JP2017091435A - Stay place prediction device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a stay place prediction device capable of improving prediction accuracy of a stay place.SOLUTION: A server 10 comprises: a feature amount storage part 17 for storing a stay ratio which is a ratio in which a user stays for every candidate of stay places, where a user stays; a detection part 18 for detecting the fact that a user starts up a transfer search application; a calculation part 22 for using the stay ratio stored in the feature amount storage part 17, for calculating a stay probability in which the user stays hereafter in the candidate, for every candidate; a prediction part 23 for predicting the stay place where the user stays, based on the stay probability calculated by the calculation part 22; and an output part 24 for outputting the stay place predicted by the prediction part 23. The calculation part 22 performs weighting processing for performing weighting to the stay probability, using a coefficient which becomes a higher value in the candidate which has lower stay ratio, according to the detection result of the detection part 18.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a stay location prediction device that predicts a stay location of a user.

  In an information distribution service that provides content to a user of a mobile terminal, it is possible to provide content more suitable for the user by predicting a stay location where the user stays. Patent Document 1 describes a behavior prediction device that predicts user behavior. This behavior prediction apparatus acquires a history of position data indicating the position of the mobile terminal, and uses the history of position data to estimate a future place of stay of the user.

JP 2011-198292 A

  As described above, in the method using the history of position data, a steady movement to a place where the user often stays is easily predicted, and an unsteady movement to a place where the user does not stay much is difficult to predict. For this reason, the prediction accuracy of the stay place did not improve.

  An object of this invention is to provide the stay location prediction apparatus which can improve the prediction precision of a stay location.

  In order to achieve the above object, the stay location prediction apparatus according to the present invention stores a stay rate, which is a rate at which a user stays, for each stay location candidate that is a place where the user stays, and a user A detection unit that detects that the information on the traffic or the map has been acquired, and a calculation unit that calculates a stay probability, which is a probability that the user will stay in the future for each candidate, using the stay ratio stored in the storage unit; A predicting unit that predicts a staying place where the user stays based on the stay probability calculated by the calculating unit, and an output unit that outputs the staying place predicted by the predicting unit, and the calculating unit includes a detecting unit In accordance with the detection result, a weighting process for weighting the stay probability is performed using a coefficient having a higher value as the stay ratio is lower.

  The stay location prediction apparatus according to the present invention detects that a user has acquired information on traffic or a map. Users tend to acquire information on traffic or maps as they go to places with a low stay rate. For this reason, when calculating the stay probability, which is the probability that the user stays, for each candidate, the stay location prediction device uses a coefficient that has a higher value for a candidate whose stay rate is lower, according to the detection result of the detection unit. A weighting process for calculating the stay probability is performed. As a result, not only steady movement of the user but also unsteady movement can be easily predicted. As a result, the prediction accuracy of the stay location can be improved.

  The calculation unit may perform the weighting process according to the number of times the user has acquired the information regarding the traffic or the map in the preset first period. Users who regularly acquire information on traffic or maps are highly likely to acquire information on traffic or maps even when they go to places with a high stay rate. According to this configuration, since the weighting process is not performed on the user who regularly acquires information on traffic or a map, it is possible to suppress a decrease in the prediction accuracy of the staying place for such a user.

  The storage unit stores the stay rate for each candidate and time zone, and the calculation unit uses the stay rate in the same time zone as the time zone for which the stay probability is calculated by the calculation unit for each candidate. May be calculated. According to this configuration, since the stay location is predicted using the stay rate for each candidate and time zone, the prediction accuracy can be further improved.

  The stay location prediction apparatus according to the present invention further includes a first position information acquisition unit that acquires first position information indicating the first position of the user, and the storage unit stores, for each candidate, the user from the candidate to another candidate. The moving ratio between candidates, which is the ratio of moving to each of the two, is stored, and the calculation unit uses the moving ratio between candidates, which is the ratio of the user moving from the candidate corresponding to the first position to each of the other candidates. The probability may be calculated. According to this configuration, since the stay location is predicted using the inter-candidate movement rate, which is the rate at which the user moves from the stay location candidate to each of the candidates, the prediction accuracy can be further improved.

  A second position information acquisition unit that acquires a plurality of second position information indicating a plurality of second positions of the user, and a plurality of second positions corresponding to the plurality of second position information acquired by the second position information acquisition unit Is included in the predetermined range, an extraction unit that extracts representative positions of the plurality of second positions may be further included, and the candidates may be set based on the representative positions extracted by the extraction unit. According to this configuration, since representative positions are extracted from a plurality of positions, for example, a plurality of positions close to each other can be indicated by one representative position and used as a candidate. Thereby, since a candidate can be set in a desired distance range, the prediction accuracy in the desired distance range can be further improved.

  A predetermined number of presets may be set based on the time when the user stayed at each of the representative positions. According to this configuration, for example, a predetermined number of candidates are set based on the time when the user stayed at the representative position even when the place where the user has stayed frequently is different from the place where the user often stays recently. Thus, candidates can be set in accordance with changes in user behavior patterns. Therefore, the prediction accuracy can be further improved.

  In the present invention, it is possible to improve the prediction accuracy of the staying place.

It is a block diagram of a server concerning an embodiment. It is a figure which shows the hardware constitutions of a server. It is a flowchart which shows the process which extracts a stay point and a candidate. It is a figure for demonstrating the process which extracts a stay point and a candidate. It is a flowchart which shows the process which produces | generates a feature-value. It is a structural diagram of the data memorize | stored in a feature-value memory | storage part. It is a flowchart which shows the process which determines continuity. It is a flowchart which shows the process which estimates a stay place. It is a figure for demonstrating the specific example of each feature-value.

  Hereinafter, embodiments of a stay location prediction apparatus according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a block diagram of a server according to the embodiment. The server 10 (stay location predicting device) is a device for predicting a stay location where the user of the terminal 30 stays. In the information distribution service, it is possible to provide content more suitable for the user's situation by predicting when and where the user is. For example, if the server 10 predicts the destination of the user and can distribute the traffic information corresponding to the destination to the terminal 30 in advance, the user can bypass the train that has stopped.

  The server 10 and the terminal 30 can transmit / receive information to / from each other via a network. This network includes, for example, a mobile communication network or the Internet.

  The terminal 30 is a portable electronic device such as a mobile phone possessed by a user. The terminal 30 may be other than a mobile phone, and may be, for example, a smartphone, a tablet computer, a personal digital assistant (PDA), or the like.

  The terminal 30 has a positioning function, and can measure the position of the terminal 30. This positioning function is realized by a known technique, for example, using a GPS receiver. The terminal 30 performs positioning at a positioning cycle (for example, several minutes) determined in advance by the terminal 30. When the terminal 30 executes positioning, the terminal 30 generates a positioning log that is information of the positioning result. Specifically, the terminal 30 stores a positioning point ID, which is information for identifying the positioning log, information indicating the position of the terminal 30 obtained by this positioning (for example, latitude and longitude), and the terminal 30 in advance. A positioning log including a user ID that is information for identifying the user of the terminal 30 and a positioning time that is a positioning time is generated. The terminal 30 transmits the generated positioning log to the server 10. The terminal 30 may transmit the positioning log to the server 10 every time the positioning log is generated, or may transmit the positioning log to the server 10 collectively after a predetermined period.

  The terminal 30 has a function of acquiring information related to traffic or a map. For example, the terminal 30 performs a transfer search application that is an application (application) for searching for a travel route by train, bus, or the like according to a user input operation, an application for displaying traffic jam information and a map, and a search using the Internet. Get information about traffic or maps. For example, when the terminal 30 activates the transfer search application, the terminal 30 generates an activation log that is information of the activation result. Specifically, the terminal 30 is an activation ID that is information for identifying the activation log, a user ID that is information for identifying a user of the terminal 30 stored in advance in the terminal 30, and an activation time. A startup log including the startup time is generated. Further, the terminal 30 may generate a start log when, for example, an application that displays traffic jam information and a map is started, or a start log when a search using a search term including a place name is performed on the Internet. You may make it produce | generate. The terminal 30 transmits the generated startup log to the server 10. The terminal 30 may transmit the activation log to the server 10 every time the activation log is generated, or may transmit the activation log collectively to the server 10 after a predetermined period has elapsed.

  The server 10 includes a position information acquisition unit (first position information acquisition unit, second position information acquisition unit) 11, a positioning log storage unit 12, an extraction unit 13, a stay log storage unit 14, and a candidate storage unit 15. , Generation unit 16, feature amount storage unit (storage unit) 17, detection unit 18, activation log storage unit 19, determination unit 20, continuity storage unit 21, calculation unit 22, and prediction unit 23 And an output unit 24.

  FIG. 2 is a diagram illustrating a hardware configuration of the server. As shown in FIG. 2, the server 10 includes a CPU 101, a RAM (Random Access Memory) 102 and a ROM (Read Only Memory) 103, which are main storage devices, a communication module 104 for performing communication, and an auxiliary storage device such as a hard disk. It is comprised as what contains a computer provided with the hardwares 105. When these components are operated by a program or the like, the functions of the functional elements of the server 10 shown in FIG. 1 are exhibited. The server 10 may be configured by a computer system including a plurality of computers. Hereinafter, each functional element shown in FIG. 1 will be described.

  The position information acquisition unit 11 receives the positioning log from the terminal 30 as position information (first position information, second position information) indicating the user's position (first position, second position), thereby obtaining the positioning log. This is the part to get. When the position information acquisition unit 11 acquires the positioning log, the position information acquisition unit 11 registers the positioning log in the positioning log storage unit 12.

  The positioning log storage unit 12 is a database that stores positioning logs. The positioning log storage unit 12 stores a positioning point ID, information indicating the position measured by the terminal 30 (for example, latitude and longitude), a user ID, and a positioning time in association with each other.

  Based on a plurality of positioning logs (second position information) corresponding to a predetermined period among a plurality of positioning logs stored in the positioning log storage unit 12, the extraction unit 13 is a candidate for a user's staying place ( Hereinafter, it is a part for extracting (setting) candidates. Specifically, the extraction unit 13 is a part that performs a process of extracting a user's stay point and a process of extracting candidates. The extraction unit 13 extracts stay points and candidates at a predetermined timing (for example, every few days). The extraction unit 13 extracts, as candidates, places where users such as homes, workplaces, and specific customers often stay. The predetermined period is, for example, N days (for example, several days) retroactive from the positioning time of the latest positioning log.

  Here, the stay of the user means that the user stays in a narrow area to some extent. That is, the stay of the user means that the user is continuously located in a predetermined area for a predetermined time or more. The user's staying point is an area where the user is located when the user stays or a position representing the area (representative position). This representative position is, for example, the barycentric position of a plurality of positions where the user was located in the area.

  First, the extraction unit 13 extracts the staying point of the user. Specifically, the extraction unit 13 refers to a plurality of positioning logs (second position information) corresponding to the predetermined period, and the plurality of positions are continuously narrow for a predetermined time (for example, several hours) or more. When included in the region, the stay points based on the plurality of positions are extracted. The extraction unit 13 generates a stay log as a stay point extraction result, and stores the stay log in the stay log storage unit 14. Details of the process in which the extraction unit 13 extracts the stay point will be described later with reference to a flowchart.

  Subsequently, the extraction unit 13 extracts candidates. Specifically, the extraction unit 13 clusters spatially close stay points from a plurality of stay points, and extracts candidates as places where the user stays well within the predetermined period. That is, the extraction unit 13 generates a stay point cluster that is a set of stay points located in a range close to each other, and extracts a representative position based on the stay point included in the stay point cluster as a candidate. The candidate is, for example, the barycentric position of the stay point included in the stay point cluster. The extraction unit 13 stores the extracted candidates in the candidate storage unit 15. Details of the process in which the extraction unit 13 extracts candidates will be described later with reference to a flowchart.

  The stay log storage unit 14 is a database that stores a stay log. The stay log storage unit 14 is generated by the extraction unit 13, for example, a stay point ID that is information for identifying the stay log, a user ID included in the positioning log corresponding to the stay point, and information indicating the position of the stay point (For example, latitude and longitude), information indicating the stay start time (the earliest positioning time among the positioning logs corresponding to the staying point), and the staying end time (the last positioning among the positioning logs corresponding to the staying points) Time) and a positioning point ID corresponding to the stay point are stored in association with each other.

  The candidate storage unit 15 is a database that stores candidates. The candidate storage unit 15 is generated by the extraction unit 13, and is a candidate ID that is information for identifying the candidate, a user ID included in the stay log of the stay point where the candidate is the representative position, and information indicating the position of the candidate (for example, , Latitude / longitude) and the stay point ID of the stay point where the candidate is the representative position.

  The generation unit 16 is a part that generates a feature amount indicating the feature of the user's stay behavior for each candidate stored in the candidate storage unit 15. The feature amount includes, for example, a movement ratio between candidates, a stay ratio by day of the week, a stay ratio by time zone, and a stay ratio. The generation unit 16 generates a feature amount at a predetermined timing. For example, upon receiving a notification from the extraction unit 13 that candidate registration has been completed, the generation unit 16 generates a feature amount. Details of the process in which the generation unit 16 generates the feature amount will be described later with reference to a flowchart.

  The inter-candidate movement ratio is a ratio of the number of times a user moves from a candidate to a candidate. For example, the inter-candidate movement ratio for moving from candidate i to candidate j is obtained by (number of times moved from candidate i to candidate j during a predetermined period) / (number of times stayed at candidate i during a predetermined period). In addition, when running from home and returning to home, the movement between candidates corresponding to the home (for example, movement from candidate i to candidate i) is performed.

  The stay ratio for each day of the week is a ratio indicating the number of stays in a candidate for the number of times or the stay in time for each day of the week. Here, the stay of the user is synonymous with the stay of the user. When the stay ratio by day of the week is, for example, the ratio of the number of times, the stay ratio of staying at the candidate i on the day m is (the number of stays at the candidate i on the day m of the predetermined period) / The sum of the number of stays in each candidate). When the stay ratio by day of the week is, for example, a time ratio, the stay ratio of staying at the candidate i on the day m is (the time at which the candidate i stays on the day m of the predetermined period) / (the day m of the predetermined period The total time spent in each candidate).

  The stay ratio by time zone is a ratio indicating the number of times the user stays at a certain candidate or the stay rate in time by time zone. The time zone is a time zone in one day, for example, a time zone every hour, a time zone in the morning and the afternoon, or the like. That is, the length of the time zone here is shorter than one day. For example, when the stay ratio by time zone is a frequency ratio, the stay ratio of staying in the candidate i in the time slot t is (the number of stays in the candidate i in the time slot t of the predetermined period) / (of the predetermined period). The sum of the number of stays in each candidate during the time period t). For example, when the stay rate by time zone is a time rate, the stay rate of staying at the candidate i in the time zone t is (the time of staying at the candidate i in the time zone t of the predetermined period) / (of the predetermined period) The total time spent at each candidate in the time zone t).

  The staying rate is the number of days that the user stays in a certain candidate (the number of staying days). The stay ratio of the candidate i is obtained by (number of days staying in the candidate i during a predetermined period) / (number of days in the predetermined period). When counting the number of days staying in the candidate i during the predetermined period, for example, the case where the user stays in the candidate i a plurality of times on the same day and the case where the user stays only once are similarly counted as one day.

  As shown in FIG. 6A, the feature amount storage unit 17 stores, for example, the inter-candidate movement ratio for each user ID included in the stay log and each candidate ID of the start point and the end point of inter-candidate movement. To do. Further, as illustrated in FIG. 6B, the feature amount storage unit 17 stores, for example, the stay ratio by day of the week for each user ID, candidate ID, and day of the week (time zone) included in the stay log. Moreover, the feature-value storage part 17 memorize | stores the stay ratio according to time slot | zone for every user ID, candidate ID, and time slot | zone included in a stay log, for example, as FIG.6 (C) shows. Moreover, the feature-value storage part 17 memorize | stores a stay ratio for every user ID and candidate ID contained in a stay log, for example, as FIG.6 (D) shows.

  The detection unit 18 is a part that receives an activation log from the terminal 30 as information indicating that the user has acquired information on traffic or a map, and detects that the user has acquired information on traffic or a map. The detection unit 18 registers the received activation log in the activation log storage unit 19.

  The activation log storage unit 19 is a database that stores activation logs. The activation log storage unit 19 stores an activation ID, a user ID, and an activation time in association with each other.

  The determination unit 20 is a part that determines whether or not the user of the terminal 30 regularly acquires information on traffic or a map (determines continuity). Specifically, the determination unit 20 refers to a plurality of activation logs stored in the activation log storage unit 19, and the user of the terminal 30 activates the transfer search application during a preset period (first period). The continuity is determined according to this number. The determination unit 20 stores the determination result in the continuity storage unit 21. The determination unit 20 determines continuity at a predetermined timing. For example, when the determination unit 20 receives a notification from the generation unit 16 that the process of generating the feature amount has been completed, the determination unit 20 determines the continuity. Details of the process in which the determination unit 20 determines continuity will be described later with reference to a flowchart.

  The continuity storage unit 21 is a database that stores continuity. The continuity storage unit 21 stores a user ID and a determination result in association with each other. Specifically, when the determination unit 20 determines that the user does not regularly use the transfer search application (no continuity), the continuity storage unit 21 stores the continuity flag as “0”. To do. When the determination unit 20 determines that the user is steadily using the transfer search application (with continuity), the continuity storage unit 21 stores the continuity flag as “1”.

  The calculation part 22 is a part which calculates the stay probability which is a probability that a user will stay at a candidate in the future for every candidate. Specifically, the calculation unit 22 calculates the movement ratio between candidates stored in the feature amount storage unit 17, the stay ratio by day of the week, the stay ratio by time zone, and the stay ratio, and the positioning acquired by the position information acquisition unit 11. Based on the positioning log (first position information) with the latest positioning time in the log and the value of the continuity flag stored in the continuity storage unit 21, the user's predetermined time (for example, It is a part for calculating the stay probability after 1 hour) for each candidate. According to the detection result of the detection unit 18, the calculation unit 22 uses a coefficient (for example, a reciprocal of the stay rate for each candidate, and (1−stay rate) for a candidate with a lower stay rate, for a stay probability for each candidate. ) Is multiplied by the exponent, etc.) to perform weighting. For example, the calculation unit 22 calculates the stay probability for all users every hour. The calculation unit 22 outputs the stay probability calculation result to the prediction unit 23. The calculation unit 22 performs a process of calculating a stay probability at a predetermined timing. Details of the process in which the calculation unit 22 calculates the stay probability will be described later with reference to a flowchart.

  The prediction unit 23 is a part that inputs the stay probability calculated by the calculation unit 22 and predicts the user's future stay location based on the stay probability. The prediction unit 23 outputs the prediction result to the output unit 24. Details of the process in which the prediction unit 23 predicts the future place of stay of the user will be described later with reference to a flowchart.

  The output unit 24 is a part that receives a prediction result from the prediction unit 23 and outputs the prediction result. In other words, the output unit 24 is a part that outputs the stay location predicted by the prediction unit 23. Specifically, the output unit 24 outputs the prediction result by registering the predicted stay location in a predetermined storage unit (not shown) in association with the user ID. As described above, since the output unit 24 registers information in which the user ID is associated with the predicted stay location, the server 10 provides the terminal 30 with content more suitable for the user's situation using the information. can do.

  With reference to FIG.3 and FIG.4, the process which the extraction part 13 extracts a stay point and a candidate is demonstrated. FIG. 3 is a flowchart showing a process for extracting stay points and candidates. FIG. 4 is a diagram for explaining the process of extracting stay points and candidates. As shown in FIG. 3, first, the extraction unit 13 acquires a positioning log for each user and date from the positioning log storage unit 12 at the above-described predetermined timing (step S11). Then, the extraction part 13 extracts a stay point based on the acquired positioning log (step S12).

  Specifically, the extraction unit 13 determines the stay point in the order of the positioning time from the oldest for the acquired positioning log. First, the extraction unit 13 sets the position indicated in the positioning log with the oldest positioning time as the reference position for determining the staying point. The extraction unit 13 compares the position where the positioning time is indicated in the next positioning log with the reference position. The extraction unit 13 determines whether or not the distance between these positions is smaller than a preset distance threshold for determining a staying point. If the extraction unit 13 determines that the distance is smaller than the distance threshold, the extraction unit 13 further compares the position indicated in the next positioning log with the reference position, and performs a determination based on the same distance threshold as described above. The extraction unit 13 sequentially performs determination until the distance becomes smaller than the distance threshold (the distance becomes equal to or greater than the distance threshold).

  The extraction unit 13 compares the positioning time related to the previous positioning log whose distance becomes smaller than the distance threshold with the positioning time related to the positioning log at the reference position. The extraction unit 13 determines whether or not the time difference between the positioning times is larger than a preset time threshold for determining the staying point. If the extraction unit 13 determines that the time difference is larger than the time threshold value, the extraction unit 13 retains those positioning logs (positioning logs from the positioning log at the reference position to the previous positioning log whose distance is larger than the distance threshold value). It is determined that it is a positioning log (positioning log corresponding to a staying point) constituting a point. That is, in this case, the extraction unit 13 determines that the stay point can be extracted. If the extraction unit 13 determines that the time difference is not larger than the time threshold value, those extraction logs (positioning logs from the positioning log at the reference position to the previous positioning log whose distance is larger than the distance threshold value) It is determined that it is not a positioning log that constitutes a stay point. That is, in this case, the extraction unit 13 determines that the stay point cannot be extracted.

  In addition, the extraction unit 13 performs the same determination as described above, using the position indicated in the positioning log whose distance is no smaller than the distance threshold as a reference position for determining a new stay point. Thus, if the user is located in the region from the reference position to the distance threshold for a time exceeding the time threshold, it is determined that there is a staying point. The distance threshold and the time threshold are appropriately set according to the purpose of the stay point extraction. The region for determining the stay point may be a circular region based on the reference position as described above, or a rectangular (mesh) region based on the reference position.

  The extraction unit 13 calculates the position (latitude and longitude) of the stay point from the positioning log that is supposed to form the stay point. The position of the stay point can be, for example, the center of gravity (average position) of the position indicated by the positioning log that constitutes the stay point. Note that the position of the stay point may be calculated by a method other than this method. The extraction unit 13 acquires a positioning log from the positioning log storage unit 12 for each day, and calculates the stay point position for each day.

  FIG. 4A shows an example in which two stay points are extracted (calculated). The extraction unit 13 extracts the staying point P1 and the staying point P2, and does not extract the positioning points p1 to p9 whose positioning points (latitude and longitude of the positioning log) are not dense as surrounding points (extracting the staying points). Judge that it is not possible). Further, the extraction unit 13 extracts the stay points for N days. When the extraction unit 13 extracts a stay point (determines that the stay point is a positioning log), the extraction unit 13 generates a stay log as the stay point extraction result. Specifically, the extraction unit 13 includes the user ID included in the positioning log corresponding to the stay point, the latitude and longitude of the position indicating the stay point, and the ID (positioning point ID) of the positioning log corresponding to the stay point. The earliest positioning time of the positioning log and the last positioning time of the positioning log are specified, and a residence log including these is generated.

  The extraction unit 13 registers the stay log in the stay log storage unit 14 when the stay point is extracted as described above based on the positioning logs for N days corresponding to the predetermined period.

  Note that the extraction unit 13 may extract the stay point by another method. For example, based on the position and time of the positioning log, well-known technologies (eg, Kyosuke Nishida, Hiroyuki Toda, Ken Kurashima, Kei Uchiyama, stochastic visit POI analysis: user modeling from spatiotemporal behavior trajectory, “multimedia, distributed, The dwell points may be extracted using a spatio-temporal mean shift clustering algorithm, which is a collaboration and mobile (DICOMO2013) symposium, 2C-6, pp. 334-345, 2013.

  Subsequently, the extraction unit 13 extracts candidates (step S13). Specifically, as described above, the extraction unit 13 acquires a positioning log for each day, extracts a staying point from the positioning log, extracts staying points for N days, and then stays based on the staying point. Generate a point cluster. This stay point cluster is a set of stay points located in a range close to each other. For example, as shown in FIG. 4B, the extraction unit 13 generates a stay point cluster by means of a known technology shift when there are stay points P1 to P9. For example, the extraction unit 13 generates the stay point cluster for each user ID.

  The extraction unit 13 generates a stay point cluster based on a preset kernel shift kernel width (width based on distance). For example, the extraction unit 13 clusters all the stay points for N days based on the latitude and longitude of the stay points. For example, as shown in FIG. 4B, when there are stay points P1 to P9, the stay points P1, the stay points P4, and the stay points P5 are dense (the stay points P4 within the kernel width from the stay points P1). Therefore, the extraction unit 13 extracts the stay point cluster C1. Similarly, the extraction unit 13 extracts each stay point cluster (stay point clusters C2 to C5).

  In this way, the extraction unit 13 clusters the stay points that are close to each other and extracts the stay point cluster (the place where the user stayed well in N days). The extraction unit 13 calculates the representative position (latitude and longitude) of the stay point cluster as a candidate from the stay points that are supposed to form the stay point cluster. The candidate position can be, for example, the barycentric position (average position) of the stay point positions that are supposed to form the stay point cluster. The representative position of the stay point cluster may be calculated by a method other than this method.

  When extracting the candidate, the extraction unit 13 stores information based on the result of extracting the candidate in the candidate storage unit 15 (step S14). When the extraction unit 13 finishes registering these pieces of information for all users in the candidate storage unit 15, the extraction unit 13 notifies the generation unit 16 that registration of candidates has been completed.

  With reference to FIGS. 5 and 6, a process in which the generation unit 16 generates a feature amount will be described. FIG. 5 is a flowchart illustrating processing for generating a feature amount. FIG. 6 is a data structure diagram showing feature amounts. As illustrated in FIG. 6, when the generation unit 16 receives notification from the extraction unit 13 that registration of candidates has been completed, the generation unit 16 stores a candidate ID and a staying point ID corresponding to the candidate ID from the candidate storage unit 15 for each user. Are acquired (step S21). Subsequently, the generation unit 16 acquires a stay log for each user from the stay log storage unit 14 (step S22).

  Subsequently, the generation unit 16 calculates an inter-candidate movement ratio (step S23). Specifically, the generation unit 16 specifies a candidate ID corresponding to the stay log having the oldest stay start time or stay end time. Subsequently, the generation unit 16 identifies a candidate ID corresponding to the second oldest staying log, starts with the candidate ID corresponding to the first oldest staying log, and sets the candidate ID corresponding to the second oldest staying log. Detect movement between candidates as the end point. Subsequently, the generation unit 16 identifies a candidate ID corresponding to the third oldest residence log, starts with a candidate ID corresponding to the second oldest residence log, and sets a candidate ID corresponding to the third oldest residence log. Detect movement between candidates as the end point. The generation unit 16 sequentially identifies candidate IDs corresponding to the fourth and subsequent old stay logs, and repeats the process of detecting inter-candidate movement. The generation unit 16 counts the number of inter-candidate movements detected in this way for each combination of the starting point candidate ID and the end point candidate ID, and calculates the inter-candidate movement ratio. The generation unit 16 calculates the inter-candidate movement ratio for all users.

  Subsequently, the generation unit 16 calculates a stay ratio by day of the week (step S25). Specifically, the generation unit 16 uses the result of specifying the candidate ID corresponding to each stay log when calculating the inter-candidate movement ratio, and stays for each day of the week indicated by the stay start time or stay end time. The number of times identified as corresponding to the log is counted for each candidate ID as the number of stays by day of the week. When the generation unit 16 counts the number of stays by day of the week for all candidate IDs, the generation unit 16 calculates the stay ratio by day of the week for each candidate ID. The generation unit 16 calculates the stay ratio by day of the week for all users.

  Then, the production | generation part 16 calculates the stay ratio classified by time slot | zone (step S26). Specifically, when calculating the inter-candidate movement ratio, the generation unit 16 uses the result of specifying the corresponding candidate ID for each stay log, and for each time period indicated by the stay start time or the stay end time. The number of times specified to correspond to the stay log is counted for each candidate ID as the number of stays by time zone. When the generation unit 16 counts the number of stays by time zone of all candidate IDs, the generation unit 16 calculates the stay rate by time zone for each candidate ID. The production | generation part 16 calculates the stay ratio classified by time zone for all users.

  Subsequently, the generation unit 16 calculates a stay ratio (step S24). Specifically, when calculating the inter-candidate movement ratio, the generation unit 16 uses the result of specifying the candidate ID corresponding to each stay log, and sets the number of days specified as corresponding to the stay log as the number of stay days. Count for each candidate ID. When the generation unit 16 counts the stay days of all candidate IDs, the generation unit 16 calculates a stay ratio for each candidate ID. The generation unit 16 calculates the stay ratio for all users.

  The generation unit 16 registers the movement ratio between candidates, the stay ratio by day of the week, the stay ratio by time zone, and the stay ratio, which are calculated and generated as described above, as feature amounts in the feature amount storage unit 17 (step S27). After completing the registration of the feature amount for all users, the generation unit 16 notifies the determination unit 20 that the registration of the feature amount is completed.

  With reference to FIG. 7, the process in which the determination part 20 determines continuity is demonstrated. FIG. 7 is a flowchart showing processing for determining continuity. As illustrated in FIG. 7, when the determination unit 20 receives a notification from the generation unit 16 that the registration of the feature amount has been completed, for each user from the activation log storage unit 19 for a predetermined period (for example, several days). Is obtained (step S31). Subsequently, the determination unit 20 refers to the acquired activation log, counts the number of activations of the transfer search application, and calculates the average number of activations per predetermined unit period (for example, one day) (step S32).

  Next, the determination unit 20 determines whether or not the average number of activation times of the transfer search application in a predetermined period is equal to or greater than a predetermined number (M times) set in advance (step S33). When the average activation count of the transfer search application in the predetermined period is less than the preset predetermined number (in the case of No in step S33), the determination unit 20 determines that there is no continuity (step S34), and the determination result is It memorize | stores in the continuity memory | storage part 21, and preserve | saves (step S35). Specifically, the continuity storage unit 21 stores a continuity flag corresponding to the user ID as “0”.

  When the average number of activations of the transfer search application in the predetermined period is equal to or greater than the predetermined number (if Yes in step S33), the determination unit 20 determines that there is continuity (step S36), and the determination result is a continuity storage unit. It memorize | stores in 21 and preserve | saves (step S37). Specifically, the continuity storage unit 21 stores a continuity flag corresponding to the user ID as “1”.

  Subsequent to step S35 or step S37, the determination unit 20 determines whether or not the continuity for all users has been determined (step S38). If determination of continuity for all users has not been completed (No in step S38), the determination unit 20 repeats the processing from step S31 to S37. When the determination of continuity for all users has been completed (Yes in step S38), the determination unit 20 ends the process.

  With reference to FIG.8 and FIG.9, the process which estimates a stay place is demonstrated. FIG. 8 is a flowchart showing a process for predicting a staying place. FIG. 9 is a diagram for explaining a specific example of each feature amount. As illustrated in FIG. 8, the calculation unit 22 acquires a positioning log, a candidate ID, and each feature amount for each user ID, for example, every hour (step S41). Specifically, the calculation unit 22 acquires a positioning log with the latest positioning time as position information (first position information) indicating the position (first position) of the user corresponding to the user ID from the positioning log storage unit 12. Then, the candidate ID is acquired from the candidate storage unit 15, and the movement ratio between candidates, the stay ratio by day of the week, the stay ratio by time zone, and the stay ratio are acquired from the feature amount storage unit 17.

  FIG. 9A is a specific example of the inter-candidate movement ratio. FIG. 9B is a specific example of the stay ratio by day of the week. FIG. 9C is a specific example of the stay ratio by time zone. FIG. 9D is a specific example of the stay ratio. Below, it demonstrates based on these specific examples. These specific examples are different from specific examples of candidate extraction by the extraction unit 13.

  Subsequently, the calculation unit 22 calculates the stay probability of the user (step S42). Specifically, first, the calculation unit 22 performs a process of acquiring a candidate ID corresponding to the position measured by the terminal 30 indicated in the positioning log with the latest positioning time as the candidate ID serving as the starting point of the inter-candidate movement. . In other words, the calculation unit 22 identifies a candidate ID having the closest representative position to the position measured by the terminal 30 whose positioning time is indicated in the latest positioning log. The calculation unit 22 acquires the candidate ID specified in this way as a candidate ID that is a starting point for inter-candidate movement. Here, for example, it is assumed that “SP1” is acquired as the candidate ID of the starting point of the inter-candidate movement.

  The calculation unit 22 sets a day of the week and a time zone for which the calculation unit 22 calculates a stay probability. For example, the calculation unit 22 refers to a clock, acquires the current date and time, and sets a date and time after a predetermined time (for example, one hour) after the acquired date and time as a calculation day of the week and time zone. Here, for example, “month” is set as the day of the week, and “9 o'clock” is set as the time zone, for example.

  Subsequently, the calculation unit 22 calculates the stay probability of each candidate by multiplying the inter-candidate movement ratio, the stay ratio by day of the week, and the stay ratio by time zone. Here, the calculation unit 22 calculates the stay probability of each candidate under the condition that the candidate ID of the starting point of the inter-candidate movement is “SP1”, the day of the week is “month”, and the time zone is “9 o'clock”. That is, for each candidate, the calculation unit 22 moves between candidates at the same start point as the start point for which the calculation unit 22 calculates the stay probability, and the same day of the week as the target day for which the calculation unit 22 calculates the stay probability. The stay probability is calculated using the stay ratio by day of the week and the stay ratio by time zone in the same time zone as the time zone for which the calculation unit 22 calculates the stay probability.

  Specifically, the stay probability of the candidate ID “SP1” is the inter-candidate movement ratio “0.08” when the candidate ID of the start point is “SP1” and the candidate ID of the end point is “SP1”, and the day of the week is “ Monthly stay ratio by day of week when candidate ID is “SP1” “0.7”, and stay ratio by time zone when time zone is “9:00” and candidate ID is “SP1” “0. By multiplying by “2”, it is calculated as “0.00112”. The stay probability of the candidate ID “SP2” is the inter-candidate movement ratio “0.6” when the starting point candidate ID is “SP1” and the end point candidate ID is “SP2”, the day of the week is “month”, and the candidate ID. Multiply the stay ratio by day of the week when “S2” is “SP2” and the stay ratio by time zone “0.5” when the time zone is “9:00” and the candidate ID is “SP2”. By combining these values, “0.06” is calculated.

  The stay probability of the candidate ID “SP3” is the inter-candidate movement ratio “0.3” when the start point candidate ID is “SP1” and the end point candidate ID is “SP3”, and the day of the week is “month” and the candidate ID. Is multiplied by the stay ratio by day of the week when “SP3” is “SP3” and the stay ratio by time of “0.3” when the time zone is “9:00” and the candidate ID is “SP3” By combining, it is calculated as “0.009”. The stay probability of the candidate ID “SP4” is the inter-candidate movement ratio “0.02” when the candidate ID of the start point is “SP1” and the candidate ID of the end point is “SP4”, the day of the week is “Month”, and the candidate ID By multiplying the stay ratio by day of the week when “SP4” is “SP4” and the stay ratio by time zone “0” when the time zone is “9:00” and the candidate ID is “SP4”, Calculated as “0”.

  In this way, the calculation unit 22 calculates the stay probability of each candidate under the condition that the candidate ID of the starting point of the inter-candidate movement is “SP1”, the day of the week is “month”, and the time zone is “9 o'clock”.

  Subsequently, the calculation unit 22 determines whether or not the user regularly uses the transfer search application (step S43). Specifically, the calculation unit 22 acquires the value of the continuity flag stored in the continuity storage unit 21 in association with the user ID of the user. When the value of the continuity flag is “0”, that is, when it is determined that the user does not regularly use the transfer search application (No in step S43), the calculating unit 22 It is determined whether or not the user has activated the transfer search application for a predetermined period (step S44). The predetermined period is, for example, a period of several hours retroactively from the current date and time. Therefore, the calculation unit 22 specifically acquires the latest activation time from the activation log of the user in the activation log storage unit 19, and the latest activation time is within one hour retroactively from the current date and time. It is determined whether or not the time is included. When it is determined that the transfer search application is activated during the predetermined period (Yes in step S44), the calculation unit 22 performs weighting processing for weighting the stay probability (step S45).

  It is known that the user tends to acquire information on traffic or a map when the user goes to a place where the stay rate is low, that is, when the user moves unsteadily (for example, Tomomi Ishimura, Ota). Tsunehira, Norio Tomii, “Detection of sudden movement demand in the near future based on the result data of the route search service,” Proceedings of the 47th Civil Engineering Research Conference, June 2013).

  Therefore, as a weighting process, the calculation unit 22 uses the reciprocal of the stay rate for each candidate to the nth power (here, n = 6) as a coefficient having a higher value for a candidate with a lower stay rate. Multiply processing. However, when the user regularly uses the transfer search application, it is considered that the transfer search application is used not only in the case of non-stationary movement but also in the case of steady movement. For this reason, the calculation unit 22 performs weighting processing for each case depending on whether or not the user regularly uses the transfer search application.

For example, the stay probability of the candidate ID “SP1” was calculated as “0.00112 (= 0.08 × 0.7 × 0.2)”. However, when the weighting process is performed, the candidate ID is “SP1”. Is multiplied by the sixth power of the reciprocal of the stay ratio “1”, and “0.00112 (= 0.08 × 0.7 × 0.2 × (1/1) ⁶ )” is obtained. The stay probability of the candidate ID “SP2” was calculated as “0.06 (= 0.6 × 0.2 × 0.5)”. However, when the weighting process is performed, the candidate ID is “SP2”. In this case, the reciprocal of the stay ratio “0.7” is multiplied by the sixth power to obtain “0.509 (= 0.6 × 0.2 × 0.5 × (1 / 0.7) ⁶ ”. When the stay probability of the ID “SP3” is calculated as “0.009 (= 0.3 × 0.1 × 0.3)”, when the weighting process is performed, the candidate ID is “SP3”. Is multiplied by the sixth power of the reciprocal of the stay ratio of “0.5” to obtain “0.576 (= 0.3 × 0.1 × 0.3 × (1 / 0.5) ⁶ ”. The stay probability of “SP4” is “0”, and does not change depending on this weighting process, so when the weighting process is not performed, the stay probability is the candidate ID “SP ”And the candidate ID“ SP1 ”, candidate ID“ SP3 ”, and candidate ID“ SP4 ”become lower in this order. When weighting is performed, the stay probability is highest for the candidate ID“ SP3 ”, and the candidate ID“ SP3 ”. “SP2”, candidate ID “SP1”, and candidate ID “SP4” decrease in this order.

  Subsequently, the calculation unit 22 outputs the calculated stay probability in association with the user ID to the prediction unit 23 (step S46). When the value of the continuity flag is not “0”, that is, when it is determined that the user regularly uses the transfer search application (No in step S43), and the transfer search application is within a predetermined period. Is determined not to be activated (No in step S44), the calculation unit 22 performs the process of step S46 without performing the weighting process.

  Here, the continuity flag is set based on the number of times the transfer search application is activated in the predetermined period counted by the determination unit 20. The determination unit 20 counts the number of activations from the activation log that is the detection result of the detection unit 18. Therefore, it can be said that the calculation unit 22 performs weighting processing according to the detection result of the detection unit 18. More specifically, the calculation unit 22 calculates the stay probability using a coefficient whose value is higher for a candidate whose stay rate is lower, according to the number of times the user has acquired information on traffic or a map during a predetermined period. It can be said that weighting processing for weighting is performed.

  When the prediction unit 23 receives the calculation result of the stay probability from the calculation unit 22, the prediction unit 23 predicts, as the stay location, a candidate having the highest stay probability among the stay probabilities input from the calculation unit 22 for each user ID (step S47). . Subsequently, the candidate ID is output to the output unit 24 as a predicted place of stay. When the output unit 24 inputs the prediction result from the prediction unit 23, the output unit 24 associates it with the user ID and outputs the stay place prediction result by registering it in a predetermined storage unit (not shown) (step S 48). Subsequently, the calculation unit 22 determines whether or not the stay probabilities for all users have been calculated (step S49).

  When the stay probabilities for all users have not been calculated (No in step S49), the calculation unit 22 repeats the processes from step S41 to S48 for another user whose stay probability has not been calculated. When calculation of the stay probabilities for all users has been completed (Yes in step S49), the process of predicting the stay location is terminated.

  As described above, in the server 10, the detection unit 18 detects that the user has started the transfer search application. The user tends to start the transfer search application as the user goes to a place where the stay rate is low. For this reason, in the server 10, when the calculation part 22 calculates a stay probability for every candidate, according to the detection result of the detection part 18, using a coefficient with a higher value for a candidate whose stay ratio is lower, the stay probability is calculated. A weighting process for calculating is performed. As a result, not only steady movement of the user but also unsteady movement can be easily predicted. As a result, the prediction accuracy of the stay location can be improved.

  In addition, a user who regularly activates the transfer search application is likely to start the transfer search application regularly even when going to a place where the stay rate is high. Therefore, in the server 10, the determination unit 20 determines whether or not the user regularly uses the transfer search application. The calculation unit 22 does not perform weighting processing on the user who is determined to have continuity by the determination unit 20. Thereby, it can suppress that the prediction precision of the stay place with respect to the user judged that there exists continuity falls.

  In the server 10, the feature amount storage unit 17 stores a movement ratio between candidates, a stay ratio by day of a week, a stay ratio by time zone, and a stay ratio as feature amounts indicating the features of the user's stay behavior for each candidate. . The calculation unit 22 calculates the stay probability using each of the feature amounts stored in the feature amount storage unit 17 and the prediction unit 23 predicts the stay location based on the stay probability, so that the prediction accuracy is further improved. It becomes possible.

  In the server 10, the extraction unit 13 refers to a plurality of positioning logs stored by the positioning log storage unit 12 and extracts candidates based on the positions indicated by the plurality of positioning logs. Since candidates are extracted from a plurality of positions, for example, a plurality of positions close to each other can be indicated by one representative position and used as a candidate. Thereby, a candidate can be set in a desired distance range. That is, if desired, candidates can be set in areas with a wide range such as the vicinity of the home and the vicinity of the workplace. As a result, the prediction accuracy in such a desired distance range can be further improved.

  In the invention described in Patent Document 1, the prediction accuracy of the stay location is improved by using the scheduler data registered by the user in the scheduler of the terminal. However, the scheduler data is not always available, and in the present invention, when the scheduler data is not available, the prediction accuracy of the staying place cannot be improved. On the other hand, in the server 10, even when the scheduler data cannot be used, the prediction accuracy of the staying place can be improved.

  In the server 10, the extracting unit 13 extracts candidates by referring to N days from the latest positioning time among the plurality of positioning logs stored by the positioning log storage unit 12. For example, even if the location where the user has often stayed is different from the location where the user often stays recently, by appropriately setting the period for referring to the positioning log, it can be adapted to the change in the user's behavior pattern Candidates can be set. Therefore, the prediction accuracy can be further improved.

  When registration of the activation log is completed, the detection unit 18 notifies the calculation unit 22 of the fact, and when the calculation unit 22 receives the notification, the detection unit 18 stays in the user corresponding to the user ID included in the activation log. It is good also as a structure which calculates a probability. In this case, the process of predicting the stay location differs from the process of the flowchart shown in FIG. 8 in the following points. That is, before the process of step S41, a process corresponding to step S44 (a process of determining whether or not the transfer search application is activated) is performed. In this case, the calculation unit 22 is performed as a process for determining whether or not there is a notification from the detection unit 18. This determination by the calculation unit 22 is repeatedly performed until a notification from the detection unit 18 is received.

  When receiving the notification from the detection unit 18, the calculation unit 22 refers to the activation log storage unit 19, acquires the user ID included in the latest activation log, and performs the above step S44 and the above for the acquired user ID. The process of step S41 and subsequent steps is performed without the process of step S49. That is, if Yes in step S43, the process of step S45 is performed. Moreover, the process which estimates a stay place is complete | finished after step S48. Thereby, the stay place of the user who started the transfer search application is predicted.

  The extraction unit 13 may set the number of candidates to a predetermined number (for example, L locations). For example, a predetermined number of candidates (for example, L locations) determined from the extracted candidates may be used as candidates for predicting the user's staying place. Specifically, this further extraction may be performed based on the stay start time or stay end time of the stay point corresponding to the candidate. More specifically, this further extraction may be performed, for example, in the order of the representative position where the residence start time or residence end time of the corresponding residence point is new. In this case, even if the extraction unit 13 does not predetermine a period for referring to the positioning log and refers to all the positioning logs, the representative positions of the L most recent locations are extracted in order of newest time the user stayed. Candidates can be set according to changes in behavior patterns. It is good also as a structure which extracts the representative position of the nearest L place in order with the time when the user stayed further in advance the period when the extraction part 13 refers a positioning log. In this case, it becomes easier to set candidates according to changes in the user's behavior pattern.

  Further, the stay probability before weighting may be obtained by machine learning using a support vector machine (SVM) or the like. That is, the generation unit 16 generates a movement rate between candidates, a stay rate by day of the week, and a stay rate by time zone, and the calculation unit 22 performs a movement rate between candidates, a stay rate by day of the week, and a stay by time zone. Instead of the process of calculating the stay probability before weighting using the ratio, the number of times of moving from candidate i to candidate j in a predetermined period, the number of times of staying in candidate i on day m of the predetermined period, and the time period t of the predetermined period Alternatively, the number of times that the user has stayed at the candidate i may be used as a feature amount to perform a process of obtaining the stay probability before weighting by machine learning.

  The inter-candidate movement ratio may be, for example, a ratio of the number of times the user moves from a candidate to a candidate through K (for example, one) candidates. According to this configuration, when the user moves from the home to the company via the station, the starting point of the inter-candidate movement can be the home and the end point is the company, and the inter-candidate movement that is more realistic is specified. It becomes easy.

  Although the stay ratio is a day ratio, it may be a number or time ratio.

  The calculation unit 22 may perform weighting processing regardless of the presence or absence of continuity. In this case, the determination unit 20 can be omitted and the processing load can be reduced. Moreover, the feature-value memory | storage part 17 should just memorize | store at least a stay rate as a feature-value, and does not need to memorize | store the movement rate between candidates, the stay rate according to a day of the week, and the stay rate according to time zone. That is, the server 10 may predict the stay location without using the inter-candidate movement ratio, the stay ratio by day of the week, and the stay ratio by time zone. When the server 10 does not use the inter-candidate movement ratio, the calculation unit 22 acquires a positioning log with the latest positioning time as position information (first position information) indicating the position (first position) of the user corresponding to the user ID. You don't have to. A predetermined place may be set as a candidate. In this case, the extraction unit 13 can be omitted and the processing load can be reduced. The extraction unit 13 may be configured to extract candidates from the positioning log, and may not necessarily be configured to extract candidates from the positioning log corresponding to a predetermined period.

DESCRIPTION OF SYMBOLS 10 ... Server (stay place prediction apparatus), 11 ... Position information acquisition part (1st position information acquisition part, 2nd position information acquisition part), 13 ... Extraction part, 17 ... Feature-value memory | storage part (storage part), 18 ... Detection unit, 22 ... calculation unit, 23 ... prediction unit, 24 ... output unit.

Claims (6)

A storage unit that stores a stay rate, which is a rate at which the user stays, for each stay place candidate where the user stays;
A detection unit for detecting that the user has acquired information on traffic or a map;
Using the stay ratio stored in the storage unit, a calculation unit that calculates a stay probability that is a probability that the user will stay in the candidate in the future for each candidate;
Based on the stay probability calculated by the calculation unit, a prediction unit that predicts the stay location where the user stays;
An output unit for outputting the stay location predicted by the prediction unit;
With
The said calculation part is a stay location prediction apparatus which performs the weighting process which weights the said stay probability using the coefficient in which the said candidate with a lower said stay rate becomes a higher value according to the detection result of the said detection part.   The stay location prediction device according to claim 1, wherein the calculation unit performs the weighting process according to the number of times the user has acquired information about traffic or a map in a preset first period. The storage unit stores the stay ratio for each candidate and time zone,
The said calculation part calculates the said stay probability using the said stay ratio of the same time slot | zone as the time slot used as the object of the calculation of the said stay probability by the said calculation part for every said candidate. Stay location prediction device. A first position information acquisition unit that acquires first position information indicating the first position of the user;
The storage unit stores, for each candidate, an inter-candidate movement ratio that is a ratio at which a user moves from the candidate to each of the other candidates.
The calculation unit calculates the stay probability using the inter-candidate movement ratio, which is a ratio of the user moving from the candidate corresponding to the first position to each of the other candidates. The stay location prediction apparatus according to any one of 3. A second position information acquisition unit that acquires a plurality of second position information indicating the plurality of second positions of the user;
An extraction unit that extracts representative positions of the plurality of second positions when the plurality of second positions corresponding to the plurality of second position information acquired by the second position information acquisition unit are included in a predetermined range. When,
Further comprising
The stay candidate prediction apparatus according to any one of claims 1 to 4, wherein the candidate is set based on the representative position extracted by the extraction unit.   The stay candidate prediction apparatus according to claim 5, wherein a predetermined number of preset candidates are set based on a time at which the user stays at each of the plurality of representative positions.

Images