Real estate valuation systems, methods, and programs

Abstract

This invention provides a real estate price appraisal system, method, and program that enable real estate agents to easily and accurately determine the fair transaction price of real estate. [Solution] The real estate price appraisal system 1 includes an information acquisition unit 610 that acquires environmental data of the target floor and the target building having the target floor, and a calculation unit 630 that uses a trained model LM that has learned to output the rent and sales price of the target floor and the sales price of the target building from the environmental data.

Description

Technical Field

[0001] The present disclosure relates to a real estate price appraisal system, method, and program.

Background Art

[0002] As a preparation stage for real estate transactions, real estate traders must appraise the appropriate transaction price of the real estate to be traded. However, this appraisal requires a deep understanding of the real estate to be traded and even a deep understanding of the general real estate market for real estate traders. Therefore, it is difficult to appraise the appropriate transaction price of real estate. Thus, it is conceivable that real estate traders entrust this appraisal to experts. However, considering that the divergence between the appropriate transaction price appraised by experts and the actual market price for real estate has been regarded as a problem over the long term, there is a problem that the appropriate transaction price may not be appraised even if real estate traders entrust the appraisal to experts.

[0003] Therefore, Patent Document 1 describes a real estate transaction price proposal program for causing a computer to execute an information acquisition step of acquiring property information regarding the content of a real estate property and a search step of searching for a transaction price to be proposed based on the degree of association with the transaction price for reference property information corresponding to the property information acquired through the information acquisition step. Also, paragraphs 0063 to 0073 of the specification of Patent Document 1 describe creating the above degree of association by causing artificial intelligence to learn past real estate transaction data.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the invention described in Patent Document 1 requires a vast amount of data to create the aforementioned correlation index, but real estate brokers do not necessarily possess sufficient information about past real estate transactions (especially those conducted by other companies). For this reason, in practice, it is difficult for real estate brokers to implement the invention described in Patent Document 1.

[0006] Therefore, the purpose of this disclosure is to provide a real estate price appraisal system, method, and program that enable real estate transactionrs to easily and accurately determine the appropriate transaction price of real estate. [Means for solving the problem]

[0007] To achieve the above-mentioned objectives, the real estate price appraisal system relating to this disclosure comprises: an information acquisition unit that acquires environmental data of a target floor and a target building having said target floor; and a calculation unit that uses a trained model that has learned to output the rent and sales price of said target floor and the sales price of said target building from said environmental data.

[0008] Furthermore, the real estate price appraisal method relating to this disclosure comprises the steps of acquiring environmental data of the target floor and the target building having the target floor, and the steps of using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building from the environmental data.

[0009] Furthermore, the real estate price appraisal program relating to this disclosure involves causing a computer to perform the following steps: acquiring environmental data of the target floor and the target building having said target floor; and using a trained model that has learned to output the rent and sales price of said target floor and the sales price of said target building, the program performs the calculation step of outputting the rent and sales price of said target floor and the sales price of said target building from the environmental data. [Effects of the Invention]

[0010] The real estate price appraisal system, method, and program described herein enable real estate traders to easily and accurately appraise the fair transaction price of real estate. [Brief explanation of the drawing]

[0011] [Figure 1] This is an overall configuration diagram showing one embodiment of the real estate price appraisal system related to this disclosure. [Figure 2] Figure 1 shows the overall configuration of the trained model stored in the memory unit of the real estate price appraisal system. [Figure 3] Figure 1 shows an example of graph data output by the SHAP calculation unit. [Modes for carrying out the invention]

[0012] Next, an embodiment of the real estate price appraisal system related to this disclosure will be described in detail with reference to Figures 1 to 3. The real estate price appraisal system 1 shown in Figure 1 is managed, for example, by a real estate company engaged in the business of selling and leasing real estate. As shown in Figure 1, the real estate price appraisal system 1 includes a control unit 100, a communication unit 200, an input unit 300, a storage unit 400, a display unit 500, and a calculation unit 600.

[0013] The control unit 100 constitutes part of the central processing unit (CPU) of the real estate price appraisal system 1 and controls each part (described in the next paragraph) that is configured to communicate with each other. In other words, each of these parts is, for example, a program module for performing each function or is equipped with a program module, and is configured to execute its respective processing under control instructions from the control unit 100.

[0014] Specifically, the control unit 100 is interconnected with the communication unit 200, the input unit 300, the storage unit 400, the display unit 500, and the calculation unit 600 in a manner that allows for mutual communication. More specifically regarding the calculation unit 600, the control unit 100 is interconnected with the information acquisition unit 610, the learning unit 620, the calculation unit 630, and the verification unit 640 within the calculation unit 600 in a manner that allows for mutual communication. More specifically regarding the learning unit 620, the control unit 100 is interconnected with the three-element learning unit 621, the index learning unit 622, the quantile regression learning unit 623, the SHAP learning unit 624, and the integrated learning unit 625 within the learning unit 620 in a manner that allows for mutual communication. More specifically regarding the calculation unit 630, the control unit 100 is interconnected with the three-element calculation unit 631, the index calculation unit 632, the quantile regression calculation unit 633, and the SHAP calculation unit 634 of the calculation unit 630 in a manner that allows for mutual communication.

[0015] The communication unit 200 is configured to communicate with the communication unit of a terminal device (not shown) via network N. Network N is a communication network capable of bidirectional information transmission using wired or wireless communication means, such as the Internet, intranet, or VPN (Virtual Private Network). For example, the communication unit 200 is a wired LAN, wireless LAN, etc., that can connect to the Internet.

[0016] The input unit 300 is an operating means that can input control instructions to the control unit 100, and is, for example, a touch panel such as a liquid crystal display or an organic EL display, a keyboard, a mouse, or an audio input device.

[0017] The memory unit 400 is configured to store various types of information, and is composed of, for example, a main memory device (hard disk, SSD, optical disk, etc.), an auxiliary memory device (ROM, RAM, etc.), a cache memory, and registers. The various types of information here include: (1) environmental data of various learning floors (learning environmental data), (2) environmental data of various learning buildings having these various learning floors (learning environmental data), (3) data on the rent and selling prices of these various learning floors (learning price data), and (4) selling price data of various learning buildings having these various learning floors (learning price data), and (5) a learned model LM (see Figure 2) into which a large amount of this learning environmental data and learning price data has been input. Note that the learning floors include both actual floors that exist in reality and virtual floors that are virtual settings that do not exist in reality. Also, the learning buildings include both actual buildings that exist in reality and virtual buildings that are virtual settings that do not exist in reality.

[0018] The memory unit 400 stores the learned model LM shown in Figure 2. The learned model LM includes a plurality of base models BM and a meta-model MM. Each base model BM includes a three-element model B1, an index model B2, a quantile regression model B3, and a SHAP model B4.

[0019] The display unit 500 is, for example, a liquid crystal display or an organic EL display, and is configured to be able to display on one screen the rent and selling price of the target floor, and the selling price of the target building having the target floor.

[0020] The arithmetic unit 600 includes an information acquisition unit 610, a learning unit 620, a calculation unit 630, and a verification unit 640.

[0021] The information acquisition unit 610 of the calculation unit 600 is configured to be able to acquire various information from the communication unit 200, the input unit 300, the storage unit 400, the calculation unit 630, etc. For example, (1) target environmental data, which is environmental data of the target floor and the target building having the target floor, (2) learning price data, which is data on the rent and selling price of the learning floor, and the selling price of the learning building having the learning floor, and (3) learning environmental data, which is environmental data of the learning floor and the learning building, and (4) population environmental data of an actual or virtual property population (an aggregate composed of floors and buildings) are acquired for each geographical division. Note that the learning floor includes both an actual floor, which is an existing floor, and a virtual floor, which is a non-existing floor in a virtual setting. Further, the learning building includes both an actual building, which is an existing building, and a virtual building, which is a non-existing floor in a virtual setting. As a geographical division, for example, one area obtained by dividing, without gaps, an area displayed on a map into meshes can be mentioned.

[0022] The learning unit 620 of the calculation unit 600 creates a learned model LM (see FIG. 2) by pseudo (single output) multi-task learning, and includes a three-element learning unit 621, an index learning unit 622, a quantile regression learning unit 623, a SHAP learning unit 624, and an integrated learning unit 625.

[0023] The three-element learning unit 621 creates a three-element model B1 (FIG. 2), which is a learned model that has learned to output numerical values (output data) of the rent and selling price of the target floor and the selling price of the target building having the target floor. Specifically, the three-element learning unit 621 creates the three-element model B1 by using (1) learning environmental data, which is environmental data of the learning floor and the learning building, and (2) learning price data, which is data on the rent and selling price of the learning floor and the selling price of the learning building, as first learning data D1. The three-element learning unit 621 creates the three-element model B1 based on GBDT-PL (gradient boosting decision tree with leaf nodes being linear regression models), for example, by using LightGBM.

[0024] Furthermore, the three-element learning unit 621 may create a three-element model B1 by learning that includes weighting the first data set, which includes the rent and environment of the learning floor; the second data set, which includes the sales price and environment of the learning floor; and the third data set, which includes the sales price and environment of the learning building. Generally, considering that the number of samples increases in the order of the first data set, the second data set, and the third data set, it is preferable to make the weighting heavier in the order of the first data set, the second data set, and the third data set.

[0025] Furthermore, the three-element learning unit 621 may create a three-element model B1 by learning that includes smoothing, which brings the statistics (e.g., mean values) of the learning environment data (e.g., year of construction, area in tsubo, number of floors) in geographical divisions with a small number of acquired learning price data (small areas) closer to the statistics of the learning environment data in geographical divisions with a large number of acquired learning environment data (many areas). Alternatively, the three-element learning unit 621 may create a three-element model B1 by learning that includes smoothing, which brings the statistics (e.g., mean values) of the learning price data in geographical divisions with a small number of acquired learning price data (e.g., rent) closer to the statistics of the learning price data in geographical divisions with a large number of acquired learning price data.

[0026] The 3-element learning unit 621 performs the smoothing described above using "ξ=(1-f(n))×ξ small +f(n)×ξ big This can be done using the formula: ξ (Xi): Smoothing statistic n: Number of data points in a small area ξ small :Statistical data for small areas ξ big : Statistics from multiple regions f(n): Sigmoid function The sigmoid function is the reciprocal of 1 + exp((na) / b). For the sigmoid function, the parameters used are, for example, a=0 and b=10.

[0027] The index learning unit 622 creates an index model B2 by using data on the rent, sales price, and environment of an actual learning floor, and data on the sales price and environment of an actual learning building as the first learning data D1. When the index model B2 is input, which is collective environmental data of an actual or virtual property group (a collection consisting of floors and buildings), it outputs an index as output data, such as the average rent and sales price of the floors in this property group, and the average sales price of the building that has these floors. This index may be output as numerical data at a single point in time, or as graph data showing its trend over time. The index learning unit 622 creates the index model B2 based on GBDT-PL (a gradient boosting decision tree where the leaf nodes are linear regression models), for example, by using LightGBM.

[0028] The quantile regression learning unit 623 creates a quantile regression model B3 that has learned to output the range of rent and sales prices for the target floor, as well as the sales price range of the target building containing the target floor, for each probability density. Specifically, the quantile regression learning unit 623 creates the quantile regression model B3 by using (1) training environment data, which is environmental data of the training floor and training building, and (2) training price data, which is data of rent and sales prices for the training floor, and the sales price of the training building, as the first training data D1. Unlike the 3-element model B1, which outputs numerical values, the quantile regression learning unit 623 outputs ranges for each probability density as output data. The quantile regression learning unit 623 creates the quantile regression model B3 based on GBDT-PL (a gradient boosting decision tree where the leaf nodes are linear regression models), for example, by using LightGBM.

[0029] The SHAP learning unit 624 creates SHAP model B4, which has learned to classify the environment of the target floor and target building into price increase factors that increase the rent and sales price of the target floor, and price decrease factors that decrease the sales price of the target building. Specifically, the SHAP learning unit 624 creates SHAP model B4 by using (1) learning environment data, which is environmental data of the learning floor and learning building, and (2) learning price data, which is data of the rent and sales price of the learning floor, and the sales price of the learning building, as the first learning data D1.

[0030] When SHAP model B4 receives target environmental data, which is environmental data for the target floor and the target building, it classifies the environmental information in the input target environmental data into price increase factors and price decrease factors and outputs it as "text data or graph data" (output data). SHAP learning unit 624 creates SHAP model B4 based on GBDT-PL (a gradient boosting decision tree where the leaf nodes are linear regression models), for example using LightGBM.

[0031] Furthermore, the SHAP learning unit 624 calculates SHAP values ​​for arbitrary factors (e.g., number of months since construction, number of floors, etc.) for each property group (learning property group) consisting of one or both of multiple learning floors and learning buildings (for example, for learning buildings in the learning property group that are 1 month old, 10 months old, and 100 months old, the SHAP values ​​for 1 month old, 5 months old, and 100 months old are calculated respectively). The SHAP learning unit 624 then outputs the distribution of individual SHAP values ​​in the learning property group for each of the multiple factors (e.g., number of months since construction, number of floors) as graph data (output data).

[0032] The integrated learning unit 625 creates a metamodel MM based on multiple (four in Figure 2) base models BM. Specifically, the integrated learning unit 625 inputs the secondA training data D2A into "multiple base models BM that have been trained on the first training data D1," and uses the secondB training data D2B, which includes the secondA training data D2A and the output data from the multiple base models BM, as training data to create the metamodel MM.

[0033] Specific examples of creating a metamodel MM using the 2B training data D2B include: (1) having the metamodel MM compare output data from multiple base models BM (for example, predicted rent for training floors included in the 2A training data D2A) with ground truth data (for example, actual rent and environment for training floors included in the 2A training data D2A); and (2) using the output data from these multiple base models BM and model information (for example, the prediction accuracy of each of the multiple base models BM) to weight the output data from the multiple base models BM in the metamodel MM.

[0034] Furthermore, the index learning unit 622, the quantile regression learning unit 623, the SHAP learning unit 624, and the integrated learning unit 625 may, if necessary, perform the "weighting" and "smoothing" processes that can be performed by the three-element learning unit 621.

[0035] The calculation unit 630 of the arithmetic unit 600 performs calculations using a trained model LM, and includes a 3-element calculation unit 631, an index calculation unit 632, a quantile regression calculation unit 633, and a SHAP calculation unit 634.

[0036] The three-element calculation unit 631 inputs base input data D3A, which includes environmental data for the target floor and the target building (target environmental data), into a plurality of three-element models B1, and outputs intermediate output data (not shown) from each of the plurality of three-element models B1, which includes data on the rent and sales price of the target floor and the sales price of the target building (target price data). Here, it is preferable that the three-element calculation unit 631 (specifically the three-element models B1) (1) output the rent for the target floor, (2A) output the sales price of the target floor from the rent for the target floor based on the estimated ratio of the sales price of the target floor to the rent for the target floor, and (2B) output the sales price of the target building from the rent for the target floor based on the estimated ratio of the sales price of the target building to the rent for the target floor. Note that the order of (2A) and (2B) is not particularly limited as long as they occur after (1).

[0037] Furthermore, the three-element calculation unit 631 inputs meta-input data D3B, which includes base input data D3A and intermediate output data output from multiple three-element models B1, into the meta-model MM, causing the meta-model MM to output final output data D3C, which is the final data for the rent and sales price of the target floor and the sales price of the target building.

[0038] The index calculation unit 632 inputs base input data D3A, which includes environmental data (collective environmental data) of real or virtual property groups (collections of floors and buildings), into multiple index models B2, and outputs indicators such as the average values ​​of floor rents and sales prices, as well as the sales prices of buildings that have these floors (i.e., indicators for floor rents, floor sales prices, and building sales prices) as intermediate output data (not shown) such as numerical data or graph data.

[0039] Furthermore, the index calculation unit 632 inputs meta-input data D3B, which includes base input data D3A and intermediate output data output from multiple index models B2, into the meta-model MM, and outputs final output data D3C from the meta-model MM, which is the final data for the rent and sales price of the floors in this property group, as well as the sales price of the building that has these floors.

[0040] The index calculation unit 632, when the number of environmental data for a real group of properties (real group environmental data) at any given time is less than or equal to a predetermined value, inputs the environmental data for a virtual group of properties (virtual group environmental data) along with the real group environmental data at that time into the index model B2 as base input data D3A, thereby outputting the index at that time as intermediate output data. Furthermore, when the index calculation unit 632 outputs the trend of the index over any given period as graph data (intermediate output data), it maintains the number of group environmental data used as base input data D3A at any given time during that period within a predetermined range (preferably maintaining it to be the same) in order to improve the accuracy of the output index.

[0041] The quantile regression calculation unit 633 inputs base input data D3A, which includes environmental data for the target floor and the target building (target environmental data), into multiple quantile regression models B3, and outputs intermediate output data (not shown) which are graph data (i.e., error distributions) for each probability density of the rent and sales price of the target floor and the sales price of the target building (target price data) from the multiple quantile regression models B3.

[0042] Here, it is preferable that the quantile regression calculation unit 633 (specifically, the quantile regression model B3) (1) outputs the rent of the target floor, (2A) outputs the sales price of the target floor from the rent of the target floor based on the estimated ratio of the sales price of the target floor to the rent of the target floor, and (2B) outputs the sales price of the target building from the rent of the target floor based on the estimated ratio of the sales price of the target building to the rent of the target floor. Note that the order of (2A) and (2B) is not particularly limited as long as they come after (1).

[0043] Furthermore, the quantile regression calculation unit 633 inputs meta-input data D3B, which includes base input data D3A and intermediate output data output from multiple quantile regression models B3, into the meta-model MM, and outputs the rent and sales price of the target floor, as well as the sales price of the target building, for each probability density as final output data D3C, which is graph data (i.e., error distribution) from the meta-model MM.

[0044] The SHAP calculation unit 634 inputs base input data D3A, which includes environmental data for the target floor and the target building (target environmental data), into multiple SHAP models B4. The SHAP models B4 classify the environmental information in the input target environmental data into price increase factors and price decrease factors, and output the classified environmental information from the SHAP models B4 as "text data or graph data" (intermediate output data not shown). For example, the SHAP calculation unit 634 outputs the classified environmental information at the same time that the 3-element calculation unit 631 outputs data for the rent and sales price of the target floor and the sales price of the target building. Here, price increase factors are factors (environmental information) that increase the rent and sales price of the target floor and the sales price of the target building, and price decrease factors are factors (environmental information) that decrease the rent and sales price of the target floor and the sales price of the target building.

[0045] Furthermore, the SHAP calculation unit 634 quantifies the impact of each of these price-increasing and price-decreasing factors on the rent and sales price of the target floor and the sales price of the target building as SHAP values ​​in the SHAP model B4, and outputs these SHAP values ​​for each price-increasing and price-decreasing factor as graph data (intermediate output data not shown in the graph).

[0046] Furthermore, the SHAP calculation unit 634 inputs meta-input data D3B, which includes base input data D3A and intermediate output data output from multiple SHAP models B4, into the meta-model MM, and outputs final output data D3C, which is the final data of the classified environmental information and its SHAP value, from the meta-model MM.

[0047] To explain in detail the calculations of the SHAP calculation unit 634, the SHAP calculation unit 634 classifies the environment of the target floor (rent of properties in the vicinity of the target floor, the number of floors of the target floor, the number of months since the target building containing the target floor was built, etc.) into multiple items (rent, number of floors, number of months since construction, etc.), and classifies these multiple items into price increase factors and price decrease factors. As the criterion for this classification, for example, the average value of environmental data of existing properties is used. In other words, for example, regarding the number of months since construction, if the number of months since construction is small (e.g., 5 months), the SHAP value of the number of months since construction will be a positive value and the number of months since construction will be classified as a price increase factor, and if the number of months since construction is large (e.g., 800 months), the SHAP value of the number of months since construction will be a negative value and the number of months since construction will be classified as a price decrease factor.

[0048] Furthermore, the SHAP calculation unit 634 calculates the SHAP value for an arbitrary factor (e.g., number of months since construction) for a group of properties (target property group) consisting of one or both of multiple target floors and target buildings. For example, the SHAP calculation unit 634 calculates the SHAP values ​​for 1 month, 5 months, and 100 months for target buildings in the target property group that are 1 month, 10 months, and 100 months old, respectively. The SHAP calculation unit 634 then outputs the correlation between the numerical value of the arbitrary factor in this target property group (e.g., the number of months since construction) and the SHAP value of this factor (e.g., the SHAP value of the number of months since construction) as graph data (intermediate output data not shown) from the SHAP model B4. Here, the SHAP calculation unit 634 may overlay the arbitrary factor (e.g., the number of months since construction) and the SHAP value of this factor (e.g., the SHAP value of the number of months since construction) in this graph data.

[0049] The SHAP calculation unit 634, by performing the functions described above, outputs graph data as shown in Figure 3, for example. In order to output this graph data, the SHAP calculation unit 634 extracts the following factors that affect the rent of the 16th floor (target floor) of the existing ABC building (target building): average rent of the surrounding 5th-order mesh, number of floors above ground, number of elevators, 4th-order mesh code, total floor area in tsubo, standard floor area in tsubo, total floor area, number of months since construction, district, and others. Of these factors, the number of floors above ground, number of elevators, total floor area in tsubo, standard floor area in tsubo, total floor area, and others are classified as price increase factors, and the other factors are classified as price decrease factors. Then, the SHAP calculation unit 634 uses the average value of environmental data of an actual property (limited to that which has been trained on the SHAP learning unit 624 shown in Figure 1) to calculate a SHAP value for each of these factors (for example, -2410 yen for surrounding 5th-order mesh rent, +1976 yen for number of floors above ground, etc.). Furthermore, the SHAP calculation unit 634 calculates the rent f(x) = 24,179 yen for the 16th floor (target floor) of the ABC building (target building) by summing the average value of the rental data for the property corresponding to this environmental data (E[f(X)] = 18,994 yen shown in Figure 3) and the SHAP value for each of these factors.

[0050] The verification unit 640 of the calculation unit 600 verifies the intermediate output data from the base model BM and the final output data D3C from the metamodel MM in the trained model LM as the data to be verified. Specifically, the verification unit 640 uses verification data different from the training data used in the training unit 620 and verifies the data to be verified using a cross-validation method or the like. The verification data includes data on the environment, rent, and sales price of the verification floor, as well as data on the environment and sales price of the verification building that houses the verification floor.

[0051] The verification criteria include, for example, (1) whether noise and bias present only in some of the training data are reflected in the data to be verified, and (2) whether universal structures common to both the training data and the verification data are reflected in the data to be verified. The verification unit 640 determines whether the data to be verified is appropriate or inappropriate. If it determines that it is appropriate, it does not perform any special processing. If it determines that it is inappropriate, it stops the learning by the learning unit 620 or issues a warning to the user via the communication unit 200 and the display unit 500, etc.

[0052] The user performs a desired real estate valuation method by freely utilizing the functions of each component in the real estate valuation system 1 shown in Figure 1. Furthermore, the real estate valuation system 1 includes a real estate valuation program that causes the computer-based real estate valuation system 1 to execute this real estate valuation method.

[0053] According to the real estate price appraisal system 1, the three-element model B1 has learned to output "the rent and sales price of the target floor, and the sales price of the target building" (three elements). Therefore, the three-element calculation unit 631 can output the three elements with high accuracy even if there are missing values ​​in the first training data D1 input to the three-element model B1 (for example, the first training data D1 includes the rent of a predetermined training floor, but does not include the sales price). Thus, real estate transaction users can output the three elements with high accuracy even if they do not possess sufficient information about the property, making it easy and highly accurate to appraise the fair transaction price of the property.

[0054] Furthermore, according to the real estate price appraisal system 1, the three-element calculation unit 631 outputs the rent of the target floor, outputs the sales price of the target floor from the rent of the target floor based on the estimated ratio of the sales price of the target floor to the rent of the target floor, and outputs the sales price of the target building from the rent of the target floor based on the estimated ratio of the sales price of the target building to the rent of the target floor. Therefore, even if there are missing values ​​in the first training data D1 input to the three-element model B1, the three-element calculation unit 631 can output the three elements with even greater accuracy and speed.

[0055] According to the real estate price appraisal system 1, the three-element learning unit 621 assigns weights to the first data, which includes the rent and environment of the learning floor; the second data, which includes the sales price and environment of the learning floor; and the third data, which includes the sales price and environment of the learning building. The weighting is weighted in the order of the first data, second data, and third data, so that the weight of the first data, which is generally numerous, becomes relatively heavier, and the weight of the third data, which is generally few in number, becomes relatively lighter. Therefore, the three-element calculation unit 631 can output the three elements with even greater accuracy.

[0056] According to the real estate price appraisal system 1, the three-element learning unit 621 brings the statistics of the learning price data (large amount of data) in geographical divisions with a small amount of acquired learning price data closer to the statistics of the learning price data (small amount of data) in geographical divisions with a large amount of acquired learning price data. The three-element calculation unit 631 suppresses the influence of noise contained in the small amount of data and can output the three elements with even higher accuracy.

[0057] According to the real estate price appraisal system 1, the index calculation unit 632 outputs an index for the property group, allowing real estate transaction users to understand not only the price of individual properties but also the price index (e.g., average value) of the property group. As a result, real estate transaction users can appraise the fair transaction price of real estate with greater accuracy and can also predict the fair transaction price of real estate from the present to the future.

[0058] According to the real estate price appraisal system 1, the three-element calculation unit 631 outputs the three elements using the metamodel MM, thus enabling the output of the three elements with even greater accuracy.

[0059] According to the real estate price appraisal system 1, the quantile regression calculation unit 633 outputs the three elements for each probability density, so that the user, a real estate transaction dealer, can appraise the fair transaction price of the property with even greater accuracy based on this probability density and their own experience.

[0060] According to the real estate price appraisal system 1, the SHAP calculation unit 634 classifies the environment of the target floor and target building into price-increasing and price-decreasing factors and quantifies them. As a result, real estate transaction users can consider the quantified impact and appraise the fair transaction price of the property with even greater accuracy.

[0061] The real estate price appraisal system 1 according to this embodiment includes a calculation unit 630 which separately comprises a three-element calculation unit 631, an index calculation unit 632, a quantile regression calculation unit 633, and a SHAP calculation unit 634. However, the real estate price appraisal system according to this disclosure is not limited to this configuration and may include a calculation unit in which the functions of these calculation units 631 to 634 are programmed together (i.e., a calculation unit in which these calculation units 631 to 634 are integrated).

[0062] Furthermore, in the real estate price appraisal system 1 according to this embodiment, the calculation unit 600 comprises a learning unit 620 and a calculation unit 630. However, the real estate price appraisal system according to this disclosure is not limited to this configuration, and the learning unit 620 and the calculation unit 630 may be provided in separate systems or calculation units, and the calculation unit 630 may have a configuration that uses only learned models already stored in the storage unit 400.

[0063] Furthermore, in the real estate price appraisal system 1 according to this embodiment, the trained model LM includes multiple base models BM and metamodels MM, where the base model BM individually includes a three-element model B1, an index model B2, a quantile regression model B3, and a SHAP model B4. However, the real estate price appraisal system according to this disclosure is not limited to such a configuration, and may use only a single base model BM, or the base model BM may be a trained model that combines some or all of the functions of these models B1 to B4.

[0064] Finally, the configuration of Real Estate Price Appraisal System 1 is illustrated below. [1] An information acquisition unit that acquires environmental data of the target floor and the target building having the said target floor, The system includes a calculation unit that outputs the rent and sales price of the target floor and the sales price of the target building from the environmental data, using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building. Real estate price appraisal system. [2] The calculation unit described above, Output the rent for the aforementioned target floor, Based on the estimated ratio of the sales price of the target floor to the rent of the target floor, the sales price of the target floor is output from the rent of the target floor. Based on the estimated ratio of the sales price of the target building to the rent of the target floor, the sales price of the target building is output from the rent of the target floor. The real estate price appraisal system described in [1] above. [3] With an additional learning section, The information acquisition unit acquires learning price data, which is data on the rent and sales price of the learning floor and the sales price of the learning building having the learning floor, and learning environment data, which is environmental data of the learning floor and the learning building. The learning unit creates the trained model by learning which includes weighting first data including the rent and environment of the learning floor, second data including the sales price and environment of the learning floor, and third data including the sales price and environment of the learning building. The weighting is such that the first data, second data, and third data are given increasing weight in that order. The real estate price appraisal system described in [1] or [2] above. [4] With an additional learning section, The information acquisition unit acquires learning price data, which is data on the rent or sales price of a learning floor, or the sales price of a learning building that has the learning floor, and learning environment data, which is environmental data of the learning floor or the learning building, for each geographical division. The learning unit creates the trained model by learning that includes bringing the statistics of the training price data in the category with a small number of acquired training price data closer to the statistics of the training price data in the category with a large number of acquired training price data. A real estate price appraisal system described in any one of the above [1] to [3]. [5] With an additional learning section, The information acquisition unit acquires data including the rent, sales price, and environment of the learning floor, and the sales price and environment of the learning building having the learning floor, and further acquires collective environmental data of real or virtual property groups. The learning unit creates an index model of the trained model by learning the rent, sales price, and environment of the learning floor and the sales price and environment of the learning building. The calculation unit inputs the collective environment data into the index model and outputs indicators from the index model for the rent and sales price of the collective floor in the property group, as well as the sales price of the collective building having the collective floor. A real estate price appraisal system described in any one of the above [1] to [4]. [6] The calculation unit uses the metamodel of the trained model to output again the rent or sales price of the target floor, or the sales price of the target building, which was output by the base model of the trained model. A real estate price appraisal system described in any one of the above [1] to [5]. [7] The calculation unit uses the quantile regression model of the trained model to output the rent or sales price of the target floor, or the sales price of the target building, for each probability density. A real estate price appraisal system described in any one of the above [1] to [6]. [8] The calculation unit uses the SHAP model of the trained model to classify the environment of the target floor or the target building into price increase factors that increase the rent or sales price of the target floor or the sales price of the target building, and price decrease factors that decrease the rent or sales price of the target floor or the sales price of the target building, and quantifies the degree of influence on the rent or sales price of the target floor or the sales price of the target building for each price increase factor and each price decrease factor. A real estate price appraisal system described in any one of the above [1] to [7]. [9] A step of acquiring environmental data of the target floor and the target building having the said target floor, The system includes a calculation step of outputting the rent and sales price of the target floor and the sales price of the target building from the environmental data, using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building. Methods for appraising real estate prices.

[10] A step of acquiring environmental data of the target floor and the target building having the said target floor, Using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building, a calculation process is performed to output the rent and sales price of the target floor and the sales price of the target building from the environmental data, and the computer is to perform this calculation process. Real estate price appraisal program. [Explanation of Symbols]

[0065] 1. Real Estate Price Appraisal System 100 Control Unit 200 Communications Department 300 Input section 400 Storage section 500 Display 600 Arithmetic section 610 Information acquisition department 620 Learning Department 621 3-Element Learning Unit 622 Index Learning Unit 623 Quantile Regression Learning Unit 624 SHAP Learning Department 625 Integrated Learning Department 630 Calculation Unit 631 3-Element Calculation Unit 632 Index Calculation Unit 633 Quantile Regression Calculation Unit 634 SHAP Calculation Unit 640 Verification Department B1 3-element model B2 Index Model B3 Quantile Regression Model B4 SHAP model BM Base Model D1: First training data D2A 2A Training Data D2B 2B Training Data D3A Base Input Data D3B Meta Input Data D3C Final Output Data LM pre-trained model MM Metamodel

Claims

1. An information acquisition unit that acquires environmental data of the target floor and the target building having the said target floor, The system includes a calculation unit that outputs the rent and sales price of the target floor and the sales price of the target building from the environmental data, using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building. Real estate price appraisal system.

2. The calculation unit described above, Output the rent for the aforementioned target floor, Based on the estimated ratio of the sales price of the target floor to the rent of the target floor, the sales price of the target floor is output from the rent of the target floor. Based on the estimated ratio of the sales price of the target building to the rent of the target floor, the sales price of the target building is output from the rent of the target floor. The real estate price appraisal system according to claim 1.

3. With an additional learning section, The information acquisition unit acquires learning price data, which is data on the rent and sales price of the learning floor and the sales price of the learning building having the learning floor, and learning environment data, which is environmental data of the learning floor and the learning building. The learning unit creates the trained model by learning which includes weighting first data including the rent and environment of the learning floor, second data including the sales price and environment of the learning floor, and third data including the sales price and environment of the learning building. The weighting is such that the first data, second data, and third data are given increasing weight in that order. The real estate price appraisal system according to claim 1.

4. With an additional learning section, The information acquisition unit acquires learning price data, which is data on the rent or sales price of a learning floor, or the sales price of a learning building that has the learning floor, and learning environment data, which is environmental data of the learning floor or the learning building, for each geographical division. The learning unit creates the trained model by learning that includes bringing the statistics of the training price data in the category with a small number of acquired training price data closer to the statistics of the training price data in the category with a large number of acquired training price data. The real estate price appraisal system according to claim 1.

5. With an additional learning section, The information acquisition unit acquires data including the rent, sales price, and environment of the learning floor, and the sales price and environment of the learning building having the learning floor, and further acquires collective environmental data of real or virtual property groups. The learning unit creates an index model of the trained model by learning the rent, sales price, and environment of the learning floor and the sales price and environment of the learning building. The calculation unit inputs the collective environment data into the index model and outputs indicators from the index model for the rent and sales price of the collective floor in the property group, as well as the sales price of the collective building having the collective floor. The real estate price appraisal system according to claim 1.

6. The calculation unit uses the metamodel of the trained model to output again the rent or sales price of the target floor, or the sales price of the target building, which was output by the base model of the trained model. The real estate price appraisal system according to claim 1.

7. The calculation unit uses the quantile regression model of the trained model to output the rent or sales price of the target floor, or the sales price of the target building, for each probability density. The real estate price appraisal system according to claim 1.

8. The calculation unit uses the SHAP model of the trained model to classify the environment of the target floor or the target building into price increase factors that increase the rent or sales price of the target floor or the sales price of the target building, and price decrease factors that decrease the rent or sales price of the target floor or the sales price of the target building, and quantifies the degree of influence on the rent or sales price of the target floor or the sales price of the target building for each price increase factor and each price decrease factor. The real estate price appraisal system according to claim 1.

9. A step of acquiring environmental data of the target floor and the target building having the said target floor, The system includes a calculation step of outputting the rent and sales price of the target floor and the sales price of the target building from the environmental data, using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building. Methods for appraising real estate prices.

10. A step of acquiring environmental data of the target floor and the target building having the said target floor, Using a trained model that has learned to output the rent and sales price of the target floor and the sales price of the target building, a calculation process is performed to output the rent and sales price of the target floor and the sales price of the target building from the environmental data, and the computer is to perform this calculation process. Real estate price appraisal program.

Images