthly-granularity price as the initial smoothing value, and use the previous moment's smoothing value and the current actual value to recursively calculate the first-order exponential smoothing value at each moment to obtain a first-order exponential smoothing sequence; Use the first element of the first-order exponential smoothing sequence as the initial smoothing value, and use the previous moment's second-order smoothing value and the current first-order smoothing value to recursively calculate the second-order exponential smoothing value at each moment to obtain a second-order exponential smoothing sequence; According to the first-order exponential smoothing sequence and the second-order exponential smoothing sequence, calculate the smoothed predicted values of the daily-granularity price, weekly-granularity price or monthly-granularity price in a future period of time to obtain the daily house price change curve, weekly house price change curve or monthly house price change curve.

4. The real estate asset operation management method based on big data analysis according to claim 2, characterized in that: Use the DBSCAN algorithm to cluster real estates to obtain real estate clusters in different locations, including: Obtain real estate data, extract the location data in the real estate data to form a location feature vector; Preprocess the location feature vector; Set the density threshold parameter ε; and set the minimum number of points included threshold MinPts according to the location data; Select any unvisited location feature point p, and calculate the number of points within the ε-neighborhood of p; If the number of points within the ε-neighborhood of p is greater than or equal to MinPts, mark p as a core point, determine a new cluster starting from p, and classify all unclassified points within the ε-neighborhood of p into the corresponding cluster; if the number of points within the ε-neighborhood of p is less than MinPts, mark p as a noise point; For the location feature points marked as noise points, if there is any core point within the corresponding ε-neighborhood, classify the core point into the corresponding cluster, otherwise mark the core point as an outlier; Output all clusters of the DBSCAN algorithm in the location feature space, each cluster corresponds to a real estate location cluster, and the real estates within the location cluster are geographically aggregated.

5. The real estate asset operation management method based on big data analysis according to claim 4, characterized in that: Set the density threshold parameter ε, and set the minimum number of points included threshold MinPts according to the location data, including: Calculate the Euclidean distance between each location feature vector and other location feature vectors, and construct a distance matrix of location feature points; Sort each row of the distance matrix in ascending order to obtain the K-nearest neighbor distance list of each location feature point; Select the Kth nearest neighbor distance of each location feature point as the KNN distance to form a KNN distance set of location feature points; Sort the KNN distance set of location feature points in ascending order, draw a KNN distance curve graph, with the horizontal axis being the serial number of the location feature point and the vertical axis being the corresponding KNN distance value; On the KNN distance curve graph, calculate the slope value of each point, obtain the first inflection point with a slope value greater than the preset threshold, and use the KNN distance value corresponding to the first inflection point as the density threshold parameter ε; Count the total number N of the lot feature vectors, and set the minimum inclusion point threshold MinPts according to N.

6. The method for operating and managing real estate assets based on big data analysis according to claim 4, wherein: Within each real estate cluster, sub-clusters are respectively divided according to the house type and floor, including: According to the real estate data within each real estate lot cluster, extract the corresponding house type attributes and floor attributes; wherein, the house type attributes include the number of rooms and the number of bathrooms, and the floor attributes include the floor and the total number of floors; Convert the house type attributes into numerical house type feature vectors, and calculate the house type Euclidean distance matrix of the real estate lot cluster according to the house type feature vectors; Use the hierarchical clustering algorithm to cluster the house type Euclidean distance matrix to obtain K1 house type sub-clusters; According to the K1 house type sub-clusters, convert the floor attributes into numerical floor feature vectors, and calculate the floor Euclidean distance matrix of the sub-clusters according to the floor feature vectors; Use the K-means clustering algorithm to cluster the floor Euclidean distance matrix to obtain K2 house type sub-clusters; According to the K1 house type sub-clusters and the K2 house type sub-clusters, divide each real estate lot cluster into K1*K2 house type-floor sub-clusters.

7. The method for operating and managing real estate assets based on big data analysis according to claim 6, wherein: Obtain the price mean distribution and price median distribution of different combinations of lot, house type and floor, including: According to each real estate lot cluster, extract the transaction price list price_list of the real estate lot cluster; Traverse the real estate lot cluster, and add the transaction price of each real estate to price_list; Calculate the arithmetic mean of price_list to obtain the average price cluster_avg_price of the real estate lot cluster; Sort price_list in ascending order, and select the element in the middle position as the median price cluster_median_price of the real estate lot cluster; According to each house type-floor sub-cluster, extract the transaction price list sub_price_list of all real estates within the house type-floor sub-cluster; Calculate the arithmetic mean of sub_price_list to obtain the average price sub_cluster_avg_price of the house type-floor sub-cluster; Sort sub_price_list in ascending order, and select the element in the middle position as the median price sub_cluster_median_price of the house type-floor sub-cluster.

8. The method for operating and managing real estate assets based on big data analysis according to claim 7, wherein: Obtaining the price mean distribution and price median distribution of different combinations of lot, house type and floor further includes: Initialize the lot average price field, the lot median price field, the house type-floor average price field and the house type-floor median price dictionary; Traverse each cluster of immovable property lots, use the lot information as the key, and store the corresponding average price and median price into the lot average price field and the lot median price dictionary respectively; Traverse each house type - floor sub - cluster, use the combination of lot, house type and floor information as the key, and store the corresponding average price and median price into the house type - floor average price dictionary and the house type - floor median price dictionary respectively; Output the average price distribution of different lots, the median price distribution of different lots, the average price distribution of different lots and house type - floor combinations, and the median price distribution of different lots and house type - floor combinations.

9. The method for operating and managing immovable property assets based on big data analysis according to claim 8, characterized in that: Construct a prediction model for immovable property transaction probability through a logistic regression algorithm, including: Extract the feature vector of each immovable property data, and the feature vector includes lot, house type and floor; Extract the timestamp of the immovable property data, and obtain the predicted prices corresponding to the corresponding time granularities from the daily house price change curve, the weekly house price change curve and the monthly house price change curve respectively as the price dimension of the feature vector; Extract the lot of the immovable property data, and obtain the corresponding price mean and median from the average price distribution of different lots and the median price distribution of different lots respectively as the lot price dimension of the feature vector; Extract the lot, house type and floor of the immovable property data, and extract the corresponding price mean and median from the average price distribution of different lots and house type - floor combinations and the median price distribution of different lots and house type - floor combinations respectively as the house type - floor price dimension of the feature vector; Construct a multi - dimensional feature vector according to the price dimension of the feature vector, the lot price dimension of the feature vector, and the house type - floor price dimension of the feature vector, and mark positive and negative class labels according to whether the immovable property is transacted to form a training sample set; Use the training sample set to train a logistic regression model through the gradient descent algorithm to predict the transaction probability of immovable property.

10. A system for operating and managing immovable property assets based on big data analysis, characterized in that it includes: At least one processing unit; used to execute instructions to implement the method for operating and managing immovable property assets based on big data analysis according to any one of claims 1 to 9.

Images