JP2010020490A - Device for providing information on unfamiliar place, and method for providing information on unfamiliar place - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for providing information on an unfamiliar place and a method for providing the information on the unfamiliar place, which provide information usable for a user by extracting a place which is highly likely to be visited by the user in the future even if the user has not visited there and providing the user with the information on the place. <P>SOLUTION: This device for providing the information on the unfamiliar place has: an action history data acquisition part 152 acquiring action history information of the user; a place group creation part 154 creating a place group composed of a plurality of places to which the user has visited from the action history information; a place group decision part 174 deciding a place group of the plurality of created place groups to which an arbitrary place belongs; and a probability calculation part 178 calculating probability that the user visits a spot in the arbitrary place under the condition that the arbitrary place belongs to the decided place group. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an unvisited location information providing apparatus and an unvisited location information providing method.

  2. Description of the Related Art Conventionally, there is a technique for acquiring a user's behavior pattern according to the situation such as time and place from the behavior history and providing information that matches the current situation of the user based on the acquired behavior pattern. Such a behavior prediction technique is disclosed in Patent Document 1, for example. In this technique, first, a logical rule that represents a user's behavior pattern is defined from the collected user's past behavior data using fuzzy set theory. And the logic rule was applied to the given fact, and the future action of the user was predicted.

JP 2001-134706 A

  However, the conventional behavior prediction technique as described in Patent Document 1 predicts the user's future behavior within the range of the situation that matches the situation recorded in the behavior history, and is limited. . Therefore, if a situation that matches the current situation of the user is not recorded in the action history, even if a recommendation for a future action is requested by the user, it is not possible to make a recommendation according to the current situation. There was a problem.

  As an example of the current situation, a case where the user is present will be described. When the user is currently in an arbitrary place, the user may match the place recorded in the action history. However, the location information is so diverse that it is possible for a user to be in a location that has never been visited before. In addition, users often want to recommend future actions in places they have never visited. In this way, in the situation where users are expected to frequently use recommendation requests for future actions, it is not possible to make recommendations according to the current situation. There was a problem that it could not be reached.

  Therefore, the present invention has been made in view of the above problems, and the object of the present invention is a point where a user is highly likely to visit in the future even in places where the user has never visited. New and improved information device for unvisited sites and information provision for unvisited sites that can provide information useful to the users by extracting the information and providing the information of the points to the users It is to provide a method.

  In order to solve the above problems, according to an aspect of the present invention, an action history data acquisition unit that acquires action history information of a user, and a set of a plurality of places that the user has visited from the action history information A location group creation unit that creates a location group, a location group determination unit that determines to which location group an arbitrary location belongs among the created multiple location groups, and a location from which an arbitrary location has been determined When it belongs to a group, an unvisited site information providing apparatus is provided that includes a probability calculating unit that calculates a probability that a user visits a point in an arbitrary place.

  You may provide the place extraction part which extracts the 1 or several point in the arbitrary places where the probability that the said user visits satisfy | fills a predetermined threshold value from the several points in an arbitrary place.

  The place group determination unit may determine a place group to which an arbitrary place belongs based on a distance between the arbitrary place and the place group.

  The place group determination unit may determine the place group to which the arbitrary place belongs based on the similarity between the attribute of the arbitrary place and the attribute of the place group.

The attribute of the arbitrary place may include types and frequencies of a plurality of points in the arbitrary place, and the attribute of the place group may include types and frequencies of a plurality of points in the place group.
The place group creation unit may create a place group based on the similarity between attributes of a plurality of places.
The attributes of the plurality of places may include types and frequencies of a plurality of points in the place.

  In order to solve the above-described problem, according to another aspect of the present invention, the behavior history data acquisition unit acquires the user's behavior history information, and the place group creation unit detects the user's behavior history information from the behavior history information. A step of creating a place group that is a set of a plurality of places that have been visited, and a step in which a place group determination unit determines to which place group an arbitrary place belongs among the created plurality of place groups And an information providing method for an unvisited location including a step of calculating a probability that a user visits a point in an arbitrary location when the probability calculating unit belongs to the determined location group. Provided.

  The place extracting unit may further include a step of extracting one or a plurality of points in an arbitrary place where the probability that the user visits satisfies a predetermined threshold from a plurality of points in the arbitrary place.

  According to the present invention, even in a place where the user has never visited, a point where the user is highly likely to visit in the future is extracted, and information on the point is provided to the user. Effective information can be provided.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

(Configuration of one embodiment)
First, the configuration of a future behavior prediction system according to an embodiment of the present invention will be described. FIG. 1 is a block diagram showing a future behavior prediction system according to this embodiment. The future behavior prediction system of this embodiment includes server devices 102 and 104 connected to a network 10 and a plurality of user terminals 106.

  First, the server apparatuses 102 and 104 will be described with reference to FIG. FIG. 2 is a block diagram illustrating the server apparatuses 102 and 104 according to the present embodiment. The server apparatuses 102 and 104 include a CPU 112, a memory 114, an input / output I / F 116, a display unit 118, a storage unit 120, a user authentication management unit 122, a network connection unit 124, and the like.

  The server apparatus 102 includes a behavior model generation unit 150, a behavior model holding unit 160, a store recommendation unit 170, a place similarity calculation unit 180, and the like which will be described later. The server apparatus 102 is an example of an information providing apparatus for an unvisited place, obtains a past behavior pattern for each user, generates a behavior model, receives the current situation of the user, Depending on the store recommendation. Here, a store is an example of a point included in a place.

  For example, the server device 104 collects a behavior pattern for each user from a credit card usage history or a blog, and sends the collected behavior pattern to the server device 102. In the present embodiment, the server apparatus 102 and the server apparatus 104 are separate computer apparatuses, but may be a single computer apparatus.

  A CPU (Central Processing Unit) 112 functions as an arithmetic processing device and a control device according to a program, and can control processing of each component provided in the server devices 102 and 104. The memory 114 is configured by a storage device such as a RAM (Random Access Memory), a ROM (Read Only Memory), and a cache memory, for example. The memory 114 has a function of temporarily storing data related to the processing of the CPU 112, an operation program of the CPU 112, and the like.

  The input / output I / F (interface) 116 includes, for example, an operation unit such as a mouse, a keyboard, a touch panel, a button, a switch, and a lever, and an input control unit that generates an input signal and outputs it to the CPU 112. For example, the administrator of the server apparatuses 102 and 104 can input data to the server apparatuses 102 and 104 and instruct processing operations by operating the operation unit of the input / output I / F 116.

  The display unit 118 includes, for example, a liquid crystal display (LCD) device, a CRT display device, a display device that displays a video signal such as a lamp, and an audio output device that outputs an audio signal such as a speaker. For example, the administrator of the server apparatuses 102 and 104 can check the result of the operation performed at the input / output I / F 116 via the display unit 118.

  The storage unit 120 includes, for example, a storage medium such as an HDD (Hard Disk Drive) and a flash memory, and a recording / reproducing device, and stores data over a long period of time. In the case of the server apparatus 102, the storage unit 120 includes, for example, an action model holding unit 160, and records an action probability model and a place group. Moreover, in the case of the server apparatus 104, the memory | storage part 120 records the action history data for every user, for example.

  The network connection unit 124 includes, for example, a communication line, a communication circuit, a communication device, and the like. The network connection unit 124 can transmit and receive data between the server apparatuses 102 and 104 and the user terminal 106 via the network 10.

  The user authentication management unit 122 confirms whether or not the user is a person who can use the service to which the future behavior prediction system according to the present embodiment is applied. The identity verification is determined, for example, based on whether or not the user ID and password transmitted from the user terminal 106 match the user ID and password stored as user information data. Also, the user authentication management unit 122 newly stores user information stored in the user information database and deletes or changes existing user information in response to a request from the user terminal 106 or another terminal, for example.

  The series of processing in the server apparatuses 102 and 104 may be processed by hardware or may be realized by software processing using a program on a computer.

  Next, the user terminal 106 will be described with reference to FIG. FIG. 3 is a block diagram showing the user terminal 106 according to the present embodiment. The user terminal 106 includes, for example, a CPU 132, a memory 134, an input / output I / F 136, a display unit 138, a storage unit 140, a network connection unit 142, and the like.

  The user terminal 106 is, for example, a personal computer, a mobile phone, or a PDA. The user terminal 106 sends a store recommendation request input from the user to the server device 102, and receives a store recommendation as a processing result from the server device 102.

  A CPU (Central Processing Unit) 132 functions as an arithmetic processing device and a control device according to a program, and can control processing of each component provided in the user terminal 106. The memory 134 is configured by a storage device such as a RAM (Random Access Memory), a ROM (Read Only Memory), and a cache memory, for example. The memory 134 has a function of temporarily storing data related to the processing of the CPU 132, an operation program of the CPU 132, and the like.

  The input / output I / F (interface) 136 includes, for example, an operation unit such as a mouse, a keyboard, a touch panel, a button, a switch, and a lever, and an input control unit that generates an input signal and outputs it to the CPU 132. For example, the user of the user terminal 106 can input data or instruct a processing operation to the server apparatuses 102 and 104 by operating the operation unit of the input / output I / F 136.

  The display unit 138 includes, for example, a liquid crystal display (LCD) device, a CRT display device, a display device that displays a video signal such as a lamp, and an audio output device that outputs an audio signal such as a speaker. For example, the user of the user terminal 106 can check the result of the operation performed at the input / output I / F 136 via the display unit 138.

  The storage unit 140 includes, for example, a storage medium such as an HDD (hard disk drive) and a flash memory, and a recording / reproducing device, and stores data over a long period of time.

  The network connection unit 142 includes, for example, a communication line, a communication circuit, a communication device, and the like. The network connection unit 142 can transmit and receive data via the network 10 between the server apparatuses 102 and 104 and the user terminal 106.

  Note that a series of processing in the user terminal 106 may be performed by hardware or may be realized by software processing by a program on a computer.

Next, in particular, a store recommendation system according to an example of the present embodiment will be described.
The configuration of the present embodiment will be described with reference to FIG. FIG. 4 is a functional block diagram when the present embodiment is applied to a store recommendation system as an example of the present embodiment.
The store recommendation system includes a behavior model generation unit 150, a behavior model holding unit 160, a store recommendation unit 170, an inter-location similarity calculation unit 180, and the like.

  The behavior model generation unit 150 and the store recommendation unit 170 are connected to the behavior model holding unit 160 and the inter-location similarity calculation unit 180. The behavior model generation unit 150 generates a behavior model representing a user's behavior pattern from the behavior history provided by the user. The behavior model holding unit 160 holds the behavior model generated by the behavior model generation unit 150. The store recommendation unit 170 inputs a user's current situation (date and time, location, etc.) to the behavior model, and makes a store recommendation according to the user's current situation. The inter-location similarity calculation unit 180 calculates the geographical distance and the similarity in the store configuration between the location appearing in the action history and the current location of the user.

  The behavior model generation unit 150 includes a behavior history data input unit 152, a place group creation unit 154, a behavior probability model generation unit 156, and the like. The action history data input unit 152 is an example of an action history data acquisition unit, and receives action history data such as a credit card use history and a blog from a user. The location group creation unit 154 classifies (groups) the locations that appear in the user's behavior history into a plurality of groups based on the similarity calculated by the inter-location similarity calculation unit 180. The behavior probability model generation unit 156 generates a behavior probability model representing the behavior pattern of the user from the behavior history in which places are grouped.

  The behavior model holding unit 160 includes a behavior probability model holding unit 162, a place group holding unit 164, and the like. The behavior probability model holding unit 162 is a database that holds a behavior probability model representing a behavior pattern generated for each user by the behavior probability model generation unit 156. The place group holding unit 164 is a database that holds the place groups generated for each user by the place group creation unit 154.

  The store recommendation unit 170 includes a current situation input unit 172, a current location group determination unit 174, a user behavior probability calculation unit 176, a store recommendation output unit 178, and the like. The current status input unit 172 receives the current status of the user as input information. The current location group determination unit 174 is an example of a location group determination unit, and calculates the probability that the current location belongs to each location group from the similarity between the current location and the location group in the action history. The user behavior probability calculation unit 176 is an example of the probability calculation unit, and inputs the probability of each place group and the current situation into the user behavior probability model, and the probability that the user will visit each store in the current situation. Is calculated. The store recommendation output unit 178 is an example of a location extraction unit, and recommends stores based on the calculated probability.

  The inter-location similarity calculation unit 180 includes an inter-location similarity calculation unit 182, a store configuration acquisition unit 184, a store configuration holding unit 186, and the like. The inter-location similarity calculation unit 182 calculates the similarity between locations based on the store configuration acquired from the store configuration acquisition unit 184 and the latitude and longitude information. The store configuration acquisition unit 184 acquires information such as how many stores have what characteristics in the vicinity of the target location. The store configuration holding unit 186 is a database that holds the acquired store configuration.

  Among the above, the calculation that generates the behavior probability model of the behavior probability model generation unit 156 and the calculation that uses the behavior probability model of the user behavior probability calculation unit 176 include, for example, a Hugin that realizes a probability model based on a Bayesian network. The company's product Hugin Expert can be used.

[Data structure diagram]
Next, data used in the present embodiment will be described with reference to FIG.
FIG. 5 is a table configuration diagram of the database according to the present embodiment. In this embodiment, there is an action probability model table 162A constituting the action probability model holding unit 162, a place group table 164A constituting the place group holding unit 164, and a store configuration table 186A constituting the store configuration holding unit 186.

  The behavior probability model table 162A includes, for example, a user ID 162A1, a date / time condition 162A2, a place group ID 162A3, a store keyword 162A4, a store business type 162A5, and a visit probability 162A6 as columns. The behavior model of each user is composed of a plurality of rows having the same user ID. Each row represents an individual action pattern taken in the past and a situation where the action was taken.

  The value of the location group ID 162A3 represents what kind of location the action pattern of the row is taken, and a single row of the location group table 164A is referred to by the ID. In the example in FIG. 5, a user with ID “U0001” has a probability of “60%” in a situation where the date is in June and is in “Kyoto” (location group ID = “L0001”). This means that you have visited a “restaurant” where the keyword “local cuisine” appears in the introductory text.

  The location group table 164A has, for example, a location group ID 164A1, a user ID 164A2, a location group center position 164A3, and a location group size 164A4 as columns. Each row represents an individual location group. In the example in FIG. 5, a place group having ID “L0001” has a radius “10 km” centered on (latitude, longitude) = (35.20, 135.37) (latitude and longitude in Kyoto Prefecture). It represents that is represented.

  The store configuration table 186A has, for example, a place group ID 186A1, a store keyword 186A2, a store type 186A3, and a frequency 186A4 as columns. The configuration of stores in each place group is represented by a plurality of rows having the same place group ID. The composition of a store refers to a combination of a keyword and a store industry that appear in a store introduction and the frequency of the combination. In the example of FIG. 5, “restaurant” in which “local cuisine” appears in the introductory text of the store indicates that only “22” stores exist in the location group of “Kyoto” (location group ID = “L0001”). Yes.

(Operation of one embodiment)
Next, after describing the outline of the operation of the example of the present embodiment, the details thereof will be described.

[1. Overview of operation of embodiment]
An example of this embodiment is a store recommendation system. First, a comparative example using a conventional store recommendation system will be described, and then an embodiment of a store recommendation system in an unvisited area where the user has not visited before will be described.

[1-1 Generation and recommendation of behavior model]
An outline when a conventional store recommendation system as a comparative example is implemented will be described.
First, as a preparatory stage for recommendation, behavior history data is received from a user, analyzed and analyzed as a behavior pattern taken by the user according to each situation in the behavior probability model table 162A of FIG. 5 of the present embodiment. The behavior probability model table is stored as follows.

  When a store is requested to recommend a store, an action pattern taken in the past when the situation is closest to the current situation of the user is taken out, and a store that is estimated to be visited from the action pattern is recommended.

  For example, as in the example shown in FIG. 5, when a recommendation is requested in Kyoto in June, an action pattern with a 60% probability of going to a restaurant where local cuisine appears in the introductory text is taken out. Search and recommend restaurants that show local cuisine. The reception of the action history data and the generation / storage of the action model based on the analysis are performed for each user. Here, the behavior model is a set of combinations of past situations and behavior patterns taken by the user in the past.

  As the action history data 208, for example, as shown in FIG. 6, data in which a store use history such as a credit card use history 204 and a blog 202 is recorded can be used, and the date and place can be used as the situation. FIG. 6 is an explanatory diagram showing a method for generating action history data in the present embodiment.

  Information necessary for generating the behavior model and generating the behavior probability model table 162A of FIG. 5 is collected from the credit card usage history 204 received from the user. As shown in FIG. 6, the location information (latitude, longitude) and date / time of the store used by the user can be acquired from the credit card. Further, store information is searched from a restaurant search site 206 such as a web page “GourNavi”, and the store address (latitude and longitude), the store introduction, and the store industry are acquired. These data are aggregated and the visit probability is obtained by aggregating the frequency of keywords appearing in the introductory text for each date and place condition.

[1-2 Outline of recommendation method in unvisited places]
Next, unlike the above-described comparative example of the store recommendation system in the prior art, a method that enables recommendation even in an unvisited place according to this embodiment will be described.

  When a user asks for recommendation, the same place as the current place may or may not exist in the action history data. The recommendation in the case of existing was also possible in the comparative example of the prior art. On the other hand, the recommendation in the case of nonexistence, which has been a problem in the prior art, can be realized by this embodiment.

  If the same location as the current location exists in the behavior history data, the store is recommended based on the behavior pattern taken in the past in a situation that matches the current situation. If it does not exist, according to the present embodiment, the place that is most similar to the current place among the places of the action history data is determined, and store recommendation is performed based on the action pattern taken in the past in the situation at that place. .

  For example, a case where a user who frequently visited Kyoto for sightseeing visited Nara for the first time will be described as an example. In this case, the action pattern in Kyoto is acquired from the past action history, but the action pattern in Nara is not acquired. For this reason, the store recommendation system of the present embodiment first determines that the current location Nara is a place similar to Kyoto, and then makes a store recommendation in Nara based on the past behavior patterns in Kyoto.

  For example, if you frequently go to a temple or local restaurant in Kyoto, if you are in Nara, recommend a temple or local restaurant. The use of local food stores is, for example, by credit card payments, and temple sightseeing, for example, a history of action is accumulated by a tourist report to a blog. Here, the store recommendation system determines that Nara requested by the user as a place similar to Kyoto. In this example, it is assumed that the user visits Nara for the same sightseeing purpose as Kyoto, and it is determined that the store structure around Nara is similar to the store structure of Kyoto in the action history. is there.

[1-3 Place grouping]
When acquiring action patterns, places in the action history are handled in groups. In the action history data, there are many places represented by different latitudes and longitudes, but these are not scattered at random places and are often related to each other. Grouping is performed in advance in order to use this relevance for acquiring behavior patterns.

  Grouping is performed based on both the proximity of geographical distances and the similarity of store configurations. The similarity between two points is defined based on the similarity in geographical distance or the similarity in store structure, and grouping is performed so that places where the similarity is large are included in the same group.

  For example, the places “Kyoto (35.20, 135.37)” and “Kyoto Prefecture XX City (35.13, 135.66)” in FIG. 6 are the same because the values of latitude and longitude are close and geographically close. Classify into Kyoto Prefecture groups. Similarly, the place “Nara Prefecture XX City (34.57, 135.79)” in FIG. 6 is a tourist spot like the above two places and is not necessarily geographically close, but has a similar store structure. Therefore, classify them into groups of the same tourist destination.

  If the action history does not include the same place as the current place in the recommendation, it is based on the action pattern in the past place group by determining which place group in the action history the current place is included in Determine if a recommendation should be made. The determination is made based on which location group the user's current location is similar to. Similarity evaluation is performed by the same method as the similarity evaluation method used when grouping places. Details of the similarity evaluation method are shown in the detailed description of the similarity calculation process below.

[2. Details of Example Operation]
Details of the operation of the store recommendation system will be described below. The operation of the store recommendation system is largely divided into three processes: grouping places and generating behavior models; recommending stores according to the current situation of users; and calculating the similarity between places. Can be separated. The three processes will be described in order.

[2-1 Behavior model generation process]
First, the behavior model generation process will be described. This process is realized by the action history data input unit 152, the place group creation unit 154, and the action probability model generation unit 156 shown in FIG.

  The action history data input unit 152 collects user action history data. The user first provides his / her action history data to the action history data input unit 152. Necessary data is extracted from the credit card usage history and converted into a format that can be used as behavior history data in the behavior model generation process.

  The location group creation unit 154 creates a location group by collecting locations that are close to each other in geographical distance and store configuration into the same group. The procedure is shown in FIG. FIG. 7 is a flowchart showing the processing operation of the place group generation unit 154 according to the present embodiment.

  First, a place is extracted from all the rows of the action history data, and a set of places is created (step S101). For all combinations of two different locations included in this set, the similarity is calculated using the inter-location similarity calculation unit 182, and a similarity matrix between any two locations (for example, location i, location j) is calculated. Create (step S102).

  Locations are clustered based on the similarity matrix (step S103). For clustering, for example, hierarchical clustering by the longest distance method is used. In hierarchical clustering, places with high similarity are repeated in order, and a place hierarchy as shown in FIG. 8 is created. FIG. 8 is an explanatory diagram showing an example of grouping places by hierarchical clustering.

  A fixed threshold is set for the similarity, and the hierarchy is divided into a plurality of partial hierarchies using this threshold (step S104). In FIG. 8, the lower the horizontal line connecting places, the greater the similarity. A broken line indicates a similarity threshold. The division of the hierarchy into partial hierarchies is performed so that places where the similarity is smaller than the threshold are not included in the same partial hierarchy. A partial hierarchy including a larger number of places is extracted from a large number of generated partial hierarchies, and a set of places included in each of the extracted partial hierarchies is defined as a place group (step S105).

  The behavior probability model generation unit 156 uses the behavior history data and the place group created by the place group creation unit 154 to generate a behavior probability model that indicates how the user's behavior pattern changes depending on the date and the place. . The procedure is shown in FIG. FIG. 10 is a flowchart showing the processing operation of the action probability model generation unit 156 according to the present embodiment.

  First, the action probability model generation unit 156 groups the places of the action history data according to the place group (step S201). Next, an action probability model is generated using action history data in which places are grouped (step S202). The action probability model generated last is recorded in the action probability model holding unit 162 (step S203).

  A Bayesian network is used as the behavior probability model representing the behavior model. In the behavior probability model, the situation at that time such as the date and time and place appearing as attributes in the behavior history data, and the store attributes of the stores visited are included as random variables. As store attributes, store business types such as restaurants and department stores, keywords appearing in store introductions, and the like are used. An example of a Bayesian network representing an action model is shown in FIG. FIG. 9 is an explanatory diagram showing an action probability model based on a Bayesian network. A node represents a random variable, and an edge (arc) represents that the variables depend on each other. A place group is used for the value range of the place random variable, and when two different places in the action history belong to the same place group, they are treated as the same value. As a result, geographically different locations are treated as the same location on the model. The behavior probability model table 162A is used for the probability function of each variable.

[2-2 Store recommendation process]
Next, the store recommendation process will be described. The store recommendation process is realized by the current situation input unit 172, the current location group determination unit 174, and the user behavior probability calculation unit 176 shown in FIG.

  The current status input unit 172 receives an input of the current status of the user. When a user requests a store recommendation from a server or the like using a mobile terminal or the like, information on the current position of the user is acquired from the mobile terminal. The date and time is automatically obtained from the current date and time. The current status input unit 172 passes the acquired data representing the current status of the user to the user behavior probability calculation unit 176 and passes the current location to the current location group determination unit 174.

  The current location group determination unit 174 calculates to which location group the current location of the user belongs is high. The procedure is shown in FIG. FIG. 11 is a flowchart showing the processing operation of the place group determination unit 174 according to this embodiment.

  First, the current location group determination unit 174 extracts a set of location groups (step S301). Then, the similarity is calculated between the current location and all location groups in the action history (step S302). For the calculation of the similarity, the inter-location similarity calculation unit 182 is used. As a result, the same number of similarities {si} as the number of place groups are obtained. Next, for each location group i, a distribution of similarity (average similarity and variance) is obtained between all locations j included in the location group i and the location group i (step S303). Thus, the same number of similarity distributions {N (μi, σi)} as the number of place groups are calculated (step S304). N (μ, σ) is a normal distribution with mean μ and variance σ.

  Next, for each location group i, the probability pi = P (Xi ≦ si) that the similarity is equal to or less than si under the distribution N (μi, σi) is obtained (step S305). However, Xi is a random variable according to N (μi, σi). Finally, normalization is performed so that the sum of the probabilities pi becomes 1 (step S306). The current location group determination unit 174 passes the probability to the user behavior probability calculation unit 176.

  The user behavior probability calculation unit 176 receives the probability of the current location (group) and the current date and time, and obtains a behavior pattern with a high probability that the user will take in the current situation. First, the user behavior probability calculation unit 176 extracts a user behavior probability model from the behavior probability model holding unit 162. Next, the current date and time are input to a random variable representing the date and time of this behavior probability model. The probability variable representing the location group is input with the probability that the current location is included in each location group, for example, the probability that the current location is included in the location group 1 is 50% and the probability that the current location is included in the location group 2 is 30%. Is done. The behavior probability model calculates the probability that the user visits a store having the attribute for each store attribute based on the value of the input random variable. Finally, the user behavior probability calculation unit 176 passes the probability value calculated by the behavior probability model to the store recommendation output unit 178.

  The store recommendation output unit 178 recommends a store to the user based on the probability value of the input store information. A store having a store attribute with a high probability value output by the behavior probability model is searched, and the found store is recommended.

[2-3 Location similarity calculation process]
Finally, the process for calculating the similarity between locations will be described. The inter-location similarity calculation process is realized by the inter-location similarity calculation unit 182 and the store configuration acquisition unit 184 shown in FIG.

  The inter-location similarity calculation unit 182 evaluates the similarity between two different locations (location i, location j) from the similarity of the geographical distance and the store configuration. The procedure is shown in FIG. FIG. 12 is a flowchart showing the calculation processing operation of the similarity between the location i and the location j by the location similarity calculation unit 182 according to the present embodiment.

  First, the inter-location similarity calculation unit 182 requests the store configuration acquisition unit 184 to acquire nearby store configurations for the location i and the location j (steps S401 and S402). Then, the acquired store configurations are compared to calculate the similarity of the store configurations (steps S403 and S404). Next, a geographical distance is calculated from the latitude and longitude values of the place i and the place j (step S405). Finally, the calculated similarity of the store configuration and the geographical distance are integrated and output as the similarity between the place i and the place j (step S406). Below, after explaining a store composition, a calculation method of similarity will be explained.

  The store configuration at a certain location refers to the attributes of the stores that exist at that location and the number of stores that have those attributes. The store attribute refers to a keyword that indicates the type of product of the store and what kind of product is handled in the store. For example, if the location is a tourist spot, it has a store structure that has many inns, shops that sell souvenirs, restaurants that handle local dishes, etc. There are many shops and cleaning shops, and there are few inns.

  Information on the types and types of products handled at each store can be obtained from the Internet. For example, a restaurant introduction site “GourNavi” provides restaurant introductions. From here, keywords such as ingredients and menus that characterize the restaurant can be obtained. In addition, a travel introduction site “Rurubu” provides an introduction to accommodation. Such cross-sectional search of store information is provided by the website “Google Map”.

  Next, the definition of the geographical distance between places and the similarity of store configurations will be described. The geographical distance between locations is defined by the Euclidean distance in latitude and longitude. The similarity of the store structure between places is defined as follows. First, for all the attributes of a store, the number of stores having that attribute in the vicinity of the target location is totaled. TF / IDF (Term Frequency / Inverse Document Frequency) is calculated from this frequency, and a vector representing TF / IDF of each attribute of the store at the place is created. This vector is created for both the two locations where the similarity is evaluated. The correlation coefficient between these two vectors is taken as the similarity of the store structure. The similarity between two locations is defined as a value that is the sum of the reciprocal of the geographical distance and the similarity of the store structure.

  Using the store structure calculation method shown above, Tokyo (35.66, 139.71), Kyoto (35.00, 135.71), Nara (34.68, 135.82) FIGS. 13A to 13C show examples of calculating the similarity of the store configuration during that time. FIG. 13 is an explanatory diagram illustrating an example of calculating the similarity of the store configuration in the present embodiment. In this example, the correlation coefficient of TF / IDF is calculated as the similarity only for the three store attributes of the inn, supermarket, and trading company.

  FIG. 13A shows the result of searching for store attributes for three locations using “Google Map”. For example, the number of hits by searching for the vicinity of Kyoto using the store attribute “ryokan” as a search keyword was 556. From this result, in Kyoto and Nara, the number of search hits for “ryokan” is about twice as large as the search hits for “trading company”, whereas in Tokyo the “trading company” is the opposite of “ryokan”. It is about twice as large, and it can be seen that there is a clear difference in store structure depending on the location.

  FIG. 13B shows the TF / IDF for each place / store attribute calculated from FIG. 13A. In FIG. 13C, the similarity of the store configuration between the three places is correlated. Compare by number. FIG. 13C shows, for example, that the correlation coefficient of TF / IDF is −0.29 between Tokyo and Nara. In contrast, the correlation coefficient between Kyoto and Nara is 0.97, which can be confirmed to be significantly higher than the others. From this, it can be seen that the similarity of the store configuration between Kyoto and Nara is significantly higher than the similarity of the other combinations. In this example, the IDF is calculated only from the information of three places, but actually it is necessary to calculate based on the information of more places in Japan.

  The place similarity calculation unit 182 also evaluates the degree of similarity between places and place groups. The procedure is shown in FIG. FIG. 14 is a flowchart showing a procedure for calculating the similarity between the location i and the location group j by the inter-location similarity calculation unit 182 according to the present embodiment.

The similarity between the place i and the place group j is defined as the smallest degree of similarity between the arbitrary places k and the place i included in the place group j.

  First, the similarity between the place i and the place k is calculated about all the places k in the place group j (step S501). The similarity (location i, location k) is calculated by the inter-location similarity calculation unit 182 according to the above-described procedure of FIG. And the minimum similarity is extracted among the similarities between the place k and the place i (step S502). The extracted minimum similarity is determined as the similarity between the place i and the place group j (step S503).

  Next, the processing operation of the store configuration acquisition unit 184 will be described. The store configuration acquisition unit 184 receives the location as input information, and holds and returns the store configuration. The procedure is shown in FIG. FIG. 15 is an explanatory diagram illustrating a processing operation of the store configuration acquisition unit 184 according to the present embodiment.

  When the store configuration acquisition unit 184 is requested to acquire the store configuration, first, it is confirmed whether the store configuration of the place (for example, location 1) has been acquired by the store configuration holding unit 186 and data is held ( Step S601). If acquired, the store configuration is acquired from the store configuration holding unit 186, and the store configuration information is returned to the store configuration acquisition unit 184 (step S607).

  On the other hand, if it has not been acquired, the store information near the location is downloaded and acquired from the Internet (step S602). From the acquired information, the business type and introduction of the store are extracted (step S603). The words constituting the extracted introduction sentence (PR sentence) are extracted by morphological analysis (step S604), and general words appearing in any store are removed from the result to be used as keywords of the store (step S605). . Then, the store business type, the keyword, and the frequency thereof are recorded in the store configuration holding unit 186 as a store configuration (step S606). Finally, the recorded store configuration information is returned.

  Here, the general word removed from the introductory sentence is a word that appears in the introductory sentence of any store such as “store” or “customer”. This calculates TF / IDF and removes words with low values as general words. For obtaining store information from the Internet, portal sites such as “GourNavi” and “Yahoo! Phonebook” can be used. For obtaining latitude and longitude information from an address, a web service such as “Geocoding” of “Google” can be used.

  As described above, according to the example of the present embodiment, the place group creation unit 154 divides places in the action history data into groups, and the action probability model generation part 156 represents action patterns in consideration of the place groups. Generate an action probability model. The current location group determination unit 174 calculates which group of locations where the user's current location has been in the past is high, and the user behavior probability calculation unit 176 calculates the calculated current location. Is input to the behavior probability model to calculate which store is likely to visit. As a result, it is determined which location in the action history data is close to the current location, and it is possible to make a recommendation according to the current situation even in a location that has not been performed in the past.

  Moreover, in this embodiment, the effect which absorbs the diversity of a place flexibly is acquired by performing action prediction by an action probability model. Since the features of places are so diverse, places that have never been in the past are unlikely to have exactly the same features as places in the action history. Therefore, if the current location has never been in the past, there is a high possibility that the current location cannot be uniquely projected onto a location (group) in the action history. A place that has not been visited in the past is likely to be determined as a place having an intermediate characteristic among a plurality of places in the action history.

  In the present embodiment, since the present location is represented by a probability to which group of locations in the past, it is flexible that the features of the current location have intermediate features among the groups of locations. Can be expressed as Since the current location probability is directly input to the behavior probability model and the store recommendation is performed, the diversity of the location can be flexibly reflected in the store recommendation.

  The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention pertains can come up with various changes or modifications within the scope of the technical idea described in the claims. Of course, it is understood that these also belong to the technical scope of the present invention.

It is a block diagram which shows the future action prediction system which concerns on one Embodiment of this invention. It is a block diagram which shows the server apparatus which concerns on the same embodiment. It is a block diagram which shows the user terminal which concerns on the same embodiment. It is a functional block diagram showing a store recommendation system as an example of the embodiment. It is a table structure figure of the database of the embodiment. It is explanatory drawing which shows the production | generation method of action history data in the embodiment. It is a flowchart which shows the processing operation of the place group production | generation part which concerns on the embodiment. It is explanatory drawing which shows an example of the grouping of the place by hierarchical clustering. It is explanatory drawing which shows the action probability model by a Bayesian network. It is a flowchart which shows the processing operation of the action probability model production | generation part which concerns on the embodiment. It is a flowchart which shows the processing operation of the place group determination part which concerns on the same embodiment. It is a flowchart which shows the calculation process operation | movement of the similarity between the place i and the place j by the similarity calculation part between places concerning the embodiment. It is explanatory drawing which shows the example which calculated the similarity of the shop structure in the same embodiment. It is a flowchart which shows the procedure of the similarity calculation between the place i and the place group j by the location similarity calculation part which concerns on the embodiment. It is explanatory drawing which shows the processing operation of the shop structure acquisition part which concerns on the same embodiment.

Explanation of symbols

10 Network 102, 104 Server device 106 User terminal 112, 132 CPU
114, 134 Memory 116, 136 I / O I / F
118, 138 Display unit 120, 140 Storage unit 122 User authentication management unit 124, 142 Network connection unit 150 Behavior model generation unit 152 Behavior history data input unit 154 Place group creation unit 156 Behavior probability model creation unit 160 Behavior model holding unit 162 Behavior Probability model holding unit 164 Location group holding unit 170 Store recommendation unit 172 Current situation input unit 174 Current location group determination unit 176 User behavior probability calculation unit 178 Store recommendation output unit 180, 182 Inter-similarity calculation unit 184 Store configuration acquisition unit 186 store composition holding department

Claims (9)

An action history data acquisition unit for acquiring user action history information;
A place group creation unit for creating a place group that is a set of a plurality of places that the user has visited from the behavior history information;
A place group determination unit that determines which place group an arbitrary place belongs to among the plurality of created place groups;
An information providing apparatus for unvisited places, comprising: a probability calculating unit that calculates a probability that the user visits a point in the arbitrary place when the arbitrary place belongs to the determined place group.   The location extraction part which extracts the 1 or several point in the said arbitrary places where the probability that the said user visits satisfy | fills a predetermined | prescribed threshold value from the several points in the said arbitrary places is provided. Information providing device for unvisited places.   The unvisited location information providing apparatus according to claim 1 or 2, wherein the location group determination unit determines the location group to which the arbitrary location belongs based on a distance between the arbitrary location and the location group. .   The location group determination unit determines the location group to which the arbitrary location belongs based on a similarity between the attribute of the arbitrary location and the attribute of the location group. An information providing device for unvisited places described in Crab.   The attribute of the arbitrary location includes the type and frequency of the plurality of points in the arbitrary location, and the attribute of the location group includes the type and frequency of the plurality of points in the location group. The information providing apparatus for unvisited places according to claim 4.   The unvisited place information providing apparatus according to claim 1, wherein the place group creation unit creates the place group based on a similarity between attributes of the plurality of places.   The unvisited site information providing apparatus according to claim 6, wherein the attributes of the plurality of places include types and frequencies of a plurality of points in the place. A step in which the behavior history data acquisition unit acquires the behavior history information of the user;
A place group creation unit creating a place group that is a set of a plurality of places that the user has visited from the behavior history information;
A place group determining unit determining which place group an arbitrary place belongs to among the created plurality of place groups;
A probability calculation unit that calculates a probability that the user visits a point in the arbitrary place when the arbitrary place belongs to the determined place group, information on unvisited places How to provide. The location extracting unit further includes a step of extracting one or a plurality of points in the arbitrary location where a probability that the user visits satisfies a predetermined threshold from a plurality of points in the arbitrary location. 8. A method for providing information on an unvisited place according to 8.

Images