JP2023177712A - Recommendation generation system, recommendation generation method, and recommendation generation program - Google Patents

Abstract

To provide a recommendation generation system capable of generating recommendations considering value information regarding a location of urban functions.SOLUTION: The recommendation generation system performs: graph creation processing of setting multiple nodes that indicate first location information and second location information in a graph by using a first table that associates the first location information and value information and a second table that associates the second location information including at least one or more out of the multiple pieces of first location information and urban functions, and creating a single graph that represents the urban functions as nodes connected to the first location information and the value information as nodes connected to the second location information; graph learning processing of representing a mutual relation between nodes in a vector space using machine learning techniques based on the graphs; and recommendation result generation processing of evaluating a path length between nodes from the vector space by evaluating the path length from distributed representation vector of the nodes of the first location information, the urban functions, the value information based on a specification of the first location information to generate a recommendation for the urban functions for the first location information.SELECTED DRAWING: Figure 12

Description

The present invention relates to a recommendation generation system, a recommendation generation method, and a recommendation generation program.

As background technology in this technical field, there is Japanese Patent Application Publication No. 2021-135722 (Patent Document 1). This bulletin states, ``We provide an information processing device and program that can quantify the relationship between information items such as analysis targets and products, even in social media where the relationship is sparse, and make it available for use.'' .” is stated.

JP 2021-135722 Publication

In recent years, technology has been developed to recommend urban functions suitable for individual locations such as various facilities and urban spaces to businesses and local governments that manage and provide urban functions, and to support the introduction or repurposing of urban functions. Attention has been paid.

One of the ways to realize this technology is to acquire location information of urban functions from map information, etc., and analyze the location relationships between urban functions, thereby determining the high degree of compatibility of individual points with surrounding city functions. One possibility is to generate recommendations based on .

The technology described in Patent Document 1 constructs a graph in which social media users are expressed as nodes and direct connection relationships between users on social media are expressed as edges. Machine learning techniques can be used to quantify relationships between all nodes, including nodes that do not exist. By applying the technology described in Patent Document 1, a graph is constructed in which urban functions implemented at a location are expressed as nodes, and urban functions located within a predetermined distance are expressed as edges, and a predetermined machine learning method is applied. By applying it, it is also possible to indirectly quantify the relationships among all city functions. Furthermore, we evaluate the degree of compatibility between the recommended city function and the surrounding city functions by comparing the estimated relationship between all city functions and the list of city functions that exist within a specified distance from the recommendation target point. It can also be calculated from This makes it possible to quantify the compatibility with surrounding city functions for all city functions that serve as recommendation candidates, so by sorting the city functions in descending order of affinity, recommendations of city functions for points can be generated. It is possible.

However, in this recommendation generation form, it is necessary to set in advance the range of surrounding city functions to be considered when calculating the affinity with the recommendation target point as a predetermined distance. In general, system users do not have information about the range of surrounding city functions that may have an affinity with the recommendation target point, so it is difficult to set an appropriate distance.

In addition, in this recommendation generation method, the graph that is input to the machine learning method is constructed using only the spatial location relationships of current urban functions. Therefore, the relationships between urban functions estimated using machine learning methods only indicate the strength of the locational correlation between urban functions, and do not reflect how urban function users evaluate the relationships between urban functions. It does not take into consideration whether or not they are doing so. For this reason, in this recommendation generation form, there is a possibility that a city function that has high location affinity with surrounding city functions but is undesirable for city function users will be recommended. To prevent this, it is necessary to consider value information regarding the location of urban functions.

An object of the present invention is to provide a technology that can make recommendations that take into consideration value information regarding the location of urban functions.

In order to solve the above problems, for example, the configurations described in the claims are adopted.
The present application includes a plurality of means for solving the above problems, and one example is a recommendation generation system that uses a computer having a processor and a memory to present urban functions for a place, the processor being , a first table that associates first place information with value information, and a second table that associates second place information that includes at least one or more of the plurality of first place information with urban functions. A node of a graph indicating the first location information and the second location information is set using a table of a graph construction process that constructs a single graph expressed as nodes connected to second location information; a graph learning process that expresses mutual relationships between nodes in a vector space using a machine learning method based on the graph; Evaluating the route length between nodes from the vector space by evaluating the route length from the node distribution expression vector of the first location information, the city function, and the value information based on the designation of the first location information. and a recommendation result generation process of generating a recommendation for the city function with respect to the first place information.

According to one aspect of the present invention, it is possible to provide a technology that allows recommendations that take into consideration value information regarding the location of urban functions.

FIG. 1 is a system configuration diagram of a recommendation generation system according to an embodiment. FIG. 1 is a hardware configuration diagram for realizing a recommendation generation system according to an embodiment. It is a figure which shows the example of the table of value information. It is a figure showing an example of a table of implemented city functions. FIG. 3 is a diagram showing an example of a spatial structure table. FIG. 2 is a diagram showing an example of a graph-structured table (node table). FIG. 3 is a diagram showing an example of a graph-structured table (edge table). FIG. 3 is a diagram illustrating an example of a distributed representation table. FIG. 3 is a diagram showing an example of a table of recommendation results. FIG. 7 is a diagram showing an example of a screen for performing initial settings for graph construction. FIG. 3 is a diagram showing an example of a screen for setting a recommendation target point. FIG. 3 is a diagram showing an example of a screen for displaying recommendation results. FIG. 3 is a flow diagram showing an operation for generating recommendations according to the first embodiment. 7 is a diagram illustrating an example of a screen for performing initial settings for graph construction in Practical Example 2. FIG. FIG. 3 is a flow diagram illustrating an operation for generating a recommendation according to a second embodiment. 12 is a diagram showing an example of a screen for performing initial settings for graph construction in Example 3. FIG. 12 is a flow diagram showing an operation for generating recommendations according to the third embodiment. FIG.

Examples will be described below with reference to the drawings. The embodiments described below do not limit the claimed invention, and all of the elements and combinations thereof described in the embodiments are essential to the solution of the invention. is not limited.

That is, the following description and drawings are illustrative for explaining the present invention, and are omitted and simplified as appropriate to clarify the explanation. The present invention can also be implemented in various other forms. Unless specifically limited, each component may be singular or plural.

The position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings.

In the following description, various information may be described using expressions such as "database," "table," and "list," but various information may be expressed using data structures other than these. "XX table", "XX list", etc. are sometimes referred to as "XX information" to indicate that they do not depend on the data structure. When describing identification information, when expressions such as "identification information", "identifier", "name", "ID", and "number" are used, these expressions can be replaced with each other.

When there are multiple components having the same or similar functions, the same reference numerals may be given different suffixes for explanation. However, if there is no need to distinguish between these multiple components, the subscripts may be omitted in the description.

In addition, in the following explanation, processing performed by executing a program may be explained, but the program is executed by a processor (for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit)), The processor may be the main body of the processing in order to perform the processing using appropriate storage resources (for example, memory) and/or interface devices (for example, communication ports). Similarly, the subject of processing performed by executing a program may be a controller, device, system, computer, or node having a processor. The main body of processing performed by executing a program may be an arithmetic unit or an arithmetic unit, and may include a dedicated circuit (for example, FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)) that performs specific processing. You can stay there.

A program may be installed on a device, such as a computer, from a program source. The program source may be, for example, a program distribution server or a computer-readable storage medium. When the program source is a program distribution server, the program distribution server includes a processor and a storage resource for storing the program to be distributed, and the processor of the program distribution server may distribute the program to be distributed to other computers. Furthermore, in the following description, two or more programs may be realized as one program, or one program may be realized as two or more programs.

In this embodiment, an example of a recommendation generation method in the recommendation generation system A1100 that simultaneously evaluates the affinity of a point with surrounding city functions and the relationship with a value index and recommends a city function will be described.

Referring to FIG. 1, the recommendation generation system of this embodiment will be described.
The recommendation generation system A1100 includes a value information DB (Database) A1111, an implemented city function DBA1112, a spatial structure DBA1113, a graph structure DBA1114, a distributed representation DBA1115, a recommendation result DBA1116, an automatic data update unit A1121, a graph construction unit A1122, and a graph learning unit A1123. , a recommendation result generation unit A1124, and a user interface A1125.

Here, the value information DBA 1111 is a DB that connects and holds value information characteristic of each area, which can be obtained from various statistical values and questionnaire surveys conducted among residents, and the aggregate area. An area ID that uniquely identifies the area name is attached to every record so that it can be linked with other DBs within the recommendation generation system A1100.

The implemented city function DBA 1112 is a DB for holding coordinates and the city functions implemented at the coordinate points, which are published by map preparation companies and public institutions that publish map information. A point ID that uniquely identifies the coordinate point is attached to every record so that it can be linked with the DB.

The spatial structure DBA1113 is information for linking the area ID included in the value information DBA1111 and the point ID included in the implemented city function DBA1112 based on the hierarchy of urban space within the recommendation generation system A1100. It also holds the area name corresponding to the area.

The graph structure DBA 1114 is information for recording the structure of a single graph including various information items of points, areas, city functions, and value information necessary for generating recommendation results, and is composed of a node table and an edge table. Ru.

The distributed representation DBA1115 is information for recording the characteristics of each graph node obtained by inputting the graph to a machine learning method, and includes the node ID and the distributed representation representing the characteristics of the graph node as a high-dimensional real vector. Retain expression.

The recommendation result DBA 1116 is information for recording the result of evaluating the relationship between points, city functions, and value information from the node distribution representation using the route length. It also holds various information used to calculate .

The automatic data update unit A1121 is a function for updating information used in this system, and collects data via a network A1300 such as a local network or the Internet, and collects data such as value information DBA1111, implemented city function DBA1112, and spatial structure DBA1113. Register your information.

The graph construction unit A1122 is a function that enables the quantification of relationships using machine learning techniques by constructing a single graph that expresses the connection relationships of points, areas, city functions, and value information necessary for recommendation generation. , graph construction processing is performed using the value information DBA 1111, implemented city function DBA 1112, and spatial structure DBA 1113, and the result is output to the graph structure DBA 1114. Initial settings for graph construction are performed by the system user using screens such as those shown in FIG. 9, FIG. 13, or FIG. 15.

The graph learning unit A1123 is a function for obtaining distributed representations for all nodes from the graph constructed by the graph construction unit A1122. The graph structure DBA 1114 is input, a distributed representation is learned by applying a machine learning method, and the result is output to the distributed representation DBA 1115.

The recommendation result generation unit A1124 receives the distributed representation DB1115 as an input, and is a function for evaluating the relationship between points, urban functions, and value information using route lengths in a vector space. The evaluation result of the urban function of the point considering the affinity with the surrounding city function and the relationship with the value information is output to the recommendation result DBA 1116.

When the user interface A1125 receives information from the user terminal A1200, it transmits the data to the graph construction unit A1122, graph learning unit A1123, and recommendation result generation unit A1124 of the recommendation generation system A1100.

The user terminal A1200 is a terminal for a user to configure settings for the recommendation generation system A1100, and includes, for example, a PC, a mobile phone, and the like. The user uses this terminal to connect to the recommendation generation system A1100.

Referring to FIG. 2, a hardware configuration for realizing the embodiment will be described.
The hardware configuration shown in FIG. 2 is a hardware configuration H2000 for implementing the recommendation generation system A1100 and the user terminal A1200.

The hardware includes a central processing unit H2002, a main storage device H2003, an internal bus H2004, a bus I/FH2005, an external bus H2006, a screen output device I/FH2007, a user input device I/FH2008, a mass storage device H2009, and a communication device I. /FH2010 and input/output device I/FH2011.

The central processing unit H2002 is a device for performing calculations such as program execution, and for realizing the operations of the graph construction unit A1122, graph learning unit A1123, and recommendation result generation unit A1124 of the recommendation generation system A1100 in FIG. 1, for example. used.

The main storage device H2003 is used as a processing area during program execution and a temporary data storage area. For example, it temporarily stores basic programs such as an OS (Operating System), programs and temporary information held by the recommendation generation system A1100 and user terminal A1200 for realizing processing in each device.

The central processing unit H2002 and the main memory H2003 are connected by an internal bus H2004, and the internal bus H2004 is connected to an external bus H2006 via a bus I/FH2005.

The external bus H2006 is connected to a screen output device I/FH2007, a user input device I/FH2008, a mass storage device I/FH2009, a communication device I/FH2010, and an input/output device I/FH2011. The bus I/FH 2005 and the internal bus H2004 are used to mediate data input/output between the central processing unit H2002 and the main storage H2003.

The screen output device I/FH2007 is an I/F for connecting to the display H2100, and outputs information obtained through the external bus H2006 to the display H2100. The display H2100 is a device that can display, for example, the setting screen shown in FIG. 9, and can also display information for operating the basic program.

The user input device I/FH2008 is an I/F for connecting with the user input device H2200, and outputs information input from the user input device to the external bus H2006. The user input device is a device for receiving input from the user, such as a keyboard or a mouse, and information input from the user input device is output to the user input device I/FH 2008.

The mass storage device I/FH2009 is an I/F for connecting with the mass storage device H2300, and mediates data input/output from the mass storage device to the external bus H2006. The large-capacity storage device is, for example, a device such as an HDD (Hard Disk Drive), and stores basic programs for realizing the functions of the recommendation generation system A1100, temporary storage information of processing results during program execution, the recommendation generation system A1100, and user terminals. Programs for realizing processing in each device such as the A1200, and information such as value information DBA1111, implemented city function DBA1112, spatial structure DBA1113, graph structure DBA1114, distributed expression DBA1115, and recommendation result DBA1116 in the recommendation generation system A1100 are transferred to the device's power supply. Save it when it is turned off or when the program is not running. Furthermore, when executing various processes, the central processing unit H2002 reads these programs and data into the main storage device H2003 and executes the programs. The mass storage device may be a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

The communication device I/FH2010 is an I/F for connecting with the communication device H2400, and mediates data input/output from the communication device to the external bus H2006. The communication device is a device for connecting to an external server device or the like via, for example, Ethernet (registered trademark), and can communicate with a system connected to the network A1300 between the recommendation generation system A1100.

The input/output device I/FH2011 is an I/F for connecting with the input/output device H2500, and mediates data input/output from the input/output device to the external bus H2006. The input/output device includes a drive device that can read and write external media, a finger vein reader for user authentication, and the like.

FIG. 3 shows an example of a table of value information DBA 1111. Value information name T10, area name linked to the above value information T11, score T12 indicating the relative evaluation value of the corresponding area name with respect to the value information, area IDT13 for uniquely identifying the area name, and prefecture to which the area belongs T14. include. By setting such a value, it is possible to make the area name and value information correspond, for example, the value index "value information: ease of raising children" corresponds to "area name: ____ town". can. In Figure 3, for example, "〇〇 town" identified by area ID "1" belonging to the prefecture "Kanagawa prefecture" has a score of "10" for "ease of raising children". It shows. The value of the score is set, for example, to a value representing the ease of raising children as a percentage obtained through a questionnaire survey.

FIG. 4 shows an example of a table of the implemented city function DBA 1112. It includes a point IDT20 for uniquely specifying the point, coordinates T21 for specifying the location of the point, an implementation city function T22 of the point, and a prefecture T23 to which the point belongs. By setting such a value, it is possible to associate locations with implemented city functions, such as "implemented city function: convenience store" for "location ID: 1", for example. For example, in Figure 4, the coordinates of the point identified by point ID "1" belonging to the prefecture "Kanagawa Prefecture" are "(36.30, 140.46)", and there is a "convenience store". It is shown that.

FIG. 5 shows an example of a table of the spatial structure DBA 1113. It includes a point IDT31 defined in FIG. 4, an area IDT32 defined in FIG. 3, and an area name T33. By setting such a value, it is possible to search for a point ID belonging to an area ID. In FIG. 5, for example, in the prefecture "Kanagawa Prefecture", the point identified by point ID "1" and the point identified by point ID "2" are located in "〇〇 town" identified by area ID "1". ” indicates that it belongs to

6A and 6B show examples of tables of the graph structure DBA 1114. The graph structure DBA 1114 manages nodes and edges, which are constituent elements of a graph, using a node table and an edge table, respectively.

The node table shown in FIG. 6A is for recording a list of nodes forming the graph, and includes a node IDT41, a node name T42, and a node attribute T43. The node ID is used to uniquely identify a node, and the same node ID is never set for multiple nodes. The node name describes the information item indicated by the node, and the same node name may be set for multiple nodes. The node attribute describes the attribute of the information item indicated by the node name, and takes one of the following values: "value information," "city function," "location," and "area." In FIG. 6A, for example, the node with the node name "Livability" identified by the node ID "1" belongs to the node attribute "Value information".

The edge table shown in FIG. 6B is for recording connection relationships between nodes, and includes an edge IDT 51 and an edge configuration node T52. Furthermore, if information regarding the degree of edge connection strength is available, the above information can be expressed by setting a numerical value to the edge weight T53 in the edge table. In FIG. 6B, for example, the edge identified by edge ID "2" records the connection relationship between the node identified by node ID "3" and the node identified by node ID "4", This indicates that edge weights are not set.

FIG. 7 shows an example of a table of the distributed representation DBA 1115. It includes a node IDT61 defined in FIG. 6A, and a distributed expression T62 that is information expressing the characteristics of the node ID as a high-dimensional real vector. In FIG. 7, for example, the node identified by the node ID "1" is represented by the distributed representation (20.73, 28.56, 20.85, . . . ).

FIG. 8 shows an example of a table of the recommendation result DBA 1116. It holds a city function T71 that is a recommendation candidate, a route length T72 that is an evaluation result of the city function, and information T73 and T74 used to calculate the route length. FIG. 8 shows, for example, that the evaluation result for the city function "Super" is expressed as a route length of "89.87". Moreover, the parameters d(p,f) and d(f,v) used in the calculation formula for calculating the predetermined path length, which will be described later, are "16.35" and "73.35", respectively. 52".

FIG. 9 shows an initial setting screen for setting the graph construction unit A1122. The setting screen W10 includes a screen W11 for selecting a target prefecture, a screen W12 for selecting value information to be introduced, and a screen W13 for previewing a graph according to the selections on the screen W11 and screen W12. have. Using this screen, the system user can select one target prefecture and one or more pieces of value information, and also check a sample of the graph generated according to the setting information on screen W11 and screen W12 on screen W13. be able to. The information displayed on the screen W13 can be displayed by performing the processes of S130 to S150 shown in FIG. 12, which will be described later. In this example, when value information such as "ease of raising children", "ease of walking around", and "liveliness" is selected for an area belonging to the prefecture "Kanagawa Prefecture", the value information DBA1111 shown in Figure 3 is selected. By reading out the implemented city function DBA 1112 shown in Figure 4 and the spatial structure DBA 1113 shown in Figure 5 and constructing a graph as described later, it is possible to connect all "points", "urban functions", and "value indicators". A single graph can be constructed to represent relationships. In addition, in Figure 9, the value information "ease of raising children" and the area name "XX shopping district" have a high relationship between nodes, so a higher weight value of "30" is set than other nodes. It is shown that.

In the past, for example, it was possible to construct a graph that expressed the relationship between "urban functions" and "value indicators," but point-specific information was missing and the same urban function was recommended at many points. It had become. However, in this example, the map graph is interpreted as a graph in which "space" and "urban functions" are fixed and does not contain value information, and the dependent characteristics of value indexes and urban functions mediated by urban space are We are constructing graphs focusing on In other words, by understanding that "urban functions" and "value information" are in a relationship where they influence each other through space, we can separate "space" and "urban functions," and furthermore, we can separate "value information" from "spatial information." It is possible to realize a graph representation linked to ``.

FIG. 10 shows a recommended location setting screen for configuring the recommendation result generation unit A1124. The setting screen W20 includes a screen W21 for selecting a target point for recommendation, a screen W22 for setting a value index to be evaluated at the time of recommendation and the importance of each value index, and settings for the screen W21 and screen W22. It has a screen W23 for visualizing the relationship between each information item according to the information. Using this screen, the system user can select one recommendation target point and one or more pieces of value information, as well as set the importance level of the selected value index numerically. In addition, the interrelationships between locations, city functions, and value indicators that are visualized according to the setting information on the screen W11 and the screen W12 can be confirmed on the screen W23. The information displayed on the screen W23 can be displayed by performing the process of S170 shown in FIG. 12, which will be described later.

FIG. 11 shows a result display screen for visualizing the output results of the recommendation result generation unit A1124. The setting screen W30 includes a screen W31 for displaying the city function recommendation results and a screen W32 for visualizing the relationship with surrounding city functions. Using this screen, the system user can check the city function recommendation results for the point set on the setting screen W20, as well as the visualization results of the relationship with surrounding city functions when the city function is introduced at the target point. be able to. The information displayed on the screen W32 can be displayed by performing the process of S180 shown in FIG. 12, which will be described later.

Referring to FIG. 12, an operation for recommending a city function in the first embodiment will be described.

Step S110 is an operation in which input of the target prefecture and introduction value information is received from the system user in order to determine the reading range of the database. The target prefecture information and target value information are read from the screen W11 and the screen W12 transmitted via the screen W11 and W12.

Step S120 is an operation of searching and reading records of the target prefecture from a plurality of databases in order to obtain data required for recommendation generation from the databases. The graph construction unit A1122 searches for and reads a record in which the value information name T10 of the value information DBA 1111 matches the target value information and the prefecture T14 matches the target prefecture information. Next, records in which the prefecture T23 of the implemented city function DBA 1112 matches the target prefecture information and records in which the prefecture T34 of the spatial structure DBA 1113 matches the target prefecture information are searched, and the corresponding records are read.

Step S130 is an operation in which the graph construction unit A1122 determines the nodes and edges that constitute the graph in order to construct a graph expressing the mutual relationships between points, city functions, and value information, and includes multiple operations starting from step S131. Consists of.

Step S131 is an operation for creating nodes that make up the graph, and information items that make up the nodes are created using the records read out from the value information DBA 1111, implemented city function DBA 1112, and spatial structure DBA 1113 in step S120. , determines the node name and node attribute, and registers them in the node table of the graph structure DBA 1114.

Specifically, the record read by the value function DBA 1111 is converted into a node with value information name T10 set as the node name, and the value of "value information" is set as the node attribute. The record read from the value function DBA 1111 is converted into a node whose node name is area IDT 13, and "area" is set as the node attribute. Furthermore, the record read from the implemented city function DBA 1112 is converted into a node whose node name is the point IDT 20, and "point" is set as the node attribute. Next, a list of values set in the implemented city function T22 is created from the records read from the implemented city function DBA 1112, and duplicate elements in the list are deleted. Next, the elements of the list are converted to nodes, and "city function" is set as the node attribute. The node name and node attribute information created by the operations up to this point are registered in the graph structure DBA 1114, unique node IDTs 41 are assigned to all records in the graph structure DBA 1114, and step S131 is ended.

Step S132 is an operation of setting an edge between nodes whose node attributes are "value information" and "area." Referring to the value information DBA 1111 read out in step S120, for all records, the node pointed to by the value information name T10 and area IDT13 of the record is specified from the node table of the graph structure DBA 1114, and the identified two nodes Repeat the process of setting an edge in between. Next, the edge constituent nodes identified by the above operation are registered in the edge table of the graph structure DBA 1114, and step S132 is ended.

Step S133 is an operation of setting weight information for the edge set in step S132. If information regarding the degree of connection strength is available for the edge connection relationship set in step S132, that information is set as the edge weight. For example, when the value of the score T12 described above is equal to or greater than a predetermined threshold, it is determined that the relationship between the nodes is high, and weight information can be set. Edges for which information about the degree of connection strength is not available may not have edge weights set.

Step S134 is an operation of setting an edge between nodes whose node attributes are "area" and "point". With reference to the spatial structure DBA 1113 read in step S120, for every record, the node pointed to by the point IDT 31 and area IDT 13 of the record is specified from the node table of the graph structure DBA 1114, and between the two specified nodes. Repeat the operation to set the edge. Next, the edge constituent nodes identified by the above operation are registered in the edge table of the graph structure DBA 1114, and step S134 is ended.

Step S135 is an operation for setting edges between nodes whose node attribute is "point". The record of the implemented city function DBA 1112 read in step S120 is input, the adjacency between nodes with the "point" attribute is determined from the coordinates T21 included in the record, and edges are set between nodes having an adjacency relationship. The process of setting edges from the above coordinates can be performed using a general method such as a neighborhood graph.

Step S136 is an operation of setting an edge between nodes whose node attributes are "point" and "urban function." Referring to the implemented city function DBA 1112 read out in step S120, the node pointed to by the point IDT 20 and the implemented city function T22 of the record is specified from the node table of the graph structure DBA 1114 for all records. Repeat the operation of setting edges between nodes. Next, the edge constituent nodes identified by the above operation are registered in the edge table of the graph structure DBA 1114, and step S136 is ended.

Steps S140 and S150 are operations for the graph learning unit A1123 to obtain distributed representations of nodes from the graph by applying a graph embedding method, which is one of the machine learning methods. A method for obtaining a distributed representation from a graph structure is, for example, the method disclosed in Grover and Leskovec, “node2vec: Scalable Feature Learning for Networks,” Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining, 2016. etc. can be used.

Graph embedding methods generally extract distributed representations from graphs by performing two processes in this order: sampling graph nodes according to certain rules and calculating node distributed representations from the sampling results. possible. In the embodiment of the present disclosure, graph information is read from the graph structure DBA 1114, nodes of the graph are sampled in step S140, and a node distribution expression is calculated from the sampling results in step S141. Subsequently, the calculated node distribution representation is recorded in the distribution representation A1115.

Step S160 is an operation of accepting input of recommended points and value information to be introduced, in which the recommendation result generation unit A1124 receives setting information of screens W21 and W22 transmitted from the user terminal A1200 via the user I/FA 1125. Read out.

Step S170 is an operation for the recommendation result generation unit A1124 to calculate the evaluation value at the recommendation point for the city function that is a recommendation candidate and is not registered in the recommendation result DBA 1116. In this disclosure, the distributed representation vector of the node recorded in the distributed representation DBA 1115 is input, and in the vector space, the path length corresponding to the travel distance when tracing multiple nodes is calculated, and this is used as the evaluation value of the urban function. .

The path length calculation formula in the present disclosure is, for example, the following formula (1).

However, d(x,y) is a function representing the length of the line segment connecting the distributed representation vector of node x and the distributed representation vector of node y in the vector space, and p is an arbitrary number whose node attribute is "point". , f indicates an arbitrary node whose node attribute is "city function", and v indicates an arbitrary node whose node attribute is "value information".

Equation (1) takes any node of location, urban function, and value information as input, and when moving from a node with the "location" attribute to a node with the "urban function" attribute, and then to a node with the "value information" Calculate the path length L.

The first term in equation (1) represents the degree of affinity between the target point and the urban function, and the higher the affinity, the smaller the value.

The second term in equation (1) represents the strength of the relationship between the urban function and the value information, and the stronger the relationship, the smaller the value.

λ included in the second term of equation (1) is an adjustment parameter, and the index importance of the screen W22 can be used as its value. The adjustment parameters can adjust the weighting of the first term indicating the degree of affinity between the target point and the urban function, and the second term indicating the strength of the relationship between the urban function and the value information.

By using equation (1), it is possible to calculate an evaluation value for any recommended city function by simultaneously considering its affinity with the function of the recommended point and its relationship with the value index. .

Although formula (1) considers only one piece of evaluation information, it is also possible to consider a plurality of value indicators. In that case, equation (1) can also be replaced by equation (2).

Here, n is the number of value indicators to be considered, and v _i is a distributed expression vector of the i-th (i is a natural number equal to or less than n) value information node.

The second term in Equation (2) represents the sum of the strengths of relationships between urban functions and all value information to be considered. λ _i is an adjustment parameter for the i-th value information, and the value of this parameter can adjust the weighting of importance between each value information.

The city function, the calculated route length, and the values of d(p,f) and d(f,v) are recorded in the recommendation result DBA 1116, and step S170 is ended.

Step S171 is an operation in which the recommendation result generation unit A1124 determines whether the route length calculation shown in step S170 has been completed for all city functions that are recommendation candidates. If the result is recorded in the DBA 1116 (step S171; Yes), the process advances to step S180; if not (step S171; No), the process returns to step S170.

Step S180 is an operation in which the recommendation result generation unit A1124 takes the recommendation result DBA1116 as input, sorts the city functions in order of shortest route length, and sends the results to the user terminal A1200 via the user I/FA 1125. . User terminal A1200 presents the recommendation results to the system user using screen W31 in result display screen W30. For example, regarding point 1 selected on screen W21 in FIG. 10, the above-mentioned path length L is the overall match degree, the above-mentioned d(p,f) is the current match, and the above-mentioned d(f,v) is the value match. Present each.

At this time, the recommendation result generation unit A1124 can also calculate the relationship between the nodes of the "urban function" attribute recorded in the distributed representation DBA1115 using the inter-node distance, and the screen in the result display screen W30 It can also be visualized on a map using W32.

For the convenience of the system user, in step S110, a preview of the graph that is immediately constructed can be displayed on the screen W13 according to the selections on the screen W11 and the screen W12.

In addition, in step S160, steps S170 and S171 are immediately executed according to the selection contents on screen W12 and screen W12, and distributed expression vectors and route lengths among points, city functions, and value information are visualized on screen W23. It can also be displayed. At this time, in order to visualize the high-order distributed representation vector on the screen, for example, the dimensionality reduction algorithm disclosed in Van der Maaten and Hinton, “Visualizing data using t-SNE”, Journal of machine learning research, 2008. t-SNE (t-Distributed Stochastic Neighbor Embedding) or the like can be used.

In Example 1, graph construction was performed based on information in the database. However, it is also conceivable that system users possess knowledge that is not included in the database but can be utilized for graph construction.

In this embodiment, an example of a recommendation generation system that can reflect knowledge held by a system user as a priori knowledge edge and automatically determine weight information of a priori knowledge edge will be described.

In the recommendation generation system A1100 in FIG. 1, descriptions of parts having the same functions as the configurations with the same reference numerals shown in FIG. 1 described above will be omitted.

FIG. 13 shows an initial setting screen for performing the settings of the graph construction unit A1122 in the second embodiment, which were performed in FIG. 9 in the first embodiment. In addition to the screens in the setting screen W10 described in FIG. 9, the setting screen W40 includes a screen W43 for accepting additions of foresight knowledge edges by the system user.

The screen W43 is provided with a setting area W431, a check box W432, an input area W433, and a button W434.

The setting area W431 is an area that accepts selection of a combination of nodes for setting a priori knowledge edge, and makes it possible to add prior knowledge as an edge of a graph.

The check box W432 is a check box for setting whether or not the recommendation generation system A1100 automatically sets the weight information for the edge set in the setting area W431. If the system user wishes to set the edge weight information himself using the check box W432, he can input the edge weight information using the input area W433.

After setting various information on screen W43, by pressing button W434, the setting information is transmitted to graph construction section A1122. The graph construction unit A1122 registers various information set on the screen W43 in the edge table of the graph structure DBA1114.
The system user can check a sample graph including the edge 1301 added according to the setting information on the screen W43 on the screen W44.

Referring to FIG. 14, an operation for recommending city functions in the second embodiment will be described.

In the second embodiment, operations similar to those in the first embodiment are performed except for steps S237, S2371, S2372, and S2373 described below.

Step S237 is an operation for the graph construction unit A1122 to determine whether the system user's foresight knowledge has been added, and if input of foresight knowledge from the user is accepted on the screen W43 (step S237; If yes), the process advances to step S2371, and if not (step S237; No), the series of processes included in step S230 is ended.

Step S2371 is an operation for the graph construction unit A1122 to reflect the system user's prior knowledge, which is not included in the DB, in the graph. Based on the prior knowledge information regarding the relationship between specific nodes received from the system user by the screen W43, the node is specified from the graph structure DBA 1114, and edges are set between the nodes. Next, the edge and its constituent nodes are registered in the edge table of the graph structure DBA 1114, and step S238 is ended.

Step S2372 is an operation for the graph construction unit A1122 to determine whether the system user wishes to automatically set the edge weight information added in step S2371. If the setting is desired (step S2372; Yes), the process advances to step S2373; if not (step S2372; No), the series of processes included in step S230 is ended.

Step S2373 is an operation for the graph construction unit A1122 to automatically set the edge weight information added in step S2371. In the automatic setting of edge weight information, first, mutual arrival probabilities between all nodes are calculated from the graph constructed up to step S236. The mutual arrival probability between nodes can be solved by considering a Markov process in which the nodes of the graph are states and the edge weights are transition probabilities, and by applying a general method for deriving the steady state of the Markov process.

Next, the attributes of the two nodes forming the edge added in step S2371 are acquired (for example, "urban function" and "value information", etc.), and all combinations between nodes having those attributes are enumerated. (For example, a combination of a node with an attribute of "urban function" and a node with an attribute of "value information"). Furthermore, among the node combinations enumerated by the above operation, the node combination with the highest probability of arrival between nodes is identified by referring to the determined mutual probability of arrival between nodes. The weight information of the a priori knowledge edge is set so that the probability of mutual arrival between nodes in the identified node combination is equal to the probability of mutual arrival between the nodes forming the a priori knowledge edge.

By the operation of step S2373, it is possible to guarantee through mutual arrival probability that the inter-node relationship for which prior knowledge exists is the most important compared to other comparable inter-node relationships.

In the first embodiment, sampling of the graph was performed for the entire graph. However, in the graph constructed in Example 1, since it is foreseen that nodes with the "area" and "urban function" attributes will be connected to a large number of nodes with the "point" attribute, nodes with the "point" attribute will be It may be difficult to extract the relationships between the two.

In this example, if there is a concern that the relationship between nodes with the same attribute will not be sufficiently estimated by machine learning methods due to tight coupling with many nodes with other attributes, a specific attribute (for example, "point" attribute) An example of a recommendation generation system that can perform priority sampling of nodes will be described.

In the recommendation generation system A1100 in FIG. 1, descriptions of parts having the same functions as the configurations with the same reference numerals shown in FIG. 1 described above will be omitted.

FIG. 15 shows an initial setting screen for performing the settings of the graph construction unit A1122 in the third embodiment, which were performed in FIG. 9 in the first embodiment. In addition to the screens in the setting screen W50 described in FIG. 9, the setting screen W50 includes a screen W53 for accepting a system user's request for prioritized sampling of nodes with specific attributes. The screen W53 is provided with a check box W531, a selection area W532, and an input area W533.

The check box W531 is a check box for the system user to set whether or not to implement priority sampling. If the system user wishes to implement priority sampling using the check box W531, he or she can input priority sampling setting information using the selection area W532 and input area W533.

The system user can confirm, on the screen W54, a graph sample in which the location where priority sampling is to be performed is highlighted 1501 according to the setting information on the screen W53.

Referring to FIG. 16, an operation for recommending city functions in the third embodiment will be described.

In the third embodiment, the same operations as in the first embodiment are performed except for steps S350, S351, and S360 described below.

Step S350 is an operation in which the graph learning unit A1123 determines whether the user requests sampling between nodes of a specific attribute. If the setting has been made (step S350; Yes), the process advances to step S351; otherwise (step S350; No), the process advances to step S360.

Step S351 is an operation in which the graph learning unit A1123 constructs a subgraph consisting only of nodes having specific attributes, and performs sampling from the subgraph. At this time, the number of times of sampling from the subgraph can be determined from the ratio specified by the system user in the input area W533.

Step S360 is an operation in which the graph learning unit A1123 calculates a node distribution representation from the sampling results. In the first embodiment, the node distribution representation was calculated using only the sampling results for the entire graph, but in this embodiment, the node distribution representation was calculated using the sampling results for the entire graph (step S340) and nodes with specific attributes. A node distribution representation is calculated using both of the sampling results (step S351) for a subgraph consisting of only as learning data.

Each of the embodiments has been described above, but in the first embodiment, as explained using FIG. The processor includes a first table (value information DBA1111) that associates first location information (points) and value information, and a second location that includes at least one or more of the plurality of first location information. Using a second table (implemented city function DBA1112) that associates information (area) and city functions, set nodes of a graph that point to the first location information and the second location information, As shown in FIGS. 6A and 6B, a graph that constructs a single graph expressing the urban function as a node connecting to the first location information and the value information as a node connecting to the second location information a construction process, a graph learning process that expresses the interrelationships between nodes in a vector space using a machine learning method based on the graph, and a graph learning process that expresses the interrelationships between nodes in a vector space based on the graph, and based on the designation of the first location information (point 1), By evaluating the route length from the node distribution expression vector of information, the above urban function, and the above value information, the route length between nodes is evaluated from the above vector space, and the above urban function is evaluated for the above first location information. Recommendation result generation processing for generating recommendations is performed. As a result, when recommending urban functions for a point, the range and degree of surrounding city functions that may have compatibility with the recommended point are automatically estimated, and both the compatibility with surrounding city functions and the relationship with value information are evaluated. can be generated by explicitly adjusting the weighting of

In addition, in the second embodiment, as explained using FIG. 14 and the like, in the graph construction process, the processor specifies between nodes of the graph based on additional designation of a priori knowledge that is knowledge possessed by the user. A prior knowledge edge (edge 1301) is added, and weight information of the added prior knowledge edge is set by comparison using arrival probabilities between graph nodes. This allows expert knowledge that cannot be obtained from data to be reflected in the graph.

Further, in the third embodiment, as explained using FIG. 16 etc., in the graph learning process, the processor extracts nodes with specific attributes (points) from the graph as a subgraph, and By executing the process of generating learning data from the graph, learning data is generated from both the graph and the subgraph. As a result, even if an area node or a function node is connected to a large number of point nodes, relationships between point nodes can be easily extracted.

The present invention is not limited to the above-described embodiments as they are, and in the implementation stage, components may be modified and embodied without departing from the gist thereof, or multiple components disclosed in the above-described embodiments may be embodied. These can be implemented in appropriate combinations.

A1100...Recommendation generation system, A1200...User terminal, A1300...Network, A1111...Value information DB, A1112...Implemented city function DB, A1113...Spatial structure DB, A1114...Graph structure DB, A1115...Distributed expression DB, A1116...Recommendation Result DB, A1121...Data automatic update unit, A1122...Graph construction unit, A1123...Graph learning unit, A1124...Recommendation result generation unit, A1125...User I/F

Claims (7)

A recommendation generation system that presents urban functions for a location using a computer having a processor and a memory,
The processor executes a program,
a first table that associates first place information with value information; and a second table that associates second place information that includes at least one or more of the plurality of first place information with urban functions. A node of a graph indicating the first location information and the second location information is set using a table, a node connecting the city function to the first location information, and a node connecting the value information to the first location information. a graph construction process that constructs a single graph expressed as a node connected to the location information of 2;
Graph learning processing that expresses mutual relationships between nodes in a vector space based on the graph using a machine learning method;
Evaluating the route length between nodes from the vector space by evaluating the route length from the node distribution expression vector of the first location information, the city function, and the value information based on the designation of the first location information. and a recommendation result generation process that generates a recommendation of the city function for the first location information;
A recommendation generation system that performs the following. In the graph construction process, the processor adds a prior knowledge edge between the nodes of the graph based on the addition designation of prior knowledge that is knowledge possessed by the user, and adds weight information of the added prior knowledge edge to Set by comparison using arrival probabilities between graph nodes,
The recommendation generation system according to claim 1, characterized in that: In the graph learning process, the processor extracts a node having a specific attribute from the graph as a subgraph, and generates learning data from the extracted subgraph. Generate learning data from both sides,
The recommendation generation system according to claim 1, characterized in that: A recommendation generation method for presenting urban functions for a location using a computer having a processor and a memory, the method comprising:
a first table that associates first place information with value information; and a second table that associates second place information that includes at least one or more of the plurality of first place information with urban functions. A node of a graph indicating the first location information and the second location information is set using a table, a node connecting the city function to the first location information, and a node connecting the value information to the first location information. Build a single graph expressed as nodes connected to the location information of 2,
Based on the graph, the mutual relationships between nodes are expressed in a vector space using machine learning techniques,
Evaluating the route length between nodes from the vector space by evaluating the route length from the node distribution expression vector of the first location information, the city function, and the value information based on the designation of the first location information. and generating a recommendation for the city function with respect to the first location information;
A recommendation generation method characterized by the following. In constructing the graph, a priori knowledge edge is added between the nodes of the graph based on the additional designation of the prior knowledge that the user has, and the weight information of the added prior knowledge edge is applied to the weight information between the graph nodes. Set by comparison using arrival probability,
5. The recommendation generation method according to claim 4. In learning the graph, nodes with specific attributes are extracted from the graph as a subgraph, and learning data is generated from both the graph and the subgraph by executing a process of generating learning data from the extracted subgraph. generate,
5. The recommendation generation method according to claim 4. A computer having a processor and memory,
a first table that associates first place information with value information; and a second table that associates second place information that includes at least one or more of the plurality of first place information with urban functions. A node of a graph indicating the first location information and the second location information is set using a table, a node connecting the city function to the first location information, and a node connecting the value information to the first location information. a graph construction process that constructs a single graph expressed as a node connected to the location information of 2;
Graph learning processing that expresses mutual relationships between nodes in a vector space based on the graph using a machine learning method;
Evaluating the route length between nodes from the vector space by evaluating the route length from the node distribution expression vector of the first location information, the city function, and the value information based on the designation of the first location information. and a recommendation result generation process that generates a recommendation of the city function for the first location information;
A recommendation generation program that executes.

Images