A method for identifying regional market structure by fusing hedonic price model and complex network

Abstract

The application discloses a method for identifying regional market structure by fusing feature price model and complex network, comprising the following steps: collecting house price data of cities in counties nationwide; combining feature price model and complex network to establish a city regional connection model; reconstructing the network based on prefecture-level city scale statistical convergence index; and calculating the network by adopting louvain algorithm to obtain the regional structure of prefecture-level cities nationwide. The application proposes a method for measuring the market convergence connection degree between cities by fitting goodness based on feature price model and complex network, and further digs out the regional structure; takes the connection network between real estate markets in counties and cities as the basis, and further reconstructs the network at the prefecture-level city scale, so as to balance the analysis accuracy and operability, capture the market similarity from micro to macro, and realize the comprehensive identification of the regional structure of the real estate market nationwide.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of regional structure research, and particularly relates to a method for identifying regional market structure by fusing a hedonic price model and a complex network. BACKGROUND

[0002] With the acceleration of urbanization and the continuous development of the real estate market, the study of regional market structure has become an important issue in urban planning and economic development. Urbanization brings population concentration and resource relocation, which promotes the closer economic ties between cities. As an important part of the urban economy, the structure of the real estate market directly affects the sustainable development of the city. The study of regional market structure helps to reveal the potential mechanism of the market, identify the key factors affecting house prices, and provide a scientific basis for policy making. In addition, the analysis of regional market structure can also reflect the interaction between cities and reveal the flow trend of economic activities within and outside the region, helping planners to rationally allocate resources and optimize the layout of urban functions.

[0003] Traditional regional division methods often rely on geographical location, administrative division, passenger flow or patent industry association. These methods, although to some extent reveal the characteristics of the region, but in the face of complex and changing real estate market, often cannot fully reflect the market correlation and interaction between different regions. Therefore, in the current rapidly changing economic environment, it is insufficient to rely only on these traditional division methods. The hedonic price model is based on the consumer's preference for different characteristics, revealing various factors affecting urban market connection, and has the advantage of quantifying complex market factors, thereby providing strong support for regional economic analysis. On the other hand, the complex network theory regards cities as nodes in the network and market connections as edges, and complex network analysis can reveal the interaction and structural characteristics between cities.

[0004] Therefore, in order to more comprehensively understand the complexity of the regional market, it is urgent to develop a regional structure identification method that takes market connection and interaction into account to achieve in-depth analysis of the regional market structure. The model that fuses the hedonic price model and the complex network can measure the complex connection between urban markets and further identify the regional market structure. However, there is currently no complete method for identifying regional structure by fusing the hedonic price model and the complex network. SUMMARY

[0005] In view of the above problems existing in the prior art, the application provides a method for identifying regional market structure by fusing a characteristic price model and a complex network, which quantifies the connection strength between city markets and then identifies the regional market structure based on the similarity degree of the internal mechanism of a real estate market and in combination with a complex network and a characteristic price model. The connection between city real estate markets is measured by the characteristic price model to re-understand the city regional structure, which provides a new perspective and method for understanding the regional market structure and policy positioning.

[0006] The technical scheme of the application is as follows:

[0007] A method for identifying regional market structure by fusing a characteristic price model and a complex network, comprising the following steps:

[0008] S1, collecting the house price data of cities in counties nationwide, defining as house price sample points, obtaining a set HP; the house price sample points comprise house unit price information, house structure information and house longitude and latitude information; the house structure information comprises area, number of bedrooms, number of bathrooms and floor;

[0009] S2, combining a characteristic price model and a complex network to establish a city regional connection model, and the specific steps are as follows:

[0010] S2-1, taking a county-level city as a research unit, constructing an initial network of city connections with mutual edges, denoted as network G0(V,E); each county-level city is a node of the network, referred to as a county city node;

[0011] S2-2, according to the house price sample points of the set HP, establishing a characteristic price model for the county-level cities, and the variables used include house structure information, location characteristics and facility distance; the location characteristics refer to the road selection degree and integration degree calculated by using a spatial syntax software Depthmap; the facility distance refers to the distance between a house and a hospital, an elementary school, a middle school and a supermarket;

[0012] S2-3, selecting any two adjacent county city nodes i and j in the network G0(V,E), calculating the goodness of fit of the respective characteristic price models, denoted as R i 2 and R j 2 ; merging the data of i and j, and calculating the goodness of fit of the merged characteristic price model again, denoted as R ij 2 ;

[0013] S2-4, calculating the network edge weight W1, namely:

[0014]

[0015] S2-5, weighting G0(V, E) network; V is the county city node, and the node weight is the goodness of fit of the county city node; E is the network edge between the county city nodes i and j, and the weight of the network edge is W1;

[0016] S2-6, removing the edge with W1 less than 1 in G0(V, E), that is, obtaining the regional contact network G1(V, E) on the county scale;

[0017] S3, based on the statistical convergence index of the prefecture scale and reconstructing the network, the specific steps are as follows:

[0018] S3-1, taking the prefecture city as the research unit, constructing the initial network of urban contact with mutual common edge, denoted as network G2(V, E); each prefecture city is a node of the network, called prefecture city node;

[0019] S3-2, counting the number of edges in G1(V, E) network in the prefecture, as the node weight of the prefecture node in G2(V, E) network;

[0020] S3-3, counting the number of edges in G1(V, E) network between any adjacent prefectures, as the edge weight W2 in G2(V, E) network;

[0021] S3-4, removing the edge with W2 being 0 in G2(V, E), obtaining the regional convergence network G3(V, E) on the prefecture scale;

[0022] S4, using Louvain algorithm to calculate G3(V, E), obtaining the regional structure of the national prefecture scale, the specific steps are as follows:

[0023] S4-1, initializing the maximum modularity and the best resolution, selecting the resolution range and the step size, dividing the resolution range into a series of resolutions with the step size as the unit;

[0024] S4-2, repeating the execution of Louvain algorithm for each resolution, and recording the community division result each time; selecting the value of random seed;

[0025] S4-3, counting the distribution times of each prefecture node in different communities, generating community distribution probability, and selecting the community to which the prefecture node is most commonly distributed as the final division;

[0026] S4-4, calculating the modularity of the division, finally outputting the maximum modularity and the corresponding best resolution, and further obtaining the regional structure of the national prefecture scale.

[0027] Further, in step S4-1, the maximum modularity and the best resolution are -1, the step size is 0.1, and the resolution range is 0.1-2.0.

[0028] In step S4-1, the maximum modularity and the best resolution are -1, the step is 0.1, and the resolution range is 0.1-2.0;

[0029] In step S4-2, the number of times of repeating the Louvain algorithm is 10, and the value of the random seed is 42.

[0030] The present application has the beneficial technical effects that:

[0031] (2) The present application proposes a method for measuring the market convergence contact degree between cities by fitting goodness-of-fit based on the characteristic price model and the complex network, and further mining the regional structure;

[0032] (3) The regional structure recognition method proposed by the present application is based on the contact network between the real estate markets of the district and county cities, and further reconstructs the network at the prefecture-level city scale. From the regional scale to the prefecture-level city scale, the analysis precision and operability are considered, the market similarity from the micro to the macro is captured, and the overall recognition of the regional structure of the national real estate market is realized. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 The figure is a schematic diagram of the steps of the present application;

[0034] Figure 2 The figure is the national regional structure identified based on the present application. DETAILED DESCRIPTION

[0035] The present application will be described in detail below in combination with the drawings and examples. Obviously, the described examples are only some of the embodiments of the present application, but not all the embodiments. Based on the examples in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0036] The example uses the present application to identify the regional structure of the cities in China. The specific steps are as follows:

[0037] S1, collect the house price data of the prefecture-level cities in China through network crawling of the second-hand house transaction websites such as Anjuke and Shell (or other data acquisition methods), including the house price information, the area, the number of bedrooms, the number of bathrooms, the floor and other house structure attributes for subsequent establishment of the characteristic price model, and the latitude and longitude information for geographic coding processing into shapefile spatial data.

[0038] The housing price data of each prefecture-level city is corrected to obtain the data in WGS1984 coordinate system, and the housing price data of each county is screened out using the spatial location screening tool in Arcgis Pro and using the administrative division data of counties in China (also in WGS1984 coordinate system). There are 1515 county cities with more than 100 housing price data.

[0039] S2, combine the hedonic price model and the complex network to establish the urban regional connection model, the specific steps are as follows:

[0040] S2-1, call the geopandas library in python to calculate the adjacency matrix of the shp file of the county-level city scale in China using the intersects function, and then call the undirected graph function in the networkx library in python to build the initial network of urban connection with common edges, denoted as network G0(V,E) (as shown in the upper left of Figure 1 );

[0041] S2-2, establish a hedonic price model for county cities with more than 100 housing price points, the variables used include: housing structure information, location characteristics and facility distance; housing structure information includes area, number of bedrooms, number of bathrooms, floor and other information collected in the data, location characteristics include road selection degree and integration calculated using the spatial syntax software Depthmap, and facility distance includes the distance to hospitals, primary schools, secondary schools and supermarkets.

[0042] S2-3, call the statsmodels library in Python to calculate the multicollinearity of all variables using the variance_inflation_factor function, and gradually remove the highest factor of VIF (i.e. variance inflation factor) and calculate VIF again until the VIF of all variables is less than or equal to 5, to obtain the final variables for building the model.

[0043] S2-4, select any two adjacent nodes i and j in the G0(V,E) network, call the statsmodels library in Python to perform multiple linear regression using the OLS function, and obtain the respective goodness of fit, denoted as R i 2 and R j 2 ; merge the data of i and j, and call the statsmodels library in Python again to perform multiple linear regression using the OLS function, and obtain the goodness of fit of the merged data, denoted as R ij 2 ;

[0044] S2-5, calculate the network edge weight W1, that is:

[0045]

[0046] S2-6, weight the network G0(V, E), where V is the city node, the node weight is the goodness of fit of the city node, E is the network edge, and the weight of the edge is W1;

[0047] S2-7, remove the edges with W1 less than 1 in the network G0(V, E), and obtain the regional contact network G1(V, E) at the county scale (as shown in the upper right of Figure 1 );

[0048] S3, reconstruct the network based on the prefecture-level city scale statistical convergence index, and the specific steps are as follows:

[0049] S3-1, call the geopandas library in python to calculate the adjacency matrix of the shp file at the prefecture-level city scale using the intersects function, and then call the undirected graph function in the networkx library in python to build the initial network of the contact between the prefecture-level cities with common edges, denoted as network G2(V, E) (as shown in the lower left of Figure 1 );

[0050] S3-2, count the number of edges in the G1 network inside the prefecture-level city as the node weight of the G2 network;

[0051] S3-3, count the number of edges in the G1 network between any adjacent prefecture-level cities as the edge weight W2 in the G2 network;

[0052] S3-4, delete the edges with W2 of 0 in the G2 network to obtain the regional convergence network G3(V, E) at the prefecture-level city scale;

[0053] S4, call the louvain algorithm in the community function in python to calculate the regional convergence network G3(V, E) at the prefecture-level city scale, initialize the maximum modularity and the best resolution to -1, and then run the Louvain algorithm multiple times in the resolution range of 0.1 to 2.0 (step size of 0.1) to find the best community division. For each resolution, the Louvain algorithm is run 10 times, and the community division result of each time is recorded. The random seed is selected as 42. Then, the number of assignments of each node in different communities is counted to generate the community assignment probability, and the community to which the node is most commonly assigned is selected as the final division. Subsequently, the modularity of the division (used to evaluate the quality of community division) is calculated, and the maximum modularity and the corresponding best resolution are finally output, and the regional structure at the prefecture-level city scale is obtained, as shown in the lower right of Figure 1 .

[0054] S5. Use the GeoPandas and Matplotlib libraries in Python to visualize the obtained regional market structure.

[0055] like Figure 2 As shown, based on existing data and the regional market structure obtained using this method, 34 communities were identified, of which 15 communities have a size (number of nodes) of 10 or more. This example demonstrates the regional structure division of the entire country (based on existing data) according to real estate market connections.

[0056] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, and for those of ordinary skill in the art, various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. Therefore, the present invention is not limited to the specific details without departing from the general concept defined by the claims and their equivalents.

Claims

1. A method for identifying regional market structure by integrating hedonic pricing models and complex networks, characterized in that, Includes the following steps: S1. Collect housing price data for cities and counties across the country, define them as housing price sample points, and obtain the set HP; the housing price sample points include: housing unit price information, housing structure information, and housing latitude and longitude information; the housing structure information includes: area, number of bedrooms, number of bathrooms, and number of floors; S2. Establish an urban regional connection model by combining the hedonic price model and complex networks. The specific steps are as follows: S2-1. Taking county-level cities as the research unit, construct an initial network of cities that share edges with each other, denoted as network G0(V,E); each county-level city is a node in this network, called a county-level city node; S2-2. Based on the housing price sample points of the set HP, establish a characteristic price model for county-level cities. The variables used include: housing structure information, location characteristics, and facility distance; the location characteristics refer to the road selectivity and integration degree calculated using the spatial syntax software Depthmap; the facility distance refers to the distance between the house and the hospital, primary school, middle school, and supermarket. S2-3. Select any two adjacent county / city nodes i and j in the G0(V,E) network, and calculate the goodness of fit of their respective feature price models, denoted as R. i 2 With R j 2 Combine the data from i and j, and recalculate the goodness of fit of the combined feature price model, denoted as Ri. ij 2 ; S2-4. Calculate the network edge weights W1, i.e.: S2-5. Assign weights to the G0(V,E) network; V is a district / county / city node, and the node weight is the goodness of fit of that district / county / city node; E is the network edge between district / county / city nodes i and j, and the weight of this network edge is W1. S2-6. Remove the edges with W1 less than 1 in G0(V,E) to obtain the regional connection network G1(V,E) at the county scale. S3. Reconstruct the network based on statistical convergence indicators at the prefecture-level city scale. The specific steps are as follows: S3-1. Taking prefecture-level cities as the research unit, construct an initial network of cities that share edges with each other, denoted as network G2(V,E); each prefecture-level city is a node in this network, called a prefecture-level city node; S3-2. Count the number of edges in the G1(V,E) network within a prefecture-level city, and use it as the node weight of the prefecture-level city node in the G2(V,E) network. S3-3. Count the number of edges in the G1(V,E) network that exist between any two adjacent prefecture-level cities, and use it as the edge weight W2 in the G2(V,E) network. S3-4. Remove the edges with W2 equal to 0 in G2(V,E) to obtain the regional convergence network G3(V,E) at the prefecture-level city scale. S4. The Louvain algorithm is used to calculate G3(V,E) to obtain the regional structure at the prefecture-level city scale across the country. The specific steps are as follows: S4-1. Initialize the maximum modularity and optimal resolution, select the resolution range and step size, and divide the resolution range into a series of resolutions in units of step size; S4-2. Repeat the Louvain algorithm for each resolution and record the community partitioning results each time; select the value of the random seed; S4-3. Count the number of times each prefecture-level city node is assigned to different communities, generate community assignment probabilities, and select the community to which the prefecture-level city node is most frequently assigned as the final division. S4-4 Calculate the modularity of the partition, and finally output the maximum modularity and its corresponding optimal resolution, thereby obtaining the regional structure at the prefecture-level city scale across the country.

2. The method for identifying regional market structure by fusing feature pricing models and complex networks according to claim 1, characterized in that: In step S4-1, the maximum modularity and optimal resolution are -1, the step size is 0.1, and the resolution range is 0.1 to 2.

0. In step S4-2, the Louvain algorithm is executed 10 times; the value of the random seed is 42.

Images