Real Estate Appraisal System Based on Generative Adversarial Network Roadshow Algorithm

Abstract

A real estate appraisal system includes a processing device, a storage device and a communication device electrically connected to the processing device, and a geographic information database and a macroeconomic indicator database electrically connected to the communication device. The processing device, based on an address and multiple facility locations of a real estate property from the geographic information database, calculates the number of adjacent facilities and multiple potentially problematic facilities within a predetermined distance of the address, and obtains a population count, a number of households, a population density value, and three corresponding rates of change. It also obtains multiple macroeconomic data from the macroeconomic indicator database, and inputs these data, along with multiple property information of the real estate property, into a generative adversarial network (GAN) algorithm appraisal model to obtain an appraised price for the real estate property.

Description

[Technical Field]

[0001] This invention relates to a real estate appraisal system, and more particularly to a real estate appraisal system based on the Generative Adversarial Network (GAN) algorithm. [Previous Technology]

[0002] Since real estate lending is a crucial business for many financial institutions, accurate real estate appraisal is particularly important for them to make appropriate lending decisions and manage related risks. Traditional real estate appraisal methods are mostly manual, such as having credit personnel search real estate transaction records to find properties with similar specifications, locations, and recent transactions as a reference, and then making corresponding price adjustments based on the actual condition of the collateral. Because manual methods are usually inefficient and prone to human intervention, with subsequent advancements in statistical, machine learning, and deep learning algorithms, financial institutions have begun to build automated real estate appraisal models using these algorithms as a reference for credit personnel to improve the efficiency of real estate lending operations. One common approach is to use a single deep learning model and train it with a large amount of real estate transaction records. However, whether there are other real estate appraisal systems with higher accuracy remains to be solved. [Summary of the Invention]

[0003] Therefore, the purpose of this invention is to provide a real estate appraisal system based on generative adversarial network algorithm with higher appraisal accuracy.

[0004] Therefore, the present invention provides a real estate appraisal system based on generative adversarial network algorithm, comprising a geographic information database, a general economic indicator database, a storage device, a communication device, and a processing device.

[0005] The geographic information database stores multiple adjacent facilities, multiple hostile facilities, and multiple facility locations corresponding to the adjacent facilities and hostile facilities in a first administrative region, as well as a population, a number of households, and a population density value of the first administrative region at the current time, and three rates of change of the population, the number of households, and the population density value.

[0006] This macroeconomic indicators database stores multiple macroeconomic data for a country at the current point in time, including the first administrative region. The storage device stores a generative adversarial network (GAN) algorithm evaluation model. The communication device is used to provide networking functionality.

[0007] The processing device is electrically connected to the storage device and the communication device, and forms an electrical connection with the geographic information database and the macroeconomic index database through the communication device.

[0008] Wherein, when the processing device needs to evaluate the appraised price of a real estate property located in the first administrative region, the processing device calculates the number of adjacent facilities and the number of dislike facilities within a predetermined distance from the address of the real estate property and the location of the facilities in the geographic information database, and obtains the population, the number of households, the population density value and the rate of change of the first administrative region, and obtains the overall economic data from the overall economic indicators database, and inputs the multiple property information of the real estate property, the number of adjacent facilities, the number of dislike facilities, the population, the number of households, the population density value, the rate of change and the overall economic data as input data into the generative adversarial network algorithm appraisal model to generate the appraised price of the real estate property.

[0009] In some implementations, the adjacent facilities include a public transport stop, a convenience store or supermarket, and a school, while the hostile facilities include a power tower or substation, a landfill or incinerator, a fault zone or soil liquefaction zone, and the predetermined distance is one kilometer.

[0010] In other implementations, such macroeconomic data include a central bank interest rate, a stock price index, a positive and negative sentiment index of online public opinion, the number of regional sales transactions in the first administrative region, and a housing price trend index of the first administrative region.

[0011] In some other embodiments, the property information includes an age of the building, a title area, a total number of floors, a transfer floor, a type of housing, a layout, a number of parking spaces, and a type of parking space.

[0012] In some other embodiments, the real estate appraisal system based on generative adversarial network algorithms is also applicable to a real estate transaction database, which is electrically connected to the communication device and stores multiple historical real estate transaction records. The geographic information database also stores the adjacent facilities, the allegedly hostile facilities, and the locations and establishment times of the facilities corresponding to the adjacent and allegedly hostile facilities in a second administrative region, as well as the population, number of households, and population density values ​​of the second administrative region at different times. The country includes the second administrative region, and the overall economic indicators database also stores the overall economic data of the country at different times. The processing device retrieves historical real estate transaction data for the second administrative region from the real estate transaction database. Based on multiple transaction times and addresses, the locations and establishment times of adjacent facilities, the locations and establishment times of problematic facilities, population figures, household numbers, population density values, and change rates at different time points, and overall economic data at different time points, the device performs a first preprocessing step on the property information from the historical real estate transaction data to obtain multiple training datasets. The processing device uses these training datasets to train the generative adversarial network (GAN) algorithm's valuation model.

[0013] In some embodiments, in the first preprocessing, the processing device, for each piece of historical real estate transaction data, determines the number of adjacent facilities and the number of undesirable facilities that are within a predetermined distance from the address of the historical real estate transaction data at the time of the transaction and whose establishment time is earlier than the time of the transaction. It also determines the population, the number of households, the population density value, the rate of change, and the overall economic data at the time of the transaction. The property information, the number of adjacent facilities, the number of undesirable facilities, the population, the number of households, the population density value, the rate of change, and the overall economic data of the historical real estate transaction data are used as one of the training data.

[0014] In some implementations, the second administrative region is greater than or equal to the first administrative region.

[0015] In some embodiments, the processing device further performs a second preprocessing on the historical real estate transaction data of the second administrative region to delete transactions belonging to an abnormal transaction, and then performs the first preprocessing accordingly. In the second preprocessing, when the processing device determines that a remarks column of any historical real estate transaction data includes any one of a plurality of first predetermined keywords, or a transaction subject of the historical real estate transaction data does not include a second predetermined keyword, or a primary use of the historical real estate transaction data does not include a third predetermined keyword, or the transaction corresponding to the historical real estate transaction data belongs to a multi-household or multi-floor transaction, it determines that it belongs to the abnormal transaction.

[0016] In some implementation formats, the first pre-selected keywords include "relatives and friends," "urgent buy and sell," "defects," "debts," "transaction price includes other sales and prices," "second-degree relatives," "customs and traditions," "land clearing," "no inheritance," "developer and landowner transaction," "house condition," "relatives," "market stall," "free of charge," "contract termination and resale," "negotiated purchase," "employees," "co-owners and sales," "co-owners and transactions," "shareholders," "unexecuted, cancelled, or abandoned decorations," "decorations," and "parking space transaction only." The second pre-selected keyword is "real estate." The third pre-selected keyword is "residential."

[0017] The advantage of the present invention is that by using the pre-trained generative adversarial network algorithm appraisal model that utilizes two deep learning models, the processing device integrates the object information of the real estate, the multiple geospatial related information obtained from the geographic information database, and the macroeconomic data obtained from the macroeconomic index database into the input data of the model, thereby obtaining a more accurate appraisal price.

Implementation Method

[0019] Before the present invention is described in detail, it should be noted that similar elements are represented by the same numbers in the following description.

[0020] Referring to Figure 1, an embodiment of the real estate appraisal system based on generative adversarial network algorithm of the present invention is applicable to a real estate registration database 6 and belongs to a financial institution, and includes a geographic information database 4, a macroeconomic indicator database 5, a storage device 2, a communication device 3, and a processing device 1.

[0021] The real estate transaction price database 6 is, for example, a server established by the financial institution and stores multiple historical real estate transaction records. Each historical real estate transaction record includes multiple property information, such as a building age, a title area, a total number of floors, a transferred floor, a housing type, a layout, a number of parking spaces, a parking space type, a transaction time, and an address. It should also be noted that in other embodiments, the real estate transaction price database 6 may also be established or provided by the government or other institutions.

[0022] The geographic information database 4 is, for example, a server established by the financial institution, and stores multiple welcoming facilities, multiple suspected facilities, and multiple facility locations and establishment times corresponding to the welcoming facilities and suspected facilities in a second administrative region, as well as multiple population numbers, multiple household numbers, and multiple population density values ​​of the second administrative region at different times. In this embodiment, the second administrative region is, for example, Taipei City. The welcoming facilities include a public transportation station (such as a bus stop, train station, or MRT station), a convenience store or hypermarket, and a school. The suspected facilities include a power tower or substation, a garbage dump or incinerator, a fault zone or soil liquefaction zone. However, it should be emphasized that the welcoming facilities and suspected facilities are not limited to the examples mentioned above.

[0023] The overall economic indicators database 5 is, for example, one or more servers, and stores multiple overall economic data for a country at different points in time, provided by at least one institution. The country is, for example, the home country, and includes the second administrative region. Such overall economic data includes a central bank interest rate, a stock price index (such as the Taiwan Stock Exchange Weighted Index), a positive and negative sentiment index for online public opinion, a regional transaction volume for the second administrative region, and a housing price trend index for the second administrative region.

[0024] The communication device 3 supports wired network technology (such as Ethernet) or wireless network technology (such as Wi-Fi), and is, for example, a network controller (such as a chip) or other network device, and is used to provide networking functionality. The processing device 1 is, for example, one or more central processing units, and is electrically connected to the storage device 2 and the communication device 3 to form an electrical connection with the geographic information database 4, the macroeconomic indicators database 5, and the real estate price registration database 6 through the communication device 3.

[0025] The storage device 2 stores a generative adversarial network (GAN) algorithm evaluation model, which is obtained by using the GAN algorithm and training it automatically and cyclically through a generator and a discriminator, i.e., two deep learning models.

[0026] More specifically, the processing device 1 obtains the historical real estate transaction data of the second administrative region from the real estate transaction database 6, and performs a first preprocessing on the property information of the historical real estate transaction data based on the transaction time and address of the historical real estate transaction data, the location and establishment time of the adjacent facilities and the location and establishment time of the disliked facilities stored in the geographic information database 4, and the population, the number of households, the population density value, and multiple change rates at different time points, and the overall economic data at different time points, to obtain multiple training data. These change rates are the percentage changes in the population, the number of households, and the population density over a predetermined period of time. In this embodiment, the predetermined period is one month, such as the percentage change in the population between February and January. These rates are calculated by the processing device 1 based on the population, the number of households, and the population density values ​​at different times, or are pre-calculated and stored in the geographic information database 4. In other embodiments, the predetermined period can be 10 days, a quarter, or other values.

[0027] In the first preprocessing, for each piece of historical real estate transaction data, the processing device 1 determines the number of adjacent facilities and the number of undesirable facilities that are within a predetermined distance from the address of the historical real estate transaction data at the time of the transaction and whose establishment time is earlier than the time of the transaction. It also determines the population, the number of households, the population density value, the rate of change, and the overall economic data at the time of the transaction. The property information, the number of adjacent facilities, the number of undesirable facilities, the population, the number of households, the population density value, the rate of change, and the overall economic data of the historical real estate transaction data are used as one of the training data. For example, if the predetermined distance is one kilometer, the information about the objects in the training data may include, for example, 10 (indicating the building's age of 10 years), 50.18 (indicating the area of ​​the title deed in square meters), 24 (indicating the total number of floors), 20 (indicating the floor to be transferred), 1 (indicating the building type is an elevator building, 2 for a mansion, 3 for an apartment, and 4 for a detached house), 3 (indicating the number of rooms in the layout), 2 (indicating the number of living rooms in the layout), 2 (indicating the number of bathrooms in the layout), 1 (indicating the number of parking spaces), and 1 (indicating the parking space type is a ramp-level surface, 2 for a ramp-mechanical area).

[0028] The processing device 1 then uses the training data to train the evaluation model of the generative adversarial network algorithm. In this embodiment, in addition to using various known parameters, such as the number of hidden layers, the number of neurons in each layer, the activation function, the learning rate, the optimizer, and the loss function of the two deep learning models (generator and discriminator), the generative adversarial network algorithm also uses normally distributed random noise instead of the known uniformly distributed random noise.

[0029] In other embodiments, the processing device 1 may first perform a second preprocessing on the historical real estate transaction data of the second administrative region to delete those belonging to an abnormal transaction, and then perform the first preprocessing accordingly. In the second preprocessing, when the processing device 1 determines that a remarks column of any historical real estate transaction data includes any one of a plurality of first predetermined keywords, it determines that it belongs to the abnormal transaction. The first predetermined keywords include "relatives and friends", "urgent buy and sell", "defects", "debts", "transaction price includes other sales + price", "second degree of kinship", "local customs", "land clearing", "no inheritance", "developer and landowner + transaction", "house condition", "relatives", "market stall", "free", "contract termination + resale", "negotiated purchase", "employee", "co-owner + sale", "co-owner + transaction", "shareholder", "unexecuted, cancelled, abandoned decoration", "decoration", and "parking space transaction only". Alternatively, when the processing device 1 determines that a transaction subject of any historical real estate transaction data does not include a second predetermined keyword, it determines that it belongs to the abnormal transaction. The second predetermined keyword is "real estate", that is, excluding transactions that only involve land or buildings. Alternatively, when the processing device 1 determines that a primary use of any historical real estate transaction data does not include a third predetermined keyword, it determines that it belongs to the abnormal transaction. The third predefined keyword is "residential," which means it will filter out purely residential or mixed-use residential and commercial properties. Alternatively, when the processing device 1 determines that any transaction corresponding to the historical real estate transaction data belongs to a multi-unit or multi-floor transaction, it will be judged as an abnormal transaction.

[0030] When the processing device 1 needs to evaluate the appraised price of a real estate property located in a first administrative region, the processing device 1 calculates the number of adjacent facilities and the number of suspected facilities within a predetermined distance of the address from the address by the geographic information database 4 based on the address of the real estate property and the location of the adjacent facilities and the suspected facilities. It also obtains the population, number of households, population density, and rate of change of the population, number of households, and population density of the first administrative region at that time. The processing device 1 obtains the current total economic data from the total economic indicators database 5. The processing device 1 inputs the property information, the number of adjacent facilities, the number of suspected facilities, the population, the number of households, the population density, the rate of change, and the total economic data as input data into the generative adversarial network algorithm appraisal model to generate the appraised price of the real estate property. In this embodiment, the second administrative region includes the first administrative region, such as Taipei City including Da'an District. In other embodiments, the first administrative region may also be equal to the second administrative region.

[0031] In summary, by using the pre-trained generative adversarial network (GAN) algorithm valuation model that utilizes two deep learning models, the processing device 1 integrates the property information, multiple geospatial related information obtained from the geographic information database 4, and the overall economic data obtained from the geographic information database 4 into the model's input data, thereby obtaining a more accurate valuation price. By using the GAN algorithm valuation model to automatically and iteratively train two deep learning models—a generator and a discriminator—it can generate more realistic property prices compared to conventional models using only a single deep learning model, thus achieving the goal of accurate valuation. Therefore, the objectives of this invention are indeed achieved.

[0032] However, the above description is only an embodiment of the present invention and should not be construed as limiting the scope of the present invention. Any simple equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the patent specification of the present invention shall still fall within the scope of the patent of the present invention. [Simplified Explanation of the Diagram]

[0018] Other features and effects of the present invention will be clearly presented in the embodiments with reference to the figures, wherein: Figure 1 is a block diagram illustrating an embodiment of the real estate appraisal system based on the generative adversarial network algorithm of the present invention.

Claims

1. A real estate appraisal system based on generative adversarial network algorithm, applicable to a real estate transaction database, the real estate transaction database storing multiple historical real estate transaction data, the real estate appraisal system based on generative adversarial network algorithm comprising: a geographic information database storing multiple adjacent facilities, multiple disadvantaged facilities, and multiple facility locations corresponding to the adjacent facilities and disadvantaged facilities in a first administrative region, and a population, a household, and a population density value of the first administrative region at the current time, and three rates of change of the population, the household, and the population density value, as well as the locations and multiple establishment times of the adjacent facilities, disadvantaged facilities, and facilities corresponding to the adjacent facilities and disadvantaged facilities in a second administrative region, and the population, the household, and the population density values ​​of the second administrative region at different times; A database of general economic indicators, storing multiple general economic data of a country at the current point in time and the same general economic data of the country at different points in time, the country including the first administrative region and the second administrative region; a storage device, storing a generative adversarial network (GAN) algorithm evaluation model; A communication device is used to provide networking functionality; and a processing device is electrically connected to the storage device and the communication device, and forms an electrical connection with the geographic information database, the macroeconomic index database, and the real estate transaction price database through the communication device. The processing device retrieves historical real estate transaction data for the second administrative region from the real estate transaction price database. Based on multiple transaction times and addresses, the locations and establishment times of adjacent facilities, the locations and establishment times of problematic facilities, population numbers, household numbers, population density values, and change rates at different time points, and macroeconomic data at different time points, the processing device performs a first preprocessing on multiple item information from the historical real estate transaction data to obtain multiple training datasets. The processing device uses these training datasets to train the generative adversarial network (GAN) algorithm valuation model. When the processing device needs to evaluate the appraised price of a real estate property located in the first administrative region, the processing device consults the geographic information database to calculate the number of adjacent facilities and the number of undesirable facilities within a predetermined distance of the address of the real estate property, and obtains the population, number of households, population density, and rate of change of the first administrative region. It also obtains the overall economic data from the overall economic indicators database. The processing device inputs the property information, the number of adjacent facilities, the number of undesirable facilities, the population, the number of households, the population density, the rate of change, and the overall economic data as input data into the generative adversarial network algorithm appraisal model to generate the appraised price of the real estate property.

2. The real estate appraisal system based on generative adversarial network algorithm as described in claim 1, wherein, The adjacent facilities include a public transport stop, a convenience store or supermarket, and a school, while the alleged facilities include an electric tower or substation, a landfill or incinerator, a fault zone or soil liquefaction zone, and the planned distance is one kilometer.

3. The real estate appraisal system based on generative adversarial network algorithm as described in claim 1, wherein, These macroeconomic data include a central bank interest rate, a stock price index, an online sentiment index (positive and negative), the number of property transactions in the first administrative region, and a housing price trend index for the first administrative region.

4. The real estate appraisal system based on generative adversarial network algorithm as described in claim 1, wherein, The property information includes an age of the building, a title area, a total number of floors, a transfer floor, a housing type, a layout, a number of parking spaces, and a parking space type.

5. The real estate appraisal system based on generative adversarial network algorithm as described in claim 1, wherein, In the first preprocessing, for each piece of historical real estate transaction data, the processing device determines the number of adjacent facilities and the number of undesirable facilities that are within a predetermined distance from the address of the historical real estate transaction data at the time of the transaction and were established earlier than the time of the transaction. It also determines the population, number of households, population density, rate of change, and overall economic data at the time of the transaction. The property information, number of adjacent facilities, number of undesirable facilities, population, number of households, population density, rate of change, and overall economic data of the historical real estate transaction data are used as one of the training data.

6. The real estate appraisal system based on generative adversarial network algorithm as described in claim 5, wherein, The second administrative region is greater than or equal to the first administrative region.

7. The real estate appraisal system based on generative adversarial network algorithm as described in claim 6, wherein, The processing device first performs a second preprocessing on the historical real estate transaction data of the second administrative region to delete those belonging to an abnormal transaction, and then performs the first preprocessing accordingly. In the second preprocessing, when the processing device determines that a remarks column of any historical real estate transaction data includes any one of a plurality of first predetermined keywords, or a transaction object of the historical real estate transaction data does not include a second predetermined keyword, or a primary use of the historical real estate transaction data does not include a third predetermined keyword, or the transaction corresponding to the historical real estate transaction data belongs to a multi-household or multi-floor transaction, it determines that it belongs to the abnormal transaction.

8. The real estate appraisal system based on generative adversarial network algorithm as described in claim 7, wherein, The first reservation keywords include "relatives and friends", "urgent buy and sell", "defects", "debts", "transaction price includes other sales + price", "second degree of kinship", "local customs", "land clearing", "no inheritance", "developer and landowner + transaction", "house condition", "relatives", "market stall", "free", "contract termination + resale", "negotiated purchase", "employees", "co-owners + sale", "co-owners + transaction", "shareholders", "unexecuted, cancelled, abandoned decoration", "decoration", and "parking space transaction only". The second reservation keyword is "real estate" and the third reservation keyword is "living".

Images