JP7162087B2 - Property information providing device, property information providing method, and property information providing program - Google Patents

Description

The present invention relates to a method of providing information on residential and architectural properties.

2. Description of the Related Art Conventionally, real estate information providing systems are known that provide information on real estate through websites and applications. In such a system, it is common for a user to first input the conditions of a desired property. Examples of these conditions include sales price or rent, type of floor plan (e.g., 2LDK, 3LDK, etc.), building type, age of building, exclusive area, area, railway line, travel time to station, presence or absence of specific facilities, etc. mentioned. Then, the real estate information providing system retrieves and provides property information that matches the input conditions. The provided property information includes, for example, various properties of the property and a floor plan. For example, Patent Literature 1 discloses a real estate information search device that acquires real estate information based on search conditions such as price, land area, building area, building age, and number of rooms.

JP 2015-129980 A

However, even if property information that matches the search conditions specified by the user is presented, there are cases where the interior state of the property indicated by the floor plan included in the property information, such as the floor plan, does not match the user's wishes. . The reason for this is that the floor plan may show various features that are not specified by the above-described search conditions. Therefore, the user is required to search for property information that meets the user's desires from among a plurality of pieces of property information that match the specified search conditions while looking at the floor plan. On the other hand, even if the user has a latent desire regarding the state of the interior of the property, the user may not recognize it as a clear condition. Therefore, even if a search condition that can be specified regarding the state of the interior of the property is newly established, it may be difficult for the user to determine whether or not the condition should be specified.

The present invention has been made in view of the above points, and one of its problems is that it is possible to provide information on a property that the user prefers without the user having to input the conditions of the property ascertained from the floor plan. To provide a property information providing device, a property information providing method, and a property information providing program.

In order to solve the above-mentioned problems, the invention according to claim 1 provides display control means for displaying a floor plan image for each of a plurality of properties, and for each of the plurality of properties, a evaluation information acquisition means for acquiring evaluation information indicating an evaluation from a user; and feature information generated based on the floor plan image for each of the plurality of properties, the feature information indicating the features of the rooms in the property. feature information acquisition means for acquiring feature information; generation means for generating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information; determination means for determining the property to be presented to the user based on the evaluation information; a first probability distribution in which the feature information obtained for a room in a property, among the plurality of properties, for which the obtained evaluation information indicates the positive evaluation is a random variable; generating the model representing a second probability distribution with the characteristic information obtained for a room in the property indicating the negative evaluation among the plurality of properties, with the characteristic information obtained as a random variable; The determining means acquires candidate property feature information generated based on a floor plan image of a candidate property to be presented to the user, and which indicates features of rooms in the candidate property. , from the first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model, in the second probability distribution indicated by the generated model, the obtained The candidate property is determined as the property to be presented to the user when a probability density difference obtained by subtracting the second probability density of the candidate property characteristic information is equal to or greater than a predetermined value .

According to this invention, the evaluation information indicating the evaluation given by the user to each of the displayed floor plan images is acquired. Further, feature information generated based on the floor plan image is acquired as information indicating features of rooms in the property. Then, based on the evaluation information and the feature information, a model for identifying properties preferred by the user is generated. As a result, a model is generated that learns, for example, what characteristics of a property having rooms are highly evaluated by users. Properties to be presented are determined based on the generated model. Therefore, even if the user does not input the conditions of the property grasped from the floor plan, it is possible to provide the information of the property that the user prefers.

In the invention according to claim 2, the display control means displays the floor plan image on the screen of a display device having a display part having a screen and a detection part for detecting a contact position on the screen. and the evaluation information obtaining means obtains the evaluation information indicating a predetermined evaluation when movement of the contact position in a predetermined direction is detected on the screen on which the floor plan image is displayed. characterized by

According to this invention, the user touches the screen with a finger or the like and moves it in a predetermined direction, thereby acquiring evaluation information indicating a predetermined evaluation. Therefore, an evaluation can be input by a simple operation.

The invention according to claim 3 is characterized in that the generating means generates the model for each user.

According to the present invention, personalized property information can be provided.

The invention according to claim 4 is characterized in that the determining means determines the property to be presented to the user based on the model generated for another user having the same attribute as the user's attribute. .

According to the present invention, models generated for other users having the same attributes as those of the user to whom the property is presented are used to determine the property to be presented. Therefore, even if a model with appropriate accuracy has not been generated for the user to whom the property is presented, it is possible to appropriately determine the property to be presented to the user.

In the invention according to claim 5, the evaluation information acquiring means provides the user with respect to the portion specified by the user in the displayed floor plan image for at least one partially evaluated property among the plurality of properties. the generating means uses the obtained partial evaluation information for a first room corresponding to the specified portion among the rooms in the partially evaluated property, The model is generated using the obtained evaluation information for a second room, which is different from the first room, among the rooms in the partially evaluated property.

According to the present invention, of the rooms in the property of the floor plan image for which the user has designated a portion and input the evaluation, for the room corresponding to the designated portion, the model is generated. Partial evaluation information that indicates the evaluation is used. On the other hand, evaluation information is used for model generation for rooms other than the room corresponding to the designated portion in the property. Therefore, even if the property as a whole can be given a specific rating, if the user wants to give a different rating to some of the rooms in the property, each rating can be assigned to the model generation. can be reflected in

According to a sixth aspect of the present invention, the display control means controls at least one of the plurality of properties on which the floor plan images are displayed to display a predetermined number of properties after the floor plan images are displayed. After the number of days has passed, the floor plan image is re-displayed, and the evaluation information acquisition means further acquires re-evaluation information indicating re-evaluation by the user of the re-displayed floor plan image, The generation means generates the model using the acquired re-evaluation information for the re-displayed property among the plurality of properties and using the acquired evaluation information for a property different from the re-displayed property. characterized by

According to this invention, after a predetermined number of days have passed, the floor plan image of any of the once-evaluated properties is re-displayed and re-evaluated. Then, for the re-evaluated property, a model is generated using the re-evaluation information. Therefore, it is possible to generate a model that reflects the user's preferences, which change with the passage of time or by looking at information on various properties.

In the invention according to claim 7, the characteristic information acquiring means identifies an area corresponding to the room from the floor plan image, and identifies the characteristics of the room based on the identified area. is characterized by obtaining

The invention according to claim 8 is characterized in that the feature information acquiring means acquires the feature of the room based on the feature of the specified area.

According to this invention, it is possible to reflect in the model the evaluation of the visual characteristics of the room obtained from the floor plan image of the room in the property.

According to a ninth aspect of the invention, the feature information acquiring means includes: the area of the room; characterized in that the feature information indicating at least one of the above is acquired.

According to the present invention, it is possible to provide property information according to the user's preference regarding the features of a room that can be specified from a floor plan.

In the tenth aspect of the invention, the model includes a plurality of types corresponding models respectively corresponding to a plurality of types of rooms, and the feature information acquiring means acquires the floor plan image for each of the rooms included in the property. the type of the room is specified from the room type, the feature information is acquired for the room for which the type is specified, and the generation means acquires the characteristic information for each type of room included in the plurality of types of rooms, the first probability distribution with the characteristic information acquired for the type of room in the property indicating the positive evaluation among the properties including the room of the type in the property being the positive evaluation; and and the feature information acquired for the room of the type in the property indicating the negative evaluation for the acquired evaluation information among the properties including the room of the type among the plurality of properties as a random variable. and the type-corresponding model representing the second probability distribution, and the determining means calculates the probability density difference for each of the plurality of type-corresponding models, and calculates the probability density difference for each of the plurality of type-corresponding models calculating the probability density difference for the candidate property based on the plurality of probability density differences, and determining whether to present the candidate property to the user based on the probability density difference for the candidate property; It is characterized by determining .

According to the present invention, the type of room in the property is specified from the floor plan image, and the characteristic information indicating the characteristics of the room and the acquired evaluation information are used to generate a type correspondence model corresponding to the type of the room. is generated. By generating a plurality of type-corresponding models, a model for identifying properties preferred by the user is generated. Therefore, the user's preferences can be reflected in the model for each room type.

According to an eleventh aspect of the invention, there is provided a property information providing method executed by a computer, comprising: a display control step of displaying a floor plan image of each of a plurality of properties; an evaluation information acquiring step of acquiring evaluation information indicating an evaluation of a user for a drawing image; a feature information acquisition step of acquiring feature information indicating a feature; a generation step of generating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information; a determination step of determining the property to be presented to the user based on the model , wherein the evaluation information obtaining step indicates any one of a plurality of evaluations including positive evaluations and negative evaluations. The evaluation information is acquired, and in the generating step, the feature information acquired for a room in the property indicating the positive evaluation among the plurality of properties is a first property information acquired as a random variable. a probability distribution; and a second probability distribution in which the characteristic information obtained for a room in the property indicating the negative evaluation among the plurality of properties is a random variable. and the determining step includes candidate property feature information generated based on a floor plan image of a candidate property of the property to be presented to the user, the candidate property feature information indicating a feature of a room in the candidate property. obtaining information, from a first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model, in the second probability distribution indicated by the generated model, When a probability density difference obtained by subtracting the second probability density of the obtained candidate property characteristic information is equal to or greater than a predetermined value, the candidate property is determined as a property to be presented to the user .

According to a twelfth aspect of the present invention, a computer is provided with display control means for displaying a floor plan image for each of a plurality of properties, and evaluation by a user of the displayed floor plan image for each of the plurality of properties. and an evaluation information obtaining means for obtaining evaluation information indicating the characteristics of each of the plurality of properties, the feature information being generated based on the floor plan image and indicating the characteristics of the rooms in the property. feature information acquisition means; creation means for creating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information; and a determining means for determining a property to be presented to the property, and the evaluation information acquisition means acquires the evaluation information indicating any one of a plurality of evaluations including positive evaluations and negative evaluations, and The generation means generates a first probability distribution with the characteristic information obtained for a room in the property indicating the positive evaluation among the plurality of properties as a random variable, and the plurality of property and a second probability distribution in which the characteristic information obtained for a room in the property indicating the negative evaluation is a random variable, and the determination means is acquiring candidate property feature information generated based on a floor plan image of a candidate property to be presented to the user, the candidate property feature information indicating characteristics of a room in the candidate property; the obtained candidate property characteristic information in the second probability distribution indicated by the generated model from the first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model is equal to or greater than a predetermined value, the candidate property is determined as the property to be presented to the user .

ADVANTAGE OF THE INVENTION According to this invention, even if a user does not input the conditions of the property grasped|ascertained from a floor plan, the information of the property which the user likes can be provided.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows an example of an outline|summary structure of property information provision system S which concerns on one Embodiment. 1 is a block diagram showing an example of a schematic configuration of an information providing server 1 according to one embodiment; FIG. 4 is a diagram showing an example of information stored in a database of the information providing server 1; FIG. 3 is a diagram showing an example of functional blocks of a system control unit 11 of the information providing server 1; FIG. (a) is a figure which shows the example of a display of a floor plan screen. (b) is a diagram showing a display example of a property details screen. (a) is a diagram showing an input example of a positive evaluation for a floor plan image and an example of evaluation for each room. (b) is a diagram showing an input example of a negative evaluation for a floor plan image and an example of evaluation for each room. FIG. 4 is a diagram showing an input example of partial evaluation and overall evaluation for a floor plan image and an example of evaluation for each room; It is a figure which shows the example of production|generation of segmented floor plan information. It is a figure which shows the structural example of the feature vector of the property|object acquired from the floor-plan image. It is a figure which shows the structural example of a property classifier. It is a figure which shows the calculation example of a characteristic item score. FIG. 10 is a diagram showing an example of the difference between the probability distribution of the feature quantity of the evaluation “like” and the probability distribution of the feature quantity of the evaluation “dislike”; It is a flow chart which shows an example of floor plan analysis processing by system control part 11 of information offer server 1 concerning one embodiment. 4 is a flow chart showing an example of property recommendation processing by the system control unit 11 of the information providing server 1 according to one embodiment. 6 is a flow chart showing an example of property selection processing by the system control unit 11 of the information providing server 1 according to one embodiment. It is a flow chart which shows an example of floor plan display processing by CPU of user terminal 2 concerning one embodiment. 4 is a flowchart showing an example of learning processing by the system control unit 11 of the information providing server 1 according to one embodiment;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[1. Configuration of Property Information Providing System]
First, the configuration and functional overview of the property information providing system S according to this embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of a schematic configuration of a property information providing system S according to this embodiment.

As shown in FIG. 1 , the property information providing system S includes an information providing server 1 and a plurality of user terminals 2 . The information providing server 1 and each user terminal 2 are connected to each other via a network NW. The network NW is constructed by, for example, the Internet, a dedicated communication line (for example, a CATV (Community Antenna Television) line), a mobile communication network (including base stations, etc.), gateways, and the like.

The information providing server 1 is a server device that distributes information on real estate properties to the user terminals 2 . A property for which information is provided may be, for example, a residence (for example, a house, a room in an apartment, a room in an apartment building, etc.) or an office. Also, the property for which the information is provided may be, for example, any of a new property, a second-hand property, and a rental property. The information providing server 1 may transmit property information in response to a request from the user terminal 2, for example. Property information may include, for example, property name, address, price, exclusive area, floor plan type, and other various information. Furthermore, the property information may include a floor plan image. The floor plan image is an image showing the floor plan of the target property. The information providing server 1 may transmit the property information on a web page, for example.

The information providing server 1 executes a process of recommending a property to a user. For example, the information providing server 1 may provide property information of a plurality of recommended properties when each user uses the property information providing system S for the first time in the day. The information providing server 1 particularly causes the user terminal 2 to display the floor plan image. The information providing server 1 allows the user to input an evaluation of the floor plan image. The information providing server 1 generates a model for identifying a property preferred by the user through learning using the floor plan image and the input evaluation. This model is called a property classifier. Then, the information providing server 1 determines a property to recommend to the user based on the generated property classifier.

Each user terminal 2 is a terminal device that can be carried by a user who can use the property information providing system S. FIG. Examples of the user terminal 2 include mobile information terminals such as smart phones and tablet computers, mobile phones, and personal digital assistants (PDAs). Each user terminal may include a display having a screen and a sensor that detects a contact position on the screen. Touching the screen means bringing the user's finger, stylus pen, or the like into contact with the screen. The contact position is the coordinates of the contact made by the user's finger, stylus pen, or the like on the screen. A touch panel is an example of a device in which such a sensor and a display are integrated. An application dedicated to the property information providing system S may be installed in each user terminal 2 . Using this application, the user can view property information. A web browser may be installed in each user terminal 2 . Property information may be browsed by using a web browser. A stationary terminal device such as a personal computer may be used as the user terminal 2 .

[2. Configuration of information providing server]
Next, the configuration of the information providing server 1 will be described with reference to FIGS. 2 and 3. FIG. FIG. 2 is a block diagram showing an example of a schematic configuration of the information providing server 1 according to this embodiment. As shown in FIG. 2 , the information providing server 1 includes a system control section 11 , a system bus 12 , an input/output interface 13 , a storage section 14 and a communication section 15 . The system control unit 11 and the input/output interface 13 are connected via a system bus 12 .

The system control unit 11 includes a CPU (Central Processing Unit) 11a, a ROM (Read Only Memory) 11b, a RAM (Random Access Memory) 11c, and the like.

The input/output interface 13 performs interface processing between the storage unit 14 and the communication unit 15 and the system control unit 11 .

The storage unit 14 is configured by, for example, a hard disk drive or the like. The storage unit 14 stores databases such as a member DB 14a, a property DB 14b, evaluation information 14c, and a model DB 14d. "DB" is an abbreviation for database.

FIG. 3 is a diagram showing an example of information stored in the database of the information providing server 1. As shown in FIG. In the member DB 14a, member information regarding users who can use the property information providing system S is stored for each user. For example, the member DB 14a stores user ID, name, gender, date of birth, address, telephone number, e-mail address, and other user attributes in association with each other as member information. A user ID is identification information that identifies a user.

The property DB 14b stores property information relating to properties to which information is provided for each combination of a property and a real estate agency that brokers or sells the property. Specifically, the property DB 14b stores, as property information, a property ID, a trader ID, a registration date, a property name, detailed information, a property image, a floor plan image, and the like. The property ID is identification information that identifies the combination of the property and the real estate agent. The trader ID is identification information that identifies a real estate trader who brokers or sells the target property. The registration date indicates the date when the property information was registered in the property information providing system S. FIG. The detailed information is detailed information about the target property. The detailed information includes attributes of the target property. Specifically, the detailed information may include the address, nearest station, travel time to the nearest station, building type, age of building, floor plan type, price, occupied area, floor on which the property is located, and other attributes. The building type indicates the type of building in which the property of interest or the property is located (eg, house, condominium, apartment, etc.). The floor plan type indicates the type and number of rooms that the property has. Examples of floor plan types include "1R", "1K", "1LDK", "2K", "2DK", "2LDK", "3LDK", and "3SLDK". The price indicates the sales price or rent. Examples of other attributes include management costs, security deposits, key money, whether the bath and toilet are separate, whether there is flooring, whether there is a washing machine storage area, whether there is a system kitchen, the type of stove, whether there is a bathroom dryer, Whether or not reheating of the bath is possible, whether or not there is a separate washroom, whether or not there is a warm water washing toilet seat, whether or not there is an air conditioner, whether or not there is floor heating, whether or not there is a walk-in closet, whether it is a corner room, whether it is an auto-lock. Presence/absence of a parking lot/presence/absence of a delivery box/whether pets are allowed/whether musical instruments are allowed/whether foreigners can move in/private lodging is allowed whether or not, the distance to a convenience store, the distance to a supermarket, and the like. The property image is a photographic image of the interior or exterior of the property.

Further, segmented floor plan information may be stored in the property DB 14b in association with the property ID. The segmented floor plan information is information obtained from the floor plan image of the target property. The segmented floor plan information is information indicating the position and range of each element that constitutes the target floor plan image. The segmented floor plan information may be, for example, information in which each pixel of the floor plan image is associated with a label indicating the element corresponding to the position of the pixel. The components of the floor plan image are divided into the interior and exterior of the property. Components inside the property may be broadly divided into partitions and internal spaces. A partition is something that divides the interior space of a property or something that separates the interior and exterior of a property. The partition may divide the space physically or may divide the space conceptually. Examples of partitions include walls, openings, and the like. Examples of openings include doorways and windows. The interior space may be an empty space surrounded by partitions. The internal space may be broadly divided into rooms and other spaces. In the present embodiment, the term "room" does not refer to a room in an apartment or a room in an apartment. It may be a space that can be classified according to usage. Examples of rooms include a general room, a living room, a dining room, a kitchen, a dining kitchen, an LDK (a room in which a living room, a dining room, and a kitchen are integrated), a closet, a loft, and the like. A general room may be a room that does not correspond to any of the living room, dining room, kitchen, dining kitchen, LDK, closet, and loft. A general room may be a room for which no particular purpose is specified. A common room is called a Japanese-style room, a Western-style room, or a bedroom. Examples of other spaces include washrooms, toilets, baths, corridors, entrances, balconies, and the like. At least one of the washroom, toilet, and bath may be included in the room.

Furthermore, the property DB 14b may store feature vectors in association with property IDs. A feature vector may be a two-dimensional array of features of rooms in the property of interest. For example, the feature vector may be information in which a one-dimensional array of feature values of each room in the target property is arranged. The one-dimensional array of features of each room may be associated with, for example, a label indicating the room type of the room and the number of the room. The room type indicates the type of room. For example, room types may include general room, living room, dining room, kitchen, dining kitchen, LDK, closet, loft, and the like. Room numbers may be assigned in order from 1 to the rooms in the property.

The evaluation information DB 14c stores evaluation information indicating the user's evaluation of the floor plan image displayed on the user terminal 2 each time an evaluation is performed. Specifically, in the evaluation information DB 14c, a user ID, a property ID, a display date, and an overall evaluation are stored in association with each other as evaluation information. User ID indicates the user who performed the evaluation. The display date indicates the date on which the floor plan image of the evaluated property was displayed. The date on which the subject's floor plan image was displayed and the date on which the evaluation was performed usually coincide. The overall evaluation is an evaluation for all rooms in the property indicated by the floor plan image. Either one of “like” and “dislike” may be set as the overall evaluation. A "like" is a positive rating for the subject's floor plan image. "Dislike" is a negative rating for the subject's floor plan image. One or a plurality of pieces of partial evaluation information may be further stored in the evaluation information DB 14c. The partial evaluation information is information regarding the evaluation of one of the rooms in the property indicated by the target floor plan image. Partial evaluation information is stored when the user designates and evaluates a part of the target floor plan image. The partial evaluation information includes the target room number, room type and partial evaluation. The target room number is the number of the room designated by the user. The room type indicates the type of room designated by the user. The partial rating indicates the user's rating for the designated room. As with the overall evaluation, either one of "like" and "dislike" may be set as the partial evaluation. Further, the evaluation information DB 14c may store responses to questionnaires. The questionnaire response is information indicating which part of the target property the user has input the evaluation of "dislike" as a response from the user when the overall evaluation is "dislike".

The model DB 14d stores, for each user, parameters of a property classifier that identifies a user's favorite property. Specifically, the model DB 14d stores user IDs and property classifier parameters in association with each other. User ID indicates the user whose preferences are identified by the property classifier. For example, when using a probability density function as a model, the parameters may include the coefficients of that function. For example, for a normal distribution, parameters may include mean and variance. In the case of a mixed normal distribution, the parameters may include the mean, variance and weight of each normal distribution that constitutes the mixed normal distribution. If a neural network is used as the model, the parameters may include the weights of the hidden layers included in the network.

The storage unit 14 further stores various programs such as an operating system, a DBMS (Database Management System), and a server program. The server program is a program that causes the system control unit 11 to execute processing related to the provision of property information, such as provision of floor plan images, acquisition of evaluations from users, generation of property classifiers, determination of properties recommended to users, and the like. is. For example, the server program may be obtained from another device via the Internet NW, or may be recorded on a recording medium such as a magnetic tape, an optical disk, or a memory card and read via a drive device. good too.

The communication unit 15 is configured by, for example, a network interface card or the like. The communication unit 15 is connected to the user terminal 2 and the like via the network NW, and controls the state of communication with these devices.

[3. Functional overview]
Next, an overview of the functions of the system control unit 11 in the information providing server 1 will be described with reference to FIGS. 4 to 12. FIG. FIG. 4 is a diagram showing an example of functional blocks of the system control unit 11 of the information providing server 1. As shown in FIG. The CPU 11a reads and executes various program codes included in the server program, so that the system control unit 11, as shown in FIG. 114, determination unit 115, and the like.

[3-1. Display of floor plan image]
The display control unit 111 causes the user terminal 2 to display a floor plan image of each of the plurality of properties. The property for which the floor plan image is displayed may be a property recommended to the target user. Recommended properties are determined using a property classifier, as described below. However, when no property classifier has been created, the recommended property may be any property. For example, a property may be recommended for which the number of days elapsed from the date of registration of the property is less than a predetermined value. Alternatively, the user may input the property search conditions in advance and save them. As a search condition, the same information as the information included in the detailed information of the property information may be designated. For example, the search conditions may be stored in the member DB 14a or the like. Properties that match the saved search criteria may then be recommended. Further, the display control unit 111 controls at least one property among a plurality of properties whose floor plan images are displayed once for each predetermined number of days (for example, one week, one month, etc.) after the floor plan image is displayed. ) has elapsed, the floor plan image may be redisplayed. While searching for a property, the user may change his/her property preferences as time passes or as he/she looks at various property information. Therefore, the evaluation once input by the user may change thereafter. The floor plan image may be redisplayed to reflect changes in user preferences in the property classifier. Here, the display control unit 111 may re-display the floor plan image only for the property for which the user has input a positive evaluation of “like”, which will be described later.

The display control unit 111 may cause the user terminal 2 to display the floor plan image, for example, when the target user logs into the property information providing system S for the first time on that day. In addition, the display control unit 111 may cause the user terminal 2 to display property information such as detailed information indicating properties of the property. For example, the display control unit 111 may transmit the property name, detailed information, property image, and floor plan image of the recommended property to the user terminal 2 . The display control unit 111 may further transmit segmented floor plan information to the user terminal 2 .

The user terminal 2 displays the floor plan image received from the information providing server 1 . FIG. 5A is a diagram showing a display example of the floor plan screen. As shown in FIG. 5( a ), the user terminal 2 displays a floor plan screen 100 . The floor plan screen 100 includes a floor plan image 110 . FIG. 5B is a diagram showing a display example of the property details screen. When the user performs a predetermined operation while the floor plan image 110 is being displayed, the user terminal 2 displays the property details screen 200 as shown in FIG. 5(b). For example, the property detail screen 200 may be displayed by the user swiping the floor plan screen 100 to the right or left. A swipe may be an operation in which the user presses the screen with a finger and slides the finger. The property details screen 200 is a screen that displays detailed information about the property. For example, the property details screen 200 may display a property image, detailed information, and information about a real estate agent that handles the target property. For example, various information may be displayed by the user scrolling the property details screen 200 up and down. Further, on the property details screen 200, it may be possible to register the target property in the user's favorites, or to request materials or make inquiries to the real estate agent.

[3-2. Input Rating]
The evaluation information acquisition unit 112 acquires evaluation information indicating the user's evaluation of the displayed floor plan image for each of a plurality of properties whose floor plan images are displayed by the user terminal 2 . User-enterable ratings may include at least positive and negative ratings. A positive evaluation may be an evaluation that the user likes the room or floor plan in the property indicated by the floor plan image. A negative evaluation may be an evaluation that the user dislikes the interior or floor plan of the property indicated by the floor plan image. A positive rating is a higher rating than a negative rating. The user may be able to input any evaluation out of three or more levels of evaluation. Even if the user potentially has a desire regarding a room or floor plan within the property, the user may not be aware of it. Alternatively, the user may have the desire in mind but find it difficult to verbalize it. By enabling input of intuitive evaluation, the user does not have to input which points are good or bad for specific reasons.

The evaluation information acquisition unit 112 acquires evaluation information indicating a positive evaluation when a movement in a predetermined first direction of a position touched by a finger or the like is detected on the screen on which the floor plan image is displayed. good too. This allows the user to easily input an evaluation. In addition, an evaluation can be input by an intuitive operation. The operation at this time may be, for example, a flick. A flick may be an operation in which the user flicks the screen with a finger or the like. The difference between the flick and the above-described swipe may be at least one of the moving distance and the length of time from when the finger or the like starts moving until it stops or until the finger or the like is released from the screen. For example, when the movement distance is less than a predetermined distance, the operation may be a flick, and when the movement distance is equal to or greater than the predetermined distance, the operation may be a swipe. Further, when the length of time is less than the predetermined time, the operation may be a flick, and when the length of time is equal to or longer than the predetermined time, the operation may be a swipe. The first direction may be up, down, right, left and diagonal. In this embodiment, rightward movement is a positive evaluation.

In addition, the evaluation information acquisition unit 112 gives a negative evaluation when the position touched by the finger or the like on the screen displaying the floor plan image is detected to move in the second direction different from the first direction. You may acquire the evaluation information shown. The operation at this time may also be a flick. Also, the second direction may be any direction different from the first direction. In this embodiment, leftward movement is a positive evaluation.

The evaluation information acquisition unit 112 may acquire the evaluation by the operation described above as the overall evaluation. That is, the input evaluation is associated with all rooms in the property indicated by the floor plan image.

FIG. 6A is a diagram showing an input example of a positive evaluation for a floor plan image and an example of evaluation for each room. As shown in FIG. 6( a ), the user flicks rightward on the floor plan image 110 . As a result, a positive evaluation "like" is input. The property indicated by the floor plan image 110 has an LDK, a general room A of 7 tatami mats, and a general room B of 5.5 tatami mats. Since the evaluation in this case is the evaluation for the entire floor plan 110, the evaluation “favorite” may be set for all the features of the LDK, the general room A, and the general room B, respectively. FIG. 6B is a diagram showing an input example of a negative evaluation for the floor plan image and an example of evaluation for each room. As shown in FIG. 6B, the user flicks leftward on the floor plan image 110 . As a result, a negative evaluation of "dislike" is entered. In this case, the evaluation “dislike” may be set for all the features of the LDK, the general room A, and the general room B, respectively.

The evaluation information acquisition unit 112 obtains the user's evaluation of the portion specified by the user in the displayed floor plan image for at least one of the plurality of properties whose floor plan images are displayed by the user terminal 2. Partial evaluation information indicating the evaluation may be further acquired. This allows the user to input different evaluations only for some rooms in the property. The user may want to enter a particular rating for the property as a whole, but may want to enter a different rating for some of the properties.

For example, the evaluation information acquisition unit 112 performs an operation in a manner different from the operation in the case of the overall evaluation, and moves the position touched by the finger or the like on the screen on which the floor plan image is displayed in a predetermined third direction. If movement is detected, partial rating information may be obtained indicating a positive rating. The operation in a mode different from the operation in the overall evaluation may be, for example, a movement direction of a finger or the like being different. For example, when a rightward flick results in a positive overall evaluation, an upward flick may result in a positive partial evaluation. Alternatively, when the user performs a flick after touching the screen with a finger or the like and stopping for a predetermined time or longer, the operation may be determined to be an input for partial evaluation. The third direction in this case may be the right direction, for example. This flick is called a charge flick. In contrast to the accumulated flick, a flick performed by the user after touching the screen with a finger or the like without stopping for a predetermined time or longer is called a normal flick. A normal flick is an operation for input for overall evaluation. A room located at the position of the finger or the like when the user starts touching the screen with the finger or the like may be the target of the partial evaluation. For example, segmented floor plan information may be sent to the user terminal 2 along with the property information, as described above. The user terminal 2 may determine where the user has specified by obtaining, from the segmented floor plan information, the label corresponding to the pixel at the position touched by the user in the floor plan image. In addition, the evaluation information acquisition unit 112 performs an operation in a mode different from the operation in the case of the overall evaluation, and the position touched by the finger or the like on the screen on which the floor plan image is displayed is different from the third direction. If movement in a predetermined fourth direction is detected, partial evaluation information indicating a negative evaluation may be obtained. This operation may also be a charge flick. Also, the fourth direction may be the left direction.

FIG. 7 is a diagram showing an input example of partial evaluation and overall evaluation for a floor plan image and an example of evaluation for each room. As shown in FIG. 7, the user puts his finger on the portion 120 on the displayed floor plan image 110 and flicks it rightward. Part 120 is a point on the LDK. After that, the user generally flicked leftward on the floor plan image 110 . In this case, the evaluation “like” may be set for all features of the LDK, and the evaluation “dislike” may be set for each of the features of the general room A and the general room B.

When the user inputs "dislike" as the overall evaluation, the user terminal 2 may display a questionnaire screen. The questionnaire screen is a screen for the user to select what aspects of the target property are disliked. In addition to the information items included in the detailed information (for example, address, nearest station, travel time, building type, building age, floor plan type, price, exclusive area, etc.), there is also "floor plan". “Floor plan” is an option when the user does not like the room or the floor plan in the property indicated by the floor plan image. The evaluation information acquisition unit 112 may acquire the user's selection from the user terminal 2 as a questionnaire answer. This questionnaire is used to determine whether or not to use the property information on which the floor plan image is displayed for model learning.

The evaluation information acquisition unit 112 may acquire the evaluation information when the user performs an operation other than a slide operation such as a flick. For example, the floor plan screen 100 may display elements such as buttons or menus for selecting evaluations. The evaluation information acquisition unit 112 may acquire information or a table indicating evaluations according to the user's operation content for the held element.

[3-3. Feature extraction]
The feature information acquisition unit 113 provides feature information generated based on the floor plan image for each of a plurality of properties whose floor plan images are displayed by the user terminal 2, and indicates the features of the rooms in the property. Get feature information. This feature may be a per-room feature. Also, the features to be acquired are not particularly limited as long as they are features that can be specified from the floor plan image. The feature to be acquired may be a feature that cannot or is difficult to specify from the detailed information of the property information. The acquired feature information may be a feature amount.

The feature information acquisition unit 113 may specify the area corresponding to the room from the floor plan image. Further, the room type may be specified from the floor plan image for each room included in the target property. The feature information acquisition unit 113 may generate segmented floor plan information from the floor plan image. For example, the feature information acquisition unit 113 may generate segmented floor plan information using semantic segmentation.

For example, the feature information acquisition unit 113 may generate segmented floor plan information using machine learning. For example, a convolutional neural network (CNN), which can be used for semantic segmentation, may be used as a model for generating segmented floor plan information. Inputting a floor plan image into this model will output segmented floor plan information from this model. For example, in the learning phase, training data may be provided with floor plan images and segmented floor plan information. The segmented floor plan information in this case may be created manually, for example. As preprocessing, the feature information acquisition unit 113 recognizes characters indicating the area of an internal space such as a room from the floor plan image by, for example, character recognition. For example, characters such as "7 tatami mats", "5 tatami mats", and "20 m ² " indicate areas. The feature information acquisition unit 113 may remove characters indicating the area from the floor plan image. Then, the feature information acquisition unit 113 trains a model using the preprocessed floor plan image as input data and the prepared segmented floor plan information as ground truth. Then, the feature information acquisition unit 113 uses the trained model to generate segmented floor plan information from the floor plan image. In the segmented floor plan information generated at this time, if there are multiple rooms of the same type, the same label may be attached to all of the rooms. In that case, the feature information acquiring unit 113 may relabel each room separately (for example, labels such as “general room A”, “general room B”, “general room C”, etc.).

The feature information acquisition unit 113 may generate segmented floor plan information from the floor plan image on a rule basis, for example. For example, the feature information acquisition unit 113 may remove characters indicating the area from the floor plan image. Further, the feature information acquisition unit 113 may emphasize necessary edges and remove unnecessary edges by applying morphological transformation or binarization to the floor plan image. The feature information acquisition unit 113 may identify the type of each internal space by pattern recognition from the floor plan image processed in this way. For example, the feature information acquisition unit 113 may identify the type by character recognition. For example, kitchens are marked with letters such as "kitchen" and "K". The dining room is labeled with letters such as "dining" and "D", for example. Characters such as "Western-style room", "Japanese-style room", "Western-style room", "Japanese style", and "Bedroom" are attached to the general rooms. Characters such as "LDK" are attached to LDK. Characters such as "Genkan" and "Gen" are attached to the entrance. For example, characters such as "storage", "storage", and "closet" are attached to the storage. Further, the feature information acquisition unit 113 may identify the type by image recognition such as object detection. For example, the kitchen has a picture of a sink. The toilet has a picture of a toilet bowl. The bath has a picture of a bathtub. After that, the feature information acquisition unit 113 may remove the recognized characters and pictures from the floor plan image. Further, the feature information acquisition unit 113 may remove the fan shape from the floor plan image. The sector drawn in the floor plan image indicates the direction in which the hinged door normally opens. Further, the characteristic information acquisition unit 113 may specify a figure formed by each line segment from the floor plan image. Then, the feature information acquisition unit 113 may label the area surrounded by each figure. A figure including a portion where the aforementioned character or picture is recognized usually corresponds to the internal space of the object. The feature information acquisition unit 113 may associate a label indicating the type identified from the recognized character or picture with all pixels within the region corresponding to the internal space. A figure in which neither a character nor a picture has been recognized may also correspond to the internal space. For example, there is a high possibility that a place adjacent to an entrance and communicating with another internal space through an opening is a corridor. Other figures that are in contact with the figure corresponding to the interior space and that are relatively thin are usually partitions. Among figures corresponding to partitions, a rectangle having one long side in contact with only one internal space and the other long side in contact with only one internal space or the exterior of the property generally corresponds to an opening. Of the rectangles corresponding to openings, rectangles whose long sides are in contact with only one internal space correspond to entrances. Other openings are windows. However, the opening leading to the exterior of the property from the entrance is the doorway, and the opening leading to the balcony is the window. Also, the opening leading to the storage is a storage door. In this way, the feature information acquisition unit 113 performs labeling and generates segmented floor plan information.

FIG. 8 is a diagram illustrating an example of generation of segmented floor plan information. As shown in FIG. 8, segmented floor plan information 300 is generated from the floor plan image 110 . Segmented floor plan information 300 includes property exterior 310, walls 320-1 to 320-5, doorways 330-1 to 330-7, storage doors 330-8 to 330-10, windows 340-1 to 340-4. , rooms 350-1 to 350-3, entrance 360-1, washroom 360-2, toilet 360-3, bath 360-4, veranda 360-5, storage 370-1 to 370-3, etc. . Room 350-1 is LDK, room 350-2 is room A, and room 350-3 is room B.

After specifying the area corresponding to the room from the floor plan image in this way, the feature information acquisition unit 113 may specify the feature of the room based on the specified area. For example, the feature information acquisition unit 113 may identify the features of the room based on the features of the area. The feature information acquisition unit 113 may extract feature amounts as features of the room. For example, the feature information acquisition unit 113 may specify the area of (the figure forming) the area, the number of vertices, the length of the sides, the internal angles (for example, minor angles only), and the like. Then, the feature information acquisition unit 113 may specify the area of the room, the number of vertices, the length of the sides, the angles of the corners of the room, and the like as feature amounts of the room. The area, vertex, side and interior angle of a room are the area, vertex, side and interior angle of a figure corresponding to the range occupied by the room in the property when the interior of the property is viewed from above. In this case, the sides of the partition correspond to the sides or part of the sides of the figure. For example, at the stage of generating the segmented floor plan information, character recognition identifies the area of at least one interior space. The characteristic information acquisition unit 113 may calculate the ratio of the area of the area corresponding to the internal space and the area of the area corresponding to another internal space in the floor plan image. Then, the characteristic information acquisition unit 113 may calculate the area of another internal space based on the area of the internal space and the calculated ratio. Further, the characteristic information acquiring unit 113 may calculate the length of the sides of the figure surrounding the area corresponding to each room. Further, the characteristic information acquisition unit 113 may calculate the square root of the ratio between the area of the area and the area of the room corresponding to the area. Then, the feature information acquisition unit 113 may calculate the length of the side surrounding the room based on the length of the side of the area and the calculated square root. For example, the feature information acquisition unit 113 may calculate the lengths of only the shortest and smallest sides of the sides surrounding the room. The feature information acquisition unit 113 may specify the number of vertices and the angle between sides of the region as the number of vertices and the angle between sides of the room, respectively. The feature information acquisition unit 113 may specify only the minimum angle and maximum angle. The feature information acquisition unit 113 may identify at least one of the area, vertex, side, and interior angle of the room. Alternatively, the feature information acquisition unit 113 may specify at least one of vertices, sides, and interior angles.

Further, the characteristic information acquisition unit 113 may specify the adjacency state between the area corresponding to the target room and other areas. Then, based on the specified adjacency situation, the feature information acquisition unit 113 may specify the adjacency situation of other internal spaces or partitions with respect to the target room as the feature amount of the room. Here, the feature information acquisition unit 113 may identify features related to openings provided in the target room. For example, the feature information acquisition unit 113 may specify the number of windows and the number of doorways in the room as feature amounts. In the floor plan image, a window adjacent to a room is the window of that room. For example, in segmented floor plan information 300, room 350-1 is bounded by windows 340-1 and 340-2. Therefore, the number of windows is two. In the floor plan image, the doorway adjacent to the room is the doorway of the room. For example, room 350-1 is bordered by doorways 330-2, 330-6 and 330-7. Therefore, the number of doorways is three. The feature information acquisition unit 113 may specify the number of rooms adjacent to the room as the feature amount. In the floor plan image, other rooms adjacent to the target room across a partition are adjacent to the target room. It does not have to be considered whether or not the rooms are connected to each other by a doorway. For example, in segmented floor plan information 300, room 350-2 is adjacent to room 350-1 across doorway 330-7 and adjacent to room 350-3 across wall 320-6. Therefore, the number of adjacent rooms is two. The feature information acquisition unit 113 may identify the storage area in the room as the feature amount. In the floor plan image, the storage adjacent to the target room across the storage door is the storage in that room. For example, room 350-2 is adjacent to storage 370-2 across storage door 330-9. Accordingly, room 350-2 has storage 370-2. The area of storage 370-2 may be identified as the storage area of room 350-2. In the case of a room having a plurality of storages, the characteristic information acquisition unit 113 may calculate the total area of these storages as the storage area of the room.

Further, the feature information acquisition unit 113 may identify the positional relationship between the area corresponding to the target room and the area corresponding to another internal space. Then, the feature information acquiring unit 113 may specify the positional relationship between the target room and other internal spaces as the feature of the room. For example, the characteristic information acquisition unit 113 may specify whether the room is connected to a corridor, whether the room is connected to the living room or the LDK, whether the room is connected to a veranda, etc., as characteristics. . In the floor plan image, the internal space adjacent to the target room across the doorway is connected to the room. Further, the feature information acquiring unit 113 may specify whether or not it is necessary to pass through an internal space other than a corridor to get from the room to the entrance. This feature can be identified by a pathfinding algorithm such as Dijkstra's algorithm. Further, the feature information acquiring unit 113 may specify whether the doorway of the room is a hinged door or a sliding door as a feature of the room. In the floor plan image, the doorway whose region is in contact with the sector may be a hinged door. Other doorways may be sliding doors. For the feature indicated by whether or not the predetermined condition is met, for example, 1 may be set as the feature amount when the condition is met, and 0 may be set as the feature amount when the condition is not met.

The feature information acquisition unit 113 may acquire feature amounts of the same items for all types of rooms. The item of the feature amount is each division when the features of the room are classified by name, such as "area of room" and "number of vertices". Alternatively, the feature information acquisition unit 113 may make all or part of the feature amount items to be acquired different from the feature amount items acquired in other rooms, depending on the room type.

FIG. 9 is a diagram showing a configuration example of a property vector acquired from a floor plan image. The feature information acquisition unit 113 may generate a feature vector composed of the extracted feature amount. Here, as an example, the features of a room are the area, the number of vertices, the length of the longest side, the length of the shortest side, the maximum angle, the minimum angle, the number of windows, the number of entrances and exits, and the number of adjacent rooms. and the area of storage shall be extracted. As shown in FIG. 9, in a feature vector 400, feature amounts for 10 items are stored for each room. In FIG. 9, feature amounts are stored for the LDK and general rooms A and B specified from the floor plan image 110 . Here, the area influences the livability of the room and how much furniture and the like can be placed. The number of vertices indicates the number of angles. The length of the longest side affects whether relatively long pieces of furniture such as beds, desks, TV stands, etc. can be placed. As for the length of the shortest side, the larger the maximum angle is from 90 degrees and the further from 90 degrees, the more likely dead space is generated when furniture or the like is installed. When the minimum angle is smaller than 90 degrees and away from 90 degrees, dead space is likely to occur when furniture or the like is installed. The number of windows affects the lighting. As the number of doorways increases, the number of wires required to enter and exit the room increases, but the installation space for furniture and the like is limited. The smaller the number of adjoining rooms, the more likely it is that the rooms are located in the corners of the property. The storage area affects the amount of things that can be stored.

The feature information acquisition unit 113 may generate the segmented floor plan information and feature vectors at or after the floor plan image of the target property is displayed. Alternatively, the feature information acquisition unit 113 may generate segmented floor plan information and feature vectors in advance and store the information in the property DB 14b.

Note that the feature information acquisition unit 113 itself does not need to generate feature information. For example, a device different from the information providing server 1 may generate feature information, and the feature information acquisition unit 113 may acquire the feature information from that device.

[3-4. Generate model]
Based on the evaluation information acquired by the evaluation information acquisition unit 112 and the feature information acquired by the feature information acquisition unit 113, the generation unit 114 generates a property classifier, which is an example of a model for identifying properties preferred by the user. . The property classifier to be generated may be a classifier whose input data is characteristic information of all the rooms of the property and whose output data is estimated evaluation information indicating the evaluation that the user is supposed to input for the property. . The estimated evaluation information to be output may indicate either a positive evaluation of “like” or a negative evaluation of “dislike”, or may indicate the degree or probability that the user likes the property. Also, the property classifier to be generated may be either a statistical model or a machine learning model as long as it is a model that uses evaluation information and feature information as learning data or training data.

The generation unit 114 may generate a property classifier so as to identify properties that have relatively many features that the user likes and that have relatively few features that the user does not like as properties that the user likes. . The generation unit 114 determines that the ratio of a certain feature to the features given the evaluation of "like" is higher than the ratio of the certain feature to the features given the evaluation of "dislike". A property classifier may be generated that identifies properties preferred by the user from among the properties. Even if a user likes a particular feature of an item of a particular type of room in a property, the user does not necessarily input a rating of "like" for the floor plan image of the property. If the user does not like other features or other rooms, the user may enter a rating of "dislike". However, as the number of properties evaluated by the user increases, the ratio of the specific feature among the features given the evaluation “favorite” for the feature item of the room of a specific type converges, for example, to the first ratio. . Along with this, the ratio of the specific feature among the features given the evaluation “dislike” for the feature item of the room of the specific type converges, for example, to the second ratio. Here, the first percentage is higher than the second percentage. For example, the user likes a regular room with two windows. However, there are fewer rooms with two windows than rooms with one window. Therefore, it is highly likely that two windows account for less than one window in the number of windows in general rooms that are rated "favorite". Similarly, the ratio of two windows to the number of windows in a general room rated "disliked" is likely to be smaller than the ratio of one window. Therefore, when comparing between the evaluation "like" and "dislike", the ratio of 2 windows to the number of evaluation windows of "like" is higher than the ratio of 2 windows to the number of evaluation windows of "dislike". will also grow. On the other hand, the ratio of one window to the number of windows rated “like” is smaller than the ratio of one window to the number of windows rated “dislike”. From this, it can be inferred that the user relatively "likes" the general room with two windows and relatively "dislikes" the general room with one window. If at least one characteristic does not satisfy such conditions, it is considered that the user does not like the property. When using statistical models as object classifiers, it is relatively easy to design such object classifiers. When a machine learning model is used as a property classifier, it is considered that the property classifier automatically learns this point by optimizing the parameters in the learning stage.

In addition, the generation unit 114 generates a plurality of probability distributions corresponding to each of a plurality of evaluations including the evaluations “like” and “dislike”, and the characteristics of the property from which the evaluation information indicating the evaluations corresponding to the probability distributions is acquired. A property classifier may be generated that indicates a plurality of probability distributions each having information as a random variable. The property classifier in this case is a model that identifies properties preferred by the user based on these multiple probability distributions. Each probability distribution can be said to be information indicating the proportion of each feature among the features evaluated corresponding to the probability distribution.

FIG. 10 is a diagram showing a configuration example of a property classifier. As shown in FIG. 10, the property classifier 500 may include models 510 for multiple rooms. Each room model 510 is a model for estimating a user's evaluation of a room of a type corresponding to the room model 510 . By generating a model for each room type, the user's evaluation of each room can be reflected in the output of the property classifier. FIG. 10 shows, as room models 510, a general room model, a living room model, a dining room model, a kitchen model, an LDK model, a dining kitchen model, a closet model, and loft models are shown. Property classifier 500 may further include rating score calculator 520 . The evaluation score calculation unit 520 generates estimated evaluation information used for identifying a property preferred by the user based on the information indicating the evaluation of the room output from each of the models for all rooms 510 .

Each room model 510 may include multiple feature models 530 . Each feature item model 530 is a model for estimating the user's evaluation of the feature quantity of the item corresponding to the feature item model. Each room model 510 outputs a plurality of feature item scores output from each of a plurality of feature item models 530 to be described later, to the evaluation score calculation unit 520 . Each feature amount has no or low relationship with other feature amounts. Therefore, by generating a model for each feature amount, the structure of the model can be simplified. In FIG. 10, as feature item models 530, a model for room area, a model for the number of vertices, a model for the length of the longest side, a model for the length of the shortest side, and a model for the maximum angle. a model for the minimum angle, a model for the number of windows, a model for the number of doorways, a model for the number of adjoining rooms, and a model for storage area.

Each feature model 530 may include a like model 540 and a dislike model 550 . The liking model 540 may output a liking score indicating the degree to which the user is estimated to input a rating of “like” for the feature amount of the target feature item of the target room type. The dislike model 550 may output a dislike score indicating the degree to which the user is presumed to input the evaluation "dislike" for the feature amount of the target feature item of the target room type. Each feature model 530 may further include a feature score calculator 560 . The feature item score calculator 560 may integrate the like score and dislike score output from the like model 540 and the dislike model 550 and output the feature item score output from the feature item model 530. .

For each of the plurality of room models 510, the generation unit 114 generates the room model 510 based on the feature amount of the room of the type corresponding to the room model 510 and the evaluation information corresponding to the property including the room. may For example, for each combination of room type and feature item, the generating unit 114 generates the favorite model 540 using the feature amount of the item of the type of room for which the user has input the evaluation “like” as learning data. good too. Further, for each combination of the room type and the feature item, the feature information acquisition unit 113 generates the dislike model 550 by using the feature amount of the item of the type of the room for which the user has input the evaluation "dislike" as learning data. You may However, the generation unit 114 may exclude, from the learning data, feature amounts for properties for which responses other than "floor plan" are input as questionnaire responses to which the evaluation "dislike" is input. In this case, it is considered that the user has not given any evaluation to the floor plan image. It is appropriate not to use such property information for learning. When a normal distribution or mixed normal distribution is used as the like model 540 and the dislike model 550, the feature information acquisition unit 113 may generate the models using, for example, maximum likelihood estimation. The number of normal distributions forming the mixed normal distribution is not particularly limited. As a result of generating all the like models 540 and all the dislike models 550, the all feature item model 530 is generated and the all room model 510 is generated. As a result, the property classifier 500 is generated.

For a property for which a partial evaluation has been input by the user, the generation unit 114 uses the partial evaluation included in the evaluation information for a room specified by the user among the rooms included in the property, and uses the partial evaluation included in the evaluation information. For rooms other than the room designated by the user, a property classifier may be generated using the overall evaluation included in the evaluation information. For example, the generation unit 114 generates a model corresponding to partial evaluation among the like model 540 and the dislike model 550 for all feature item models 530 of the room model 510 corresponding to the room designated by the user. You may use the feature-value of the room. On the other hand, the generation unit 114 generates a model corresponding to the overall evaluation, out of the like model 540 and the dislike model 550, for all the feature item models 530 of the room models 510 corresponding to rooms not specified by the user. , the feature amount of the room may be used. As a result, the property classifier 500 can be made to reflect the evaluation of the room for the property for which the evaluation for some rooms is different from the evaluation for the whole property.

The generation unit 114 may generate the property classifier 500 for each user. This makes it possible to personalize which property is recommended to the user.

Note that when a neural network is used as the property classifier, the property classifier may include, for example, a plurality of fully connected layers as intermediate layers. The generation unit 114 inputs the feature information to this property classifier and causes it to output estimated evaluation information. The generation unit 114 may use the evaluation information acquired by the evaluation information acquisition unit 112 as ground truth to train the property classifier by error backpropagation.

[3-5. Determination of recommended properties]
The determination unit 115 determines the property to be presented to the user based on the property classifier 500 generated by the generation unit 114 . The determining unit 115 selects candidate properties according to, for example, predetermined conditions. For example, the determining unit 115 may select candidates at random, or may select properties whose elapsed days from the date of registration of the properties are less than a predetermined value as candidates. Further, the determining unit 115 may select a property that matches the saved search condition as a candidate. The determining unit 115 acquires feature information of the selected candidate, such as a feature vector. The determination unit 115 inputs the feature amount included in the feature vector to each of the like model 540 and the dislike model 550 in the feature item model 530 of the room model 510 corresponding to the feature amount. Then, the determination unit 115 acquires a liking score and a disliking score from the liking model 540 and the disliking model 550, respectively. If the candidate property does not have a specific type of room, all the like models 540 and all the dislike models 550 of the room models 510 corresponding to that type output 0 as a score. good too.

The determination unit 115, as the characteristic item score calculation unit 560 of each characteristic item model 530, may calculate the characteristic item score by, for example, subtracting the dislike score from the like score. FIG. 11 is a diagram showing a calculation example of feature item scores. In FIG. 11, graph 610 shows the probability distribution exhibited by trained liking model 540 . Graph 620 shows the probability distribution exhibited by trained dislike model 550 . Each of graphs 610 and 620 shows a mixed normal distribution composed of two normal distributions. In a mixed normal distribution, the integral of the probability density is one. The probability density y of the feature quantity x satisfies 0<y<1. Assume that graphs 610 and 620 show probability distributions corresponding to, for example, the area of a living room. Here, the area x of the living room of a certain property is input as a feature amount to the room area model included in the living room model. In graph 610, the probability density for area x is 0.3. Like model 540 may output this probability density as a like score. The higher the liking score, the higher the possibility that the user likes the living area. In graph 620, the probability density for area x is 0.1. Dislike model 550 may output this probability density as a dislike score. The higher the dislike score, the higher the possibility that the user dislikes the living area. From these scores, the determination unit 115 determines 0.2 as the feature item score for the area of the living room.

If the feature item score is greater than 0, it is likely that the user likes the target feature of the target room. On the other hand, if the feature item score is less than 0, the user may dislike the target feature of the target room. The higher the absolute value of the feature item score, the more likely that the user will like or dislike the subject feature of the subject room. Therefore, the smaller the absolute value, the more likely that the user neither likes nor dislikes the subject feature of the subject room.

The difference in probability density (the difference between the like score and the dislike score) between the probability distribution of the feature value with the input rating of “like” and the probability distribution of the feature value with the rating of “dislike” is The larger it is, the more likely it is that the user has input the evaluation with emphasis on the target feature of the target room. For example, the greater the difference between the like score and the dislike score in the feature value that maximizes the like score, or the greater the difference between the like score and the dislike score in the feature value that maximizes the dislike score. , users are likely to attach importance to FIG. 12 is a diagram showing an example of the difference between the probability distribution of the feature quantity of the evaluation “like” and the probability distribution of the feature quantity of the evaluation “dislike”. The upper part of FIG. 12 shows superimposed graphs 610 and 620 respectively showing the probability density corresponding to the area of the living room. The upper part of FIG. 13 shows superimposed graphs 630 and 640 respectively showing the probability density corresponding to the storage area of the general room. A graph 630 shows the probability distribution exhibited by the liking model 540 included in the storage space model in the general room model. A graph 640 shows the probability distribution indicated by the dislike model 550 included in the storage area model in the general room model. Overall, the difference between the like score and the dislike score for the storage area of the living room is large compared to the difference between the like score and the dislike score for the storage area of the general room. Therefore, it is highly likely that the user places more importance on the area of the living room than on the storage area of the general room.

The determination unit 115, as the evaluation score calculation unit 520, may integrate the feature item scores output from the models for all feature items of the model for all rooms to calculate the evaluation score as the estimated evaluation information. For example, the evaluation score calculator 520 may calculate the total value or average value of all feature item scores as the evaluation score. Alternatively, evaluation score calculation section 520 determines that at least one of the like score and dislike score before calculating the characteristic item score is equal to or greater than a first predetermined value, and the absolute value of the characteristic item score is the second A feature item score that is equal to or greater than a predetermined value of is extracted. Then, the evaluation score calculation unit 520 may calculate the total value or average value of only the extracted feature item scores as the evaluation score. This is a measure to prevent the final evaluation score from reflecting the evaluation of features that the user does not attach importance to.

If the evaluation score is equal to or greater than a predetermined threshold, the determination unit 115 may determine that the selected candidate property is identified as the user's favorite property, and that the property be presented to the user. The threshold may be, for example, 0 or a value greater than 0. By setting the threshold value to 0 or more, the ratio of a certain feature to the features given the evaluation "like" is higher than the ratio of the certain feature to the features given the evaluation "dislike". It is ensured that the property preferred by the user is identified among the properties that have rooms with features.

The determining unit 115 may determine the property to be presented to the target user based on the property classifier 500 generated for other users having the same attribute as the target user's attribute. Users with the same attribute may have the same or similar property preferences. For example, when the property classifier 500 corresponding to the target user has not been generated, or when the estimation accuracy by the property classifier 500 is low due to a small number of properties for which evaluations have been input by the target user, the same attribute By using the property classifier 500 corresponding to the user having the , it is possible to appropriately identify properties that the target user prefers. For example, the determination unit 115 may calculate the estimation accuracy of the generated property classifier 500 using property information already evaluated and its evaluation information. That is, the determination unit 115 may input the feature information to the property classifier 500 and acquire the estimated evaluation information. The determining unit 115 may calculate the rate of matching between the estimated evaluation information and the evaluation information as the estimation accuracy. If the estimation accuracy is less than a predetermined value, the determination unit 115 may use the property classifier 500 generated for another user.

The determining unit 115 may determine which user's property classifier 500 is to be used, using attributes such as gender, date of birth, and address stored in the member DB 14a. Further, the determination unit 115 may determine which user's property classifier 500 is to be used based on the attributes of the target user registered in a system different from the property information system S. FIG. For example, a user who uses the property information system S may be able to use other systems using the same user ID as that user's user ID. In this case, the determination unit 115 may acquire information on other users who have the same attributes as those of the target user by transmitting a request including the user ID to the system. The other system may be, for example, a website, or a system available through a dedicated application that can be installed on the user terminal 2 . Examples of such systems include online shopping malls, hotel reservation systems, job search support systems, marriage support systems, web searches, flea markets, and ticket sales sites. For example, assume that the target user is using a job hunting support system. Therefore, even if the determining unit 115 identifies a user who is in the same university and on the same campus as the target user, who is applying for the same type of business, and where the applying company is located, good. The determination unit 115 may acquire, from the system, information indicating other users who have the same attributes as the attributes registered by the user in the system, for example. In addition, the determination unit 115, for example, information indicating actions taken by the target user in the system or information indicating other users having the same attributes as the attributes of the target user identified from the history of actions taken, may be obtained from the system.

When the display control unit 111 causes the user terminal 2 to display the floor plan images of a plurality of properties, the determination unit 115 may determine the properties whose floor plan images are to be displayed. Alternatively, the determination unit 115 may determine the property to be presented to the user, for example, every time a predetermined period of time elapses (for example, every 12 hours, every day, every 3 days, etc.).

The system control unit 11 presents information on the property determined by the determination unit 115 as the property to be presented to the user to the target user by some method. The determined property may be presented as a property recommended to the user. For example, as described above, the display control unit 111 may cause the user terminal 2 of the user to display the floor plan image and property information of the determined property when the user logs in for the first time on the day. Alternatively, the display control unit 111 displays the property information of the decided property and the A floor plan image may be displayed on the user terminal 2, or a list of determined properties may be displayed on the user terminal 2. FIG. Alternatively, the system control unit 11 may send an e-mail including property information and a floor plan image of the decided property to the user. Alternatively, the system control unit 11 may transmit a message to the effect that there is a recommended property to the user terminal 2 by push notification when the property to be presented to the user is found. When the user selects the message displayed on the user terminal 2, the user terminal 2 activates the application dedicated to the property information providing system S, and in response, the system control unit 11 outputs the property information of the determined property. or the floor plan image may be displayed on the user terminal 2 .

[4. Operation of Property Information Providing System]
Next, the operation of property information providing system S will be described with reference to FIGS. 13 to 17. FIG. The system control unit 11 of the information providing server 1 executes the processes shown in FIGS. 13, 14, 15 and 17 according to various program codes included in the server program.

FIG. 13 is a flowchart showing an example of floor plan analysis processing by the system control unit 11 of the information providing server 1 according to this embodiment. For example, when the property information input by the real estate agent is stored in the property DB 14b, the system control unit 11 may execute the floor plan analysis process for the property information.

As shown in FIG. 13, the feature information acquisition unit 113 acquires a floor plan image from property information (step S101). Next, the feature information acquisition unit 113 generates segmented floor plan information based on the floor plan image (step S102). Next, the feature information acquisition unit 113 selects one of the rooms in the property indicated by the segmented floor plan information (step S103). Next, the feature information acquisition unit 113 selects one of a plurality of predetermined feature items (step S104). Next, the feature information acquisition unit 113 identifies an area corresponding to the selected room from the segmented floor plan information. For example, the feature information acquisition unit 113 identifies the positions of all pixels labeled with the selected room. Then, the feature information acquisition unit 113 acquires a feature quantity corresponding to the selected feature item based on the specified area (step S105). Next, the feature information acquisition unit 113 determines whether or not there is a feature item that has not yet been selected among a plurality of predetermined feature items (step S106). If there is a feature item that has not been selected yet (step S106: YES), the process proceeds to step S104, and the feature information acquisition unit 113 selects one of the feature items that have not yet been selected. On the other hand, if there is no feature item that has not been selected yet (step S106: NO), the feature information acquisition unit 113 determines that there is a room that has not been selected among the rooms in the property indicated by the segmented floor plan information. It is determined whether or not (step S107). If there is a room that has not been selected yet (step S107: YES), the process proceeds to step S103, and the feature information acquisition unit 113 selects one of the rooms that have not been selected yet. On the other hand, if there is no room that has not been selected yet (step S107: NO), the feature information acquisition unit 113 generates a feature vector (step S108). For example, the feature information acquisition unit 113 generates a one-dimensional array including feature amounts of predetermined feature items for each room. At this time, the feature information acquisition unit 113 associates the room number and room type of the corresponding room with each one-dimensional array. Then, the feature information acquisition unit 113 arranges the one-dimensional arrays to generate a feature vector. Next, the feature information acquisition unit 113 associates the generated segmented floor plan information and segmented feature vectors with the property information and stores them in the property DB 14b (step S109), and the floor plan analysis process ends. .

FIG. 14 is a flowchart showing an example of property recommendation processing by the system control unit 11 of the information providing server 1 according to this embodiment. For example, the target user activates an application dedicated to the property information providing system S. In response to this, the user terminal 2 transmits a login request including, for example, the user ID and password stored in the user terminal 2 to the information providing server 1 . When determining that the login has succeeded based on the login request, the system control unit 11 determines whether or not the current login by the target user is the first login today. If this login is the first login, the system control unit 11 may execute property recommendation processing.

As shown in FIG. 14, the determination unit 115 determines whether or not the parameter of the property classifier 500 is stored in the model DB 14d in association with the user ID of the target user (step S201). If the parameters are not stored (step S201: NO), the determining unit 115 identifies other users who have the same attributes as the target user's attributes (step S202). For example, the determination unit 115 may acquire at least one attribute of the target user's attributes stored in the member DB 14a as a search condition. Then, the determination unit 115 may search the member DB 14a for a user having an attribute that matches the search condition. Alternatively, the determining unit 115 may request a list of user IDs of users having the same attribute as the target user from a server device of a system different from the property information providing system S. At this time, the determining unit 115 may designate an attribute item as a search condition. Next, the determining unit 115 determines whether or not the parameter of the property classifier 500 is stored in the model DB 14d in association with the specified user ID of the other user (step S203). If the parameters of the property classifier 500 of the target user or the parameters of the property classifier 500 of other users are stored (step S201: YES or step S203: YES), the determining unit 115 uses the parameters as models. Acquired from DB 14d (step S204). After step S204, or when the parameters of the property classifier 500 of another user are not stored (step S203: NO), the determining unit 115 sets the property number i to 0 (step S205). Next, the determination unit 115 executes property selection processing (step S206).

FIG. 15 is a flowchart showing an example of property selection processing by the system control unit 11 of the information providing server 1 according to this embodiment. As shown in FIG. 15, the determination unit 115 determines whether or not the parameters of the property classifier 500 have been acquired in the property recommendation process (step S301). That is, the determination unit 115 determines whether step S204 has been performed. If the parameters have not been acquired (step S301: NO), the determination unit 115 selects any property whose property information is stored in the property DB 14b as a display property (step S302), and the property selection process ends. do. For example, the determining unit 115 acquires the property ID of the displayed property. At this time, the determination unit 115 refers to the evaluation information DB 14c, and excludes properties for which the target user has already input an evaluation from the displayed properties.

On the other hand, when parameters have been acquired (step S301: YES), the determination unit 115 determines whether the acquired parameters are parameters of the property classifier 500 of the target user (step S303). If the acquired parameter is the parameter of the property classifier 500 of the target user (step S303: YES), the determination unit 115 determines whether the target user has been displayed the floor plan image for a predetermined number of days or more. It is determined whether or not there is an object that has passed (step S304). For example, the determination unit 115 extracts the evaluation information stored in association with the user ID of the target user from the evaluation information DB 14c. The determination unit 115 searches the extracted evaluation information for evaluation information in which the number of days elapsed from the display date of the floor plan image to today is equal to or greater than a predetermined number of days. Here, the determining unit 115 may search only for evaluation information for which the overall evaluation is set to “like”. When the corresponding evaluation information is found, the determination unit 115 determines that there is a property for which a predetermined number of days or more have passed since the day when the floor plan image was displayed (step S304: YES). In this case, the determining unit 115 generates a pseudo-random number within the range of 0 to 100, for example (step S305). Next, the determining unit 115 determines whether or not the generated pseudorandom number is less than a predetermined probability (step S306). This probability may be set in advance based on the percentage of properties for which the user has input an evaluation in the past to be included in the properties presented to the user. If the pseudo-random number is less than the predetermined probability (step S306: YES), the determination unit 115 selects one of the properties for which a predetermined number of days or more have passed since the date the floor plan image was displayed as the display property. (Step S307), the property selection process ends. For example, the determining unit 115 acquires the property ID of the displayed property from any of the retrieved evaluation information.

If the acquired parameter is not the parameter of the property classifier 500 of the target user (step S303: NO), if there is no property for which a predetermined number of days or more have passed since the date the floor plan image was displayed (step S304 : NO), or if the pseudo-random number is equal to or greater than the predetermined probability (step S306: NO), the determining unit 115 selects one of the properties whose property information is stored in the property DB 14b as a candidate property (step S308). . At this time, the determination unit 115 refers to the evaluation information DB 14c, and excludes the property for which the target user has already input the evaluation from the candidate properties.

Next, the feature information acquisition unit 113 acquires the feature vector of the candidate property from the property DB 14b. Determining unit 115 determines each feature amount included in the acquired feature vector in model 510 for the type of room corresponding to the feature amount. Input to the model 540 and the dislike model 550 to acquire the like score and dislike score (step S309). For example, the determination unit 115 uses the parameters of the liking model 540 among the acquired parameters to calculate the probability density of the feature quantity as the liking score. Further, the determination unit 115 uses the parameters of the dislike model 550 to calculate the probability density of the feature quantity as the dislike score.

Next, the determination unit 115 excludes combinations of room types and feature items for which both the like score and the dislike score are less than the first predetermined value from the combinations for which the feature item scores are calculated (step S310). Next, the determining unit 115 subtracts the dislike score from the like score for each of the combinations of the room type and the characteristic item that were not excluded in step S310 to calculate the characteristic item score (step S311). Next, the determining unit 115 excludes, from among the combinations for which the feature item scores have been calculated, the combination whose absolute value of the feature item score is less than the second predetermined value from the combinations whose feature item scores are used in the calculation of the evaluation score. (step S312). Next, the determining unit 115 calculates an evaluation score by adding together the feature item scores of the combinations that were not excluded in step S312 among the combinations of room types and feature items (step S313).

Next, the determination unit 115 determines whether or not the evaluation score is equal to or greater than the threshold (step S314). If the evaluation score is less than the threshold (step S314: NO), the process proceeds to step S303. On the other hand, if the evaluation score is equal to or greater than the threshold (step S313: YES), the determination unit 115 selects the candidate property as the display property (step S315), and ends the property selection process.

As shown in FIG. 14, when the property selection process ends, the display control unit 111 acquires property information, floor plan information, and segmented floor plan information of the displayed property from the property DB 14b. Then, the display control unit 111 transmits the acquired information to the user terminal 2 of the target user (step S207). Next, the evaluation information acquisition unit 112 receives evaluation information transmitted from the user terminal 2 of the target user (step S208). Next, the evaluation information acquisition unit 112 adds the user ID of the target user, the property ID of the displayed property, and the display date indicating today's date to the received evaluation information. Then, the evaluation information acquisition unit 112 stores the evaluation information in the evaluation information DB 14c (step S209). When the evaluation information of the displayed property is already stored in the evaluation information DB 14c, the evaluation information acquisition unit 112 may rewrite the evaluation information with new evaluation information.

Next, the display control unit 111 determines whether or not the number i of properties is less than the predetermined upper limit number N of properties (step S210). If the number of properties i is less than the upper limit number N (step S2120: YES), the display control unit 111 increases the number of properties i by 1 (step S211), and the process proceeds to step S206. On the other hand, if the number of properties i is equal to or greater than the upper limit number N (step S2120: NO), the property recommendation process ends.

FIG. 16 is a flowchart showing an example of floor plan display processing by the CPU of the user terminal 2 according to this embodiment. For example, after the target user activates the application dedicated to the property information providing system S, the information providing server 1 transmits property information, floor plan information, and segmented floor plan information to the user terminal 2 of the user. The user terminal 2 may execute the floor plan display process when receiving such information from the information providing server 1 .

As shown in FIG. 16, the user terminal 2 displays the received floor plan image on the display (step S401). Next, the user terminal 2 acquires the coordinates of the touched pixel on the screen at predetermined time intervals, for example, from a touch panel provided in the user terminal 2 . The user terminal 2 determines the user's operation based on the acquired coordinate information.

Then, the user terminal 2 determines whether or not the user operation is a rightward or leftward charge flick (step S402). If the user operation is a charge flick (step S402: YES), the user terminal 2 identifies the room designated by the user (step S403). For example, the user terminal 2 identifies the coordinates of the first touched pixel in the charge flick operation. The user terminal 2 obtains the label of the pixel at the specified coordinates from the segmented floor plan information. The user terminal 2 identifies the room number and room type of the designated room based on the acquired label. Next, the user terminal 2 determines whether or not the direction of the accumulation flick is rightward (step S404). If the direction of the charging flick is rightward (step S404: YES), the user terminal 2 generates partial evaluation information including the room number and room type of the designated room and the partial evaluation "like" (step S405 ), the process proceeds to step S402. On the other hand, if the charging flick direction is not the right direction (step S404: NO), the user terminal 2 generates partial evaluation information including the room number and room type of the designated room and the partial evaluation "dislike" ( step S406), and the process proceeds to step S402.

If the user operation is not a charge flick (step S402: NO), the user terminal 2 determines whether or not the user operation is a rightward or leftward normal flick (step S407). If the user operation is not a normal flick (step S407: NO), the user terminal 2 determines whether the user operation is another valid operation (step S408). If the user operation is other operation (step S408: YES), the user terminal 2 executes processing according to the operation (step S409), and the process proceeds to step S401. For example, when the user operation is a rightward or leftward swipe, the user terminal 2 displays property information such as property name, property image, and detailed information on the property detail screen. Then, when the user swipes rightward or leftward again, the process proceeds to step S401.

If the user operation is the normal flick (step S407: YES), the user terminal 2 determines whether the direction of the normal flick is rightward (step S410). If the normal flick direction is the right direction (step S410: YES), the user terminal 2 sets the overall evaluation to "like" (step S411). On the other hand, if the direction of the normal flick is not the right direction (step S410: NO), the user terminal 2 sets the overall evaluation to "dislike" (step S412). Next, the user terminal 2 displays a questionnaire screen (step S413). Next, the user terminal 2 acquires the questionnaire answers entered by the user on the questionnaire screen (step S414).

After step S411 or step S414, the user terminal 2 generates evaluation information including overall evaluation. If partial evaluation information has been generated in step S405 or S406, the user terminal 2 adds the partial evaluation information to the evaluation information. Moreover, when the questionnaire answer is acquired in step S414, the user terminal 2 adds the questionnaire answer to the evaluation information. Then, the user terminal 2 transmits the generated evaluation information to the information providing server 1 (step S415), and the floor plan display process ends.

FIG. 17 is a flowchart showing an example of learning processing by the system control unit 11 of the information providing server 1 according to this embodiment. For example, after the property recommendation process ends, the system control unit 11 may perform the learning process on the target user to whom the property was presented.

As shown in FIG. 17, the generation unit 114 acquires all evaluation information associated with the user ID of the target user from the evaluation information DB 14c (step S501). Next, the feature information acquisition unit 113 acquires the feature vector associated with the property ID included in each acquired evaluation information from the property DB 14b (step S502). Next, generation unit 114 selects one of a plurality of predetermined room types (step S503). Next, the generation unit 114 selects one of a plurality of predetermined feature items (step S504).

Next, the generation unit 114 acquires partial evaluation information in which the selected room type and the partial evaluation "like" are set from among the evaluation information acquired in step S501. Then, for each piece of acquired partial evaluation information, the generation unit 114 extracts a feature vector corresponding to the combination of the room number included in the partial evaluation information and the selected feature item from the feature vector of the property corresponding to the partial evaluation information. The amount is acquired (step S505). Next, the generation unit 114 acquires the evaluation information in which the overall evaluation “like” is set from among the evaluation information acquired in step S501. The feature information acquisition unit 113 identifies the room of the selected room type for the property corresponding to each piece of evaluation information acquired here. Of the specified rooms, the feature information acquisition unit 113 acquires the feature amount corresponding to the combination of the room number and the selected feature item for the property for which there is no partial evaluation information for the room (step S506). Next, the generation unit 114 uses the feature values acquired in steps S505 and S506 to determine the parameters of the liking model 540 (step S507). For example, the generation unit 114 may calculate the average value, the variance, and the weight of each normal distribution that constitutes the mixed normal distribution indicating the probability distribution of the acquired feature values, by maximum likelihood estimation.

Next, the generation unit 114 acquires partial evaluation information in which the selected room type and the partial evaluation "dislike" are set from the evaluation information acquired in step S501. Then, for each piece of acquired partial evaluation information, the generation unit 114 extracts a feature vector corresponding to the combination of the room number included in the partial evaluation information and the selected feature item from the feature vector of the property corresponding to the partial evaluation information. The amount is acquired (step S508). Next, the generating unit 114 acquires the evaluation information in which the overall evaluation “dislike” is set from among the evaluation information acquired in step S501. The feature information acquisition unit 113 identifies the room of the selected room type for the property corresponding to each piece of evaluation information acquired here. Among the specified rooms, the feature information acquisition unit 113 acquires feature amounts corresponding to combinations of room numbers and selected feature items for properties for which there is no partial evaluation information for the rooms (step S509). Next, the generation unit 114 determines the parameters of the dislike model 550 in the same manner as in step S507 using the feature values acquired in steps S508 and S509 (step S510).

Next, the generation unit 114 determines whether or not there is a feature item that has not yet been selected among a plurality of predetermined feature items (step S511). If there are feature items that have not been selected yet (step S511: YES), the process proceeds to step S504, and the generation unit 114 selects one of the feature items that have not yet been selected. On the other hand, if there is no feature item that has not been selected yet (step S511: NO), the feature information acquisition unit 113 determines whether or not there is a room type that has not been selected among a plurality of predetermined room types. Determine (step S512). If there is a room type that has not been selected yet (step S512: YES), the process proceeds to step S503, and the generator 114 selects one of the room types that have not been selected yet. On the other hand, if there is no room type that has not been selected yet (step S512: NO), the generation unit 114 associates the determined parameter with the user ID of the target user and stores it in the model DB 14d (step S513), The learning process ends. At this time, the generation unit 114 stores a parameter in association with each combination of the room type and the feature item.

As described above, according to the present embodiment, the information providing server 1 displays floor plan images of each of a plurality of properties. In addition, the information providing server 1 acquires evaluation information indicating the user's evaluation of the displayed floor plan image for each of the plurality of properties. In addition, the information providing server 1 acquires, for each of the plurality of properties, feature information that is generated based on the floor plan image and that indicates the features of the rooms within the property. Also, the information providing server 1 generates a property classifier based on the acquired evaluation information and the acquired feature information. Also, the information providing server 1 determines the property to be presented to the user based on the generated property classifier. Therefore, even if the user does not input the conditions of the property grasped from the floor plan, it is possible to provide the information of the property that the user prefers.

Here, the information providing server 1 may display the floor plan image on the screen of a touch panel having a display having a screen and a sensor for detecting a contact position on the screen. Further, the information providing server 1 may acquire evaluation information indicating a predetermined evaluation when movement in a predetermined direction of the contact position is detected on the screen on which the floor plan image is displayed. In this case, the evaluation can be input by a simple operation.

Also, the information providing server 1 may generate a property classifier for each user. In this case, personalized property information can be provided.

Further, the information providing server 1 may determine the property to be presented to the target user based on the property classifier generated for another user having the same attribute as the target user. In this case, even if a property classifier with appropriate accuracy has not been generated for the user to whom the property is presented, the property to be presented to the user can be determined appropriately.

Further, the information providing server 1 may further acquire partial evaluation information indicating the user's evaluation of the portion specified by the user in the displayed floor plan image for at least one of the plurality of properties. . In addition, the information providing server 1 uses the acquired partial evaluation information for a room corresponding to the specified portion of the rooms in the property, A property classifier may be generated using the overall evaluation of the obtained evaluation information for a room different from the room. In this case, even if the property as a whole can be given a specific rating, if the user wants to give a different rating to some of the rooms in the property, each rating can be assigned to the property classification. It can be reflected in the generation of vessels.

In addition, the information providing server 1 reproduces the floor plan image of at least one of the plurality of properties whose floor plan images are displayed after a predetermined number of days have passed since the floor plan image was displayed. may be displayed. In addition, the information providing server 1 may further acquire value information indicating the user's re-evaluation of the re-displayed floor plan image. In addition, the information providing server 1 uses the re-evaluation information obtained later for the property whose floor plan image is re-displayed among the plurality of properties, and uses the obtained evaluation information for the property different from the property. The information may be used to generate property classifiers. In this case, it is possible to generate a property classifier that reflects the user's preferences, which change with the passage of time or by looking at information on various properties.

Further, the information providing server 1 may specify the area corresponding to the room from the floor plan image. Also, the information providing server 1 may identify the features of the room based on the identified area. Here, the information providing server 1 may specify the features of the room based on the features of the specified area. In this case, the property classifier can reflect the evaluation of the visual features of the rooms in the property obtained from the floor plan image.

Further, the information providing server 1 provides feature information indicating at least one of the area of the room, the length of the side of the figure corresponding to the range occupied by the room in the property, the interior angle of the figure, and the number of vertices of the figure. may be obtained. In this case, it is possible to provide property information according to the user's preferences regarding the characteristics of rooms that can be specified from the floor plan.

In addition, the information providing server 1 has a certain feature in which the proportion of the features to which the evaluation "like" is given is greater than the proportion of the features to which the evaluation "dislikes" is given. A property classifier may be generated that identifies properties that the user prefers among properties that have rooms. In this case, a property with at least a room having characteristics preferred by the user can be identified as a property favored by the user.

In addition, the information providing server 1 uses a plurality of probability distributions corresponding to each of a plurality of evaluations of the property preferred by the user, and the feature information obtained by obtaining the evaluation information indicating the evaluation corresponding to the probability distribution is set as a random variable. A property classifier may be generated for identification based on a plurality of probability distributions. In this case, a property classifier is generated that identifies properties preferred by the user based on the occupancy or probability of appearance in the feature given that rating in each of the multiple ratings.

In addition, the information providing server 1 identifies the room type from the floor plan image for each room included in the property, and indicates, for each of the plurality of room models, the characteristics of the room type corresponding to the room model. A room model may be generated based on the feature information and the evaluation information. In this case, the user's preferences can be reflected in the property classifier for each room type.

1 information providing server 2 user terminal 11 system control unit 12 system bus 13 input/output interface 14 storage unit 14a member DB
14b Property DB
14c Evaluation information DB
14d Model DB
15 communication unit 111 display control unit 112 evaluation information acquisition unit 113 feature information acquisition unit 114 generation unit 115 determination unit NW network S property information provision system

Claims (12)

a display control means for displaying a floor plan image for each of a plurality of properties;
evaluation information acquiring means for acquiring, for each of the plurality of properties, evaluation information indicating a user's evaluation of the displayed floor plan image;
feature information acquiring means for acquiring, for each of the plurality of properties, feature information generated based on the floor plan image, the feature information indicating the features of rooms in the property;
generating means for generating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information;
determining means for determining properties to be presented to the user based on the generated model;
with
The evaluation information acquisition means acquires the evaluation information indicating any one of a plurality of evaluations including positive evaluations and negative evaluations,
The generating means generates, among the plurality of properties, a first probability distribution in which the characteristic information obtained for a room in the property indicating the positive evaluation is a random variable; a second probability distribution in which the characteristic information obtained for a room in the property indicating the negative evaluation is generated from among the properties;
The determining means acquires candidate property feature information generated based on a floor plan image of a candidate property to be presented to the user, and which indicates features of rooms in the candidate property. , from the first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model, in the second probability distribution indicated by the generated model, the obtained A property information providing device , wherein when a probability density difference obtained by subtracting a second probability density of candidate property characteristic information is equal to or greater than a predetermined value, the candidate property is determined as a property to be presented to the user . The display control means displays the floor plan image on the screen of a display device having a display unit having a screen and a detection unit that detects a contact position on the screen,
The evaluation information acquisition means acquires the evaluation information indicating a predetermined evaluation among a plurality of evaluations when movement of the contact position in a predetermined direction is detected on the screen displaying the floor plan image. 2. The property information providing device according to claim 1, wherein the property information providing device 3. The property information providing apparatus according to claim 1, wherein said generating means generates said model for each user. 4. The property information provision according to claim 3, wherein said determining means determines the property to be presented to said user based on said model generated for another user having the same attribute as said user's attribute. Device. The evaluation information obtaining means further obtains partial evaluation information indicating the user's evaluation of the portion specified by the user in the displayed floor plan image for at least one partially evaluated property among the plurality of properties. Acquired,
The generation means uses the acquired partial evaluation information for a first room corresponding to the designated portion among the rooms in the partially evaluated property, and generates the first room among the rooms in the partially evaluated property. 5. The property information providing device according to any one of claims 1 to 4, wherein the model is generated using the acquired evaluation information for a second room different from the first room. The display control means controls, for at least one of the re-displayed properties among the plurality of properties each having the floor plan image displayed, the floor plan to be displayed after a predetermined number of days has elapsed since the floor plan image was displayed. Redisplay the figure image,
The evaluation information acquisition means further acquires re-evaluation information indicating a re-evaluation by the user of the re-displayed floor plan image,
The generating means generates the model by using the acquired re-evaluation information for the re-displayed property among the plurality of properties, and by using the acquired evaluation information for a property different from the re-displayed property. 6. The property information providing device according to claim 1, wherein the property information providing device generates the property information. The feature information acquiring means specifies an area corresponding to the room from the floor plan image, and acquires the feature of the room based on the specified area. Item 7. The property information providing device according to any one of Items 1 to 6. 8. The property information providing apparatus according to claim 7, wherein said characteristic information acquiring means acquires the specified characteristics of said room based on the characteristics of said specified area. The feature information acquiring means is the feature indicating at least one of the area of the room, the length of a side of a figure corresponding to the range occupied by the room in the property, the interior angle of the figure, and the number of vertices of the figure. 9. The property information providing device according to any one of claims 1 to 8, wherein information is acquired. The model includes a plurality of type corresponding models corresponding to each of a plurality of types of rooms,
The feature information acquiring means specifies the type of the room from the floor plan image for each room included in the property, acquires the feature information for the room whose type is specified,
For each type of room included in the plurality of types of rooms, the generating means determines that the acquired evaluation information indicates the positive evaluation among the properties including the room of the type among the plurality of properties. The first probability distribution with the feature information acquired for the type of room in the property as a random variable, and the acquired evaluation information among the properties including the room of the type among the plurality of properties. generates the type correspondence model indicating the second probability distribution with the feature information obtained for the type of room in the property indicating the negative evaluation as a random variable;
The determining means calculates the probability density difference for each of the plurality of type-corresponding models, and calculates the probability density for the candidate property based on the plurality of probability density differences calculated for the plurality of type-corresponding models. 10. The method according to any one of claims 1 to 9, wherein a difference is calculated and whether or not to present the candidate property to the user is determined based on the probability density difference for the candidate property. Property information providing device described. In the property information providing method executed by a computer,
a display control step of displaying a floor plan image of each of a plurality of properties;
an evaluation information acquisition step of acquiring evaluation information indicating a user's evaluation of the displayed floor plan image for each of the plurality of properties;
a feature information acquisition step of acquiring feature information, which is generated based on the floor plan image for each of the plurality of properties and indicates features of rooms in the property;
a generation step of generating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information;
a determination step of determining properties to be presented to the user based on the generated model;
including
The evaluation information acquisition step acquires the evaluation information indicating any one of a plurality of evaluations including positive evaluations and negative evaluations,
In the generating step, a first probability distribution having as a random variable the characteristic information obtained for a room in a property in which the obtained evaluation information indicates the positive evaluation among the plurality of properties; a second probability distribution in which the characteristic information obtained for a room in the property indicating the negative evaluation is generated from among the properties;
The determining step acquires candidate property feature information generated based on a floor plan image of a candidate property to be presented to the user, the candidate property feature information indicating characteristics of rooms in the candidate property. , from the first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model, in the second probability distribution indicated by the generated model, the obtained A property information providing method, wherein, when a probability density difference obtained by subtracting a second probability density of candidate property characteristic information is equal to or greater than a predetermined value, the candidate property is determined as a property to be presented to the user. the computer,
a display control means for displaying a floor plan image for each of a plurality of properties;
Evaluation information acquisition means for acquiring evaluation information indicating a user's evaluation of the displayed floor plan image for each of the plurality of properties;
feature information acquiring means for acquiring, for each of the plurality of properties, feature information generated based on the floor plan image, the feature information indicating the features of rooms in the property;
generating means for generating a model for identifying a property preferred by the user based on the acquired evaluation information and the acquired feature information;
determining means for determining properties to be presented to the user based on the generated model;
function as
The evaluation information acquisition means acquires the evaluation information indicating any one of a plurality of evaluations including positive evaluations and negative evaluations,
The generating means generates, among the plurality of properties, a first probability distribution in which the characteristic information obtained for a room in the property indicating the positive evaluation is a random variable; a second probability distribution in which the characteristic information obtained for a room in the property indicating the negative evaluation is generated from among the properties;
The determining means acquires candidate property feature information generated based on a floor plan image of a candidate property to be presented to the user, and which indicates features of rooms in the candidate property. , from the first probability density of the obtained candidate property characteristic information in the first probability distribution indicated by the generated model, in the second probability distribution indicated by the generated model, the obtained A property information providing program, wherein when a probability density difference obtained by subtracting a second probability density of candidate property characteristic information is equal to or greater than a predetermined value, the candidate property is determined as a property to be presented to the user.

Images