JP2017204174A - Destination prediction apparatus, destination prediction method, and destination prediction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a destination prediction apparatus which can accurately predict a destination of a user in consideration of spatial proximity, a destination prediction method, and a destination prediction program.SOLUTION: A destination prediction apparatus includes a destination prediction model parameter database 18 which stores RNN model parameters for predicting a route from a grid ID sequence, and operates as follows: extracting a grid ID sequence indicating a grid, which is a sub-area obtained by dividing a spatial position, from a sequence of positioning data indicating the spatial position; acquiring RNN model parameters from the destination prediction model parameter database 18, to constitute an RNN model; calculating probabilities of visit to destination candidates by sampling simulation with the RNN model and grid distance from the extracted grid ID sequence; and predicting a destination on the basis of the probabilities of visit to the destination candidates.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a destination prediction apparatus, a destination prediction method, and a destination prediction program, and in particular, to accurately estimate a user's next destination from positioning data that is a series of observed position information. The present invention relates to a destination prediction apparatus, a destination prediction method, and a destination prediction program for realizing the above.

  With the widespread use of mobile terminals such as GPS (Global Positioning System) and Wi-Fi chips and mobile terminals such as smartphones, it has become possible to easily acquire positioning data obtained by measuring the position of the user. Therefore, if the user's destination can be predicted using positioning data from the past to the present, it can be applied to a wide range of services such as information recommendation according to the prediction result and life support.

  A method using sequence mining is known as a conventional method for predicting a user's destination using the positioning data. In this method using series mining, for example, a user's stay area is first extracted by mining processing from GPS positioning data from a portable terminal carried by the user. Furthermore, by predicting the transition behavior between multiple stay areas as a Markov model and calculating the transition probability, a destination that is a stay area where a user staying at a certain place will stay next is predicted. (For example, see Non-Patent Document 1 and Non-Patent Document 2)

  In the conventional method described above, the destination is predicted based on a model that takes into account the short-term state dependence, such as a first-order Markov model, for the transition behavior between the places of stay. On the other hand, it is expected to predict the destination with a smaller error by taking into account long-term dependence.

  As an efficient destination prediction method that takes into account long-term state dependency, a series of positioning data is expressed in a discrete grid space, and transitions between grids are modeled using an RNN (recurrent neural network). A method for predicting the destination based on the transition probability obtained from the RNN model is conceivable.

Andy Yuan Xue, Rui Zhang, Yu Zheng, Xing Xie, Jin Huang, Zhenghua Xu, "Destination prediction by sub-trajectory synthesis and privacy protection against such prediction", In Proc. Of ICDE'13, pp.254-265 (2013 ). Andy Yuan Xue, Jianzhong Qi, Xing Xie, Rui Zhang, Jin Huang, Yuan Li, "Solving the data sparsity problem in destination prediction", VLDB J.24 (2), pp.219-243 (2015).

  However, in the method using the above-described RNN model, spatial proximity between grids is not taken into account when modeling the transition between discrete grids. For this reason, there is a problem that the model predicts a spatially separated transition that cannot physically occur, which deteriorates a prediction error.

  The present invention has been made in view of the circumstances as described above, and is a destination prediction apparatus, a destination prediction method, and the like that can accurately predict a user's destination while considering spatial proximity. The purpose is to provide a destination prediction program.

  In order to achieve the above object, the destination prediction apparatus of the present invention includes a destination prediction model parameter database for storing parameters of an RNN (recurrent neural network) model for predicting a movement path from a grid ID sequence, and a spatial position. A grid ID extraction unit for extracting a grid ID series indicating a grid which is a divided region obtained by dividing a spatial position from a series of positioning data indicating a parameter of the RNN model from the destination prediction model parameter database And the RNN model is constructed, and each destination candidate can be visited by a sampling simulation using the RNN model and the distance between the grids from the grid ID sequence extracted by the grid ID extraction unit. Calculated and calculated A destination prediction unit that predicts a destination based on the possibility of visiting each destination candidate.

  A destination prediction model learning unit that learns the RNN model from a grid ID sequence extracted from the positioning data sequence history and stores the learned RNN model parameters in the destination prediction model parameter database; You may do it.

  Further, the destination prediction unit uses the RNN model based on the grid ID sequence extracted by the grid ID extraction unit in the sampling simulation to correspond to the last point of the positioning data sequence. The transition probability from each grid to each grid and the state vector are calculated, the calculated transition probability is corrected by the weight according to the distance between the grids, and the grid that moves to the next step is sampled based on the corrected transition probability Then, the transition probability and the state vector may be input to the RNN model to repeat the calculation of the transition probability and the state vector in the next step.

  The destination prediction unit may correct the transition probability using a Hadamard product of a weight corresponding to the distance between grids and the transition probability when correcting the transition probability.

  In addition, the destination prediction unit may use the RNN from the grid ID sequence extracted by the grid ID extraction unit using the spatial position of the final point of each series in the history of the positioning data series as the destination candidate. The possibility of visiting a destination candidate may be calculated by a sampling simulation using a model and the distance between grids, and the destination may be predicted based on the calculated possibility of visiting each destination candidate.

  In order to achieve the above object, a destination prediction method of the present invention is a destination prediction method performed by a destination prediction apparatus including a grid ID extraction unit and a destination prediction unit, wherein the grid ID extraction unit includes: Extracting a grid ID series indicating a grid that is a divided region obtained by dividing the spatial position from a series of positioning data indicating a spatial position; and The RNN model parameters are acquired from a destination prediction model parameter database that stores parameters of an RNN (recurrent neutral network) model that predicts a movement route, the RNN model is constructed, and the grid ID extraction unit extracts the parameters. A sump using the RNN model and the distance between grids from the grid ID series Calculating the possibility of visiting each destination candidate by ring simulation and predicting the destination based on the calculated possibility of visiting each destination candidate.

  A destination prediction method performed by a destination prediction apparatus including a grid ID extraction unit, a destination prediction model learning unit, and a destination prediction unit, wherein the destination prediction model learning unit is a sequence of the positioning data. The method may further comprise the step of learning the RNN model from the grid ID sequence extracted from the history and storing the learned RNN model parameters in the destination prediction model parameter database.

  In order to achieve the above object, the destination prediction program of the present invention is a program for causing a computer to function as each part of the destination prediction apparatus of the present invention.

  According to the present invention, it is possible to accurately predict a user's destination while considering spatial proximity.

It is a functional block diagram which shows an example of a structure of the destination prediction apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the positioning data series historical information which concerns on embodiment. It is a schematic diagram which shows an example of the destination prediction model parameter information which concerns on embodiment. It is a schematic diagram which shows an example of the destination candidate information which concerns on embodiment. It is a schematic diagram which shows an example of the input information which concerns on embodiment. It is a flowchart which shows the flow of the process in the learning phase performed by the destination prediction apparatus which concerns on embodiment. It is a flowchart which shows the flow of the process in the prediction phase performed by the destination prediction apparatus which concerns on embodiment. It is a flowchart which shows the flow of the initial extraction process performed by the destination prediction apparatus which concerns on embodiment. It is a flowchart which shows the flow of the extraction process at the time of the data input performed by the destination prediction apparatus which concerns on embodiment. It is a flowchart which shows the flow of the learning process performed by the destination prediction apparatus which concerns on embodiment. It is a schematic diagram which shows the structure of RNN which concerns on embodiment. It is a flowchart which shows the flow of the prediction process performed by the destination prediction apparatus which concerns on embodiment. It is a flowchart which shows the flow of the proximity transition probability calculation process performed by the destination prediction apparatus which concerns on embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the present embodiment, a case will be described in which a user travels to a destination with a portable terminal equipped with a GPS, a Wi-Fi chip, or the like.

  Further, in the present embodiment, an RNN (recurrent neutral network) model (see Non-Patent Document 3 and Non-Patent Document 4 below) that predicts the next destination of the user is learned from the history of the positioning data series. After that, the possibility of visiting the destination candidate is calculated by sampling simulation using the RNN model. As a result, it is possible to solve the data sparseness problem while taking into account the dependence of the state over a long period of time, and the future destination of the user can be predicted with higher accuracy.

      [Non-Patent Document 3] Sepp Hochreiter and Jurgen Schmidhuber, "Long short-term memory", Neural Computation 9 (8): pp.1735-1780 (1997).

      [Non-Patent Document 4] Felix A. Gers, Nicol N. Schraudolph, and Jurgen Schmidhuber, "Learning precise timing with LSTM recurrent networks", Journal of Machine Learning Research, vol. 3, pp. 115-143 (2002).

  As shown in FIG. 1, the destination prediction apparatus 10 according to the present embodiment includes a positioning data series history database 12, a grid ID extraction unit 14, a destination prediction model learning unit 16, a destination prediction model parameter database 18, a destination A candidate database 20 and a destination prediction unit 22 are provided.

  As an example, as shown in FIG. 2, the positioning data series history database 12 includes positioning data (in this embodiment) that indicates a spatial position measured by the mobile terminal when the user moves the mobile terminal. , Latitude and longitude), the positioning number represented by an integer value, the latitude measured as part of the positioning data, the longitude measured as part of the positioning data, and the positioning time The associated positioning data series history information is stored. In the positioning data series history information, a series of positioning data is constituted by a plurality of positioning data each positioned at a plurality of different times in a series of movements of the user.

  The grid ID extraction unit 14 divides the spatial position into a plurality of parts, and assigns a grid ID, which is an identification ID, to each grid that is the divided area by an integer value. In addition, the grid ID extraction unit 14 associates a grid ID with a set of latitude and longitude of each positioning data based on the positioning data series stored in the positioning data series history database 12, thereby obtaining a grid ID. Extract the series.

  The destination prediction model learning unit 16 learns an RNN model for predicting the next destination grid from the grid ID series extracted by the grid ID extraction unit 14. In addition, the destination prediction model learning unit 16 stores the learned parameters of the RNN model in the destination prediction model parameter database 18. The RNN model and the learning method for the RNN model will be described later.

  The destination prediction model parameter database 18 stores, as an example, as shown in FIG. 3, destination prediction model parameter information in which the parameter name of the parameter of the RNN model and the parameter value of the parameter of the RNN model are associated with each other. The

  As shown in FIG. 4 as an example, the destination candidate database 20 corresponds to the destination candidate number in which the destination candidate identification ID is represented by an integer value, the latitude of the destination candidate, and the longitude of the destination candidate. The attached destination candidate information is stored.

  In this embodiment, a set of latitude and longitude of the final point of each series of positioning data is extracted as a destination candidate from a plurality of series of positioning data, and the set of extracted latitude and longitude is the latitude of the destination candidate. And the destination candidate information is constructed by using the longitude and the longitude.

  However, the method of constructing the destination candidate information is not limited to this, and the final point of each series of positioning data is clustered by K-means clustering, Mean-shift clustering, etc., and the clustered subset is used as the destination candidate. Also good. In this case, the destination candidate information may be constructed by setting the average and median values of the latitude and longitude of the final point included in the subset as the latitude and longitude of the destination candidate, respectively.

  The destination prediction unit 22 is learned based on input positioning information that is input as positioning data to be predicted when the destination is predicted, that is, positioning data indicating at least a part of the movement route to the current position of the user. A sampling simulation is performed using the RNN model constructed by using the parameters of the RNN model, destination candidate information, and distance information between grids, and the possibility of visiting each destination candidate is calculated. The user's destination is predicted based on each calculated purpose and the possibility of visiting the candidate.

  The input information described above is a series of positioning data to be predicted when the destination is predicted. As an example, as shown in FIG. 5, the identification ID of the positioning data in which a set of latitude and longitude is positioned. Is information in which a positioning number represented by an integer value, a latitude that is a part of positioning data, a longitude that is a part of positioning data, and a positioning time are associated with each other. The destination prediction unit 22 is constructed by using the parameters of the RNN model as described above based on the grid ID sequence extracted by the grid ID extraction unit 14 from the positioning data sequence included in the input information. The user's destination is predicted using the RNN model, destination candidate information, and distance information between grids.

The destination prediction unit 22 includes the proximity transition probability calculation unit 4.
The proximity transition probability calculation unit 24 calculates a proximity transition probability according to the distance between the grids for each grid using a weight vector based on the distance between the grids. Moreover, the proximity transition probability calculation unit 24 updates the transition probability to each grid so as to correct the calculated proximity transition probability.

  In addition, when performing the sampling simulation, the destination predicting unit 22 uses the RNN model on the basis of the grid ID sequence extracted from the positioning data series, and each of the grids corresponding to the final point of the positioning data series. The transition probability to the grid and the state vector are calculated. Then, the destination prediction unit 22 corrects the calculated transition probability with a weight according to the distance between the grids, samples the grid that moves to the next step based on the corrected transition probability, Is input to the RNN model, and the transition probability and state vector in the next step are calculated repeatedly.

  The destination prediction apparatus 10 according to the present embodiment is configured by a computer apparatus including, for example, a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory) that stores various programs. . Moreover, the computer which comprises the destination prediction apparatus 10 may be provided with memory | storage parts, such as a hard disk drive and a non-volatile memory. In the present embodiment, the CPU reads and executes a program stored in a storage unit such as a ROM or a hard disk, whereby the hardware resources and the program cooperate to realize the above-described function.

  In the present embodiment, the positioning data series history database 12, the destination prediction model parameter database 18, and the destination candidate database 20 are provided in the destination prediction apparatus 10, but not limited thereto. It may be provided outside the destination prediction apparatus 10.

  The processing executed by the destination prediction apparatus 10 having the functions as described above includes a learning phase for storing parameters of the RNN model, a prediction phase for predicting the destination based on input input information, , Divided into

  First, the outline of the processing flow in the learning phase by the destination prediction apparatus 10 according to the present embodiment will be described with reference to the flowchart shown in FIG.

  In step S101, the grid ID extraction unit 14 acquires the positioning data sequence history information from the positioning data sequence history database 12, and based on the positioning data sequence included in the acquired positioning data sequence history information, the grid ID sequence is obtained. Performs extraction processing when inputting data to be extracted. The extraction process at the time of data input will be described later with reference to the flowchart shown in FIG.

  In step S103, the destination prediction model learning unit 16 performs a learning process of learning the RNN model from the extracted grid ID series. The learning process will be described later with reference to the flowchart shown in FIG.

  In step S <b> 105, the destination prediction model learning unit 16 stores the learned parameters of the RNN model in the destination prediction model parameter database 18.

  Next, an outline of the flow of processing in the prediction phase by the destination prediction apparatus 10 according to the present embodiment will be described using the flowchart shown in FIG.

  In step S <b> 201, the grid ID extraction unit 14 inputs input information, and performs an extraction process at the time of data input for extracting a grid ID series based on the positioning data series included in the input information. The extraction process at the time of data input will be described later with reference to the flowchart shown in FIG.

  In step S <b> 203, the destination prediction unit 22 acquires parameters of the RNN model from the destination prediction model parameter database 18 and acquires destination candidate information from the destination candidate database 20. In addition, the destination prediction unit 22 uses the RNN model constructed using the acquired RNN model parameters from the extracted grid ID series, and the acquired destination candidate information to determine the user's destination. Perform prediction processing to predict. The prediction process will be described later using the flowchart shown in FIG.

  Here, the process of the grid ID extraction part 14 in this embodiment is demonstrated in detail. The grid ID extraction unit 14 performs an initial extraction process performed at the start of destination prediction and an extraction process at the time of data input performed when a series of positioning data is input.

  First, the flow of the initial extraction process of the grid ID extraction unit 14 will be described with reference to the flowchart shown in FIG. The initial extraction process is a process performed prior to the extraction process at the time of data input.

  In step S <b> 301, the grid ID extraction unit 14 divides the latitude / longitude space into a lattice having a predetermined interval, thereby creating a plurality of grids, each of which is a rectangular region.

  Note that the method of dividing the latitude / longitude space is not limited to this, and may be divided so that each set of latitude and longitude is included in one of the divided regions. For example, it is disclosed in Non-Patent Document 5 below. You may divide | segment using the well-known technique geo-hash. In the present embodiment, a plurality of grids having the same shape are created by dividing the latitude / longitude space into a lattice at regular intervals. However, the shape of the grid is not limited to this, and a plurality of grids having different sizes or shapes are used. May be created.

[Non-Patent Document 5] Kisung Lee, Raghu K. Ganti, Mudhakar Srivatsa, Ling Liu, "Efficient spatial query processing for big data", In Proc. Of SIGSPATIAL'14, pp.469-472 (2014).

  In step S303, the grid ID extraction unit 14 assigns an integer value as a grid ID to each grid obtained by the division. In this embodiment, the grid ID extraction part 14 avoids duplication of grid ID in each of a plurality of grids by assigning an integer value to each grid while increasing the value by 1 in order from 0.

  Next, the flow of extraction processing when data is input by the grid ID extraction unit 14 will be described with reference to the flowchart shown in FIG.

  In step S401, the grid ID extraction unit 14 determines a corresponding grid for each positioning data in the positioning data series included in the input information.

  In step S403, the grid ID extraction unit 14 outputs the integer value series assigned to the corresponding grid to the destination prediction unit 22 as a grid ID series.

  Next, learning processing executed by the destination prediction model learning unit 16 in the present embodiment will be described in detail with reference to the flowchart shown in FIG.

  Hereinafter, the vector is expressed in bold in the mathematical expression, and in the text, the vector symbol is represented by “^”.

  In step S501, the grid ID extraction unit 14 acquires the positioning data sequence history information from the positioning data sequence history database 12, and based on the positioning data sequence included in the acquired positioning data sequence history information, the above-described learning time. A series of grid IDs obtained by the extraction process is extracted. In addition, the grid ID extraction unit 14 calculates a series of feature vectors of one-hot expression including dimensions of all grid numbers. The feature vector of the one-hot expression here is a binary vector in which the element of the corresponding grid ID is 1 and the other elements are 0.

  In step S503, the destination prediction model learning unit 16 learns an RNN model that predicts a feature vector at the next positioning timing with respect to a feature vector at an arbitrary positioning timing, using the calculated feature vector series as teacher data. In addition, as the technique of RNN, RNN provided with LSTM (long short term memory) which is a known technique disclosed in Non-Patent Document 3 described above, and a forgetting gate which is a known technique disclosed in Non-Patent Document 4 described above It is possible to use a known technique such as an RNN provided with an attached LSTM.

As shown in FIG. 11, in the prediction of the feature vector of the next positioning timing using the RNN model, first, the feature vector of the one-hot expression at the current positioning timing is obtained based on the parameters (S30), and the obtained feature is obtained. The vector is embedded in a low dimension (S32). Specifically, a feature vector embedded in another vector space is obtained by taking the product of one of the model parameters w _embed ^ and the feature vector of the one-hot expression.

  Next, the obtained feature vector and the potential state vector (S34) obtained using the RNN model at the previous positioning timing are input, and the LSTM disclosed in Non-Patent Document 3 described above is used. The state vector at the current positioning timing is calculated (S36).

  Finally, the feature vector at the next positioning timing, that is, the probability of movement to each grid is predicted from the calculated state vector using a softmax function (S38).

  For the learning of the RNN, a technique such as Truncated BPTT (Truncated back propagation through time) disclosed in Non-Patent Document 3 described above can be used.

In step S505, the learned parameters of the RNN model are stored in the destination prediction model parameter database 18. Note that the destination prediction model parameter information shown in FIG. 3 is an example of a parameter of the RNN model provided with the above-described LSTM. That is, w _embed ^ is a parameter for embedding the input feature vector of the one-hot expression in another low-dimensional vector space. In addition, w _in is a parameter in the LSTM input gate, w _out is a parameter in the output gate, and w _c is a parameter in the memory cell. W _softmax ^ is a parameter for calculating a feature vector at the next positioning timing from the state vector of the RNN using a _softmax function.

  Next, prediction processing executed by the destination prediction unit 22 in the present embodiment will be described in detail using the flowchart shown in FIG.

  Originally, the RNN model is a model for estimating a transition destination state when transitioning from the current positioning timing to the next positioning timing. However, when estimating the destination, it is not always possible to reach the destination from the current location in one step. For this reason, it is impossible to predict the destination by using a general RNN model.

  Therefore, in the present embodiment, the destination prediction unit 22 inputs a grid ID sequence up to the current positioning timing, simulates a multi-step transition by random sampling using an RNN model, and each destination candidate. Count the number of times to reach. The counted value can be considered as a visitability to each next destination candidate for the input. Therefore, the destination is determined based on the counted value.

  Hereinafter, a specific method for determining the destination will be described in detail.

  In step S601, the destination predicting unit 22 initializes a visit probability vector P ^ representing the possibility of visiting each destination candidate with a zero vector.

  In step S603, the destination prediction unit 22 initializes the initial state vector ＾ with a zero vector.

  In step S605, the destination prediction unit 22 acquires the parameters of the RNN model from the destination prediction model parameter database 18, and constructs the RNN model using the acquired parameters of the RNN model. At this time, as described above, the destination prediction unit 22 uses the feature vector of the one-hot expression generated from the grid ID at the current positioning timing and the state vector obtained from the RNN model at the previous positioning timing as the RNN model. To output the feature vector and state vector of the prediction result at the next positioning timing.

  Here, the feature vector of the prediction result can be regarded as a transition probability to each grid. Hereinafter, the feature vector of the one-hot expression generated from the grid ID of the grid g to be processed, and the state vector h ^ obtained from the RNN model at the previous positioning timing of the grid g to be processed , A function RNN (grid g, state vector h ^) obtained using the RNN model.

  In step S <b> 607, the destination prediction unit 22 starts loop A that repeats the processing in the loop until the variable i, which is the number of simulation trials, changes from 1 to N. For example, N = 1000.

  In step S609, the destination prediction unit 22 uses the grid ID extraction unit 14 to obtain a grid ID sequence obtained by executing the above-described extraction process at the time of prediction on the positioning data sequence included in the input information. Assuming G, a loop B that repeats the processing in the loop is started for each grid g of the grid ID of each element in the grid ID series G.

  In step S611, the destination predicting unit 22 uses the RNN model and the transition probability p ^ to each grid from the grid g and the state vector h ^ at the previous positioning timing using the RNN model. And further update the state vector ＾.

  In step S613, the destination prediction unit 22 determines whether or not the processing in the loop has been executed for all the grids g in the grid ID series G. If the determination is negative, the processing in the loop is repeated. If the determination is affirmative, loop B is terminated and the process proceeds to step S615.

  In step S615, the destination prediction unit 22 starts a loop C that repeats the processing in the loop until the variable j, which is the number of samplings, changes from 1 to M. Note that M is the upper limit number of samplings, for example, M = 100.

  In step S616, the proximity transition probability calculation unit 24 of the destination prediction unit 22 executes a proximity transition probability calculation process for updating the transition probability p ^ to each grid. Hereinafter, the flow of the proximity transition probability calculation process will be described with reference to a flowchart showing the flow of the proximity transition probability calculation process shown in FIG.

  In step S701, the proximity transition probability calculation unit 24 calculates a weight vector ＾ based on the distances from the grid g to all | G | grids based on the following equation (2).

  Here, Dist (..., ...) represents the distance [meter] between the two grids. σ is a fixed value that determines the degree of variation in the weight of each element, and in this embodiment, σ = 200 [meters].

  In step S703, the proximity transition probability calculation unit 24 calculates the proximity transition probability to each grid according to the following equation (3) based on the weight vector s ^ and the transition probability p ^ to each grid. The transition probability p ^ to each grid is updated, and the process proceeds to step S617 in FIG.

  Here, “◯” in the above equation (3) represents a Hadamard product of vectors.

  In this way, the transition probability to each grid is corrected according to the distance between the grids, and the grid moving to the next step is sampled based on the corrected transition probability to each grid, as will be described later. Thus, it is possible to avoid predicting a transition that is physically separated that cannot occur physically, and thereby it is possible to predict the destination with a smaller error.

  In step S617 of FIG. 12, the destination predicting unit 22 randomly samples the grid g, which is a grid when moved to the next step, according to the transition probability p ^ to each grid.

  In step S619, the destination prediction unit 22 determines whether the distance to the destination candidate d closest to the grid g is closer to the threshold ε. In the present embodiment, the average value of the distance that the user moves in one step is set as a threshold value ε, for example, ε = 100 m (meters). If the determination in step S619 is affirmative, the process proceeds to step S625. If the determination is negative, the process proceeds to step S621.

  In step S621, the destination prediction unit 22 calculates the transition probability p ^ to each grid from the grid g and the state vector h ^ one step before using the RNN model based on the above equation (1). Further, the state vector h ^ is updated.

  In step S623, the destination prediction unit 22 determines whether or not the variable j has become M. If the determination is negative, the processing in the loop is repeated, and if the determination is affirmative, loop C is terminated. Then, the process proceeds to step S627.

  In step S625, the destination prediction unit 22 adds 1 to the element Pd of the visit probability vector P ^ corresponding to the destination candidate d based on the following equation (4).

  In step S627, the destination predicting unit 22 determines whether or not the variable i is N. If the determination is negative, the processing in the loop A is repeated, and if the determination is positive, the loop A is terminated. Then, the process proceeds to step S629.

  In step S629, the destination predicting unit 22 calculates the grid ID of the destination candidate most likely to visit from the visit probability vector P ^ representing the possibility of visiting each destination candidate.

  In step S631, the destination prediction unit 22 acquires the latitude and longitude of the destination candidate that matches the calculated grid ID from the destination candidate information acquired from the destination candidate database 20, and displays it on the display means. It is output by storing it in the storage means.

  As described above, the destination prediction apparatus 10 according to the present embodiment receives a history of positioning data series that is a set of latitude and longitude from the positioning data series history information, and is included in the received history of positioning data series. A grid ID series is extracted based on the positioning data series. Further, the destination prediction apparatus 10 learns the RNN model from the grid ID series, and stores the learned parameters of the RNN model as destination prediction model parameter information.

  The destination prediction apparatus 10 receives a series of positioning data as input information, and extracts a grid ID series based on the received positioning data series. Further, the destination prediction apparatus 10 acquires the parameters of the destination prediction model, constructs an RNN model, and visits each destination candidate by sampling simulation using the RNN model from the extracted grid ID series. , And the calculated visitability is corrected according to the distance between the grids, and the latitude and longitude of the destination are output as the destination prediction results based on the corrected visitability.

  In this way, the RNN model is learned using the history of the user's positioning data series, and the sampling simulation using the RNN model is performed from the newly acquired user's positioning data series. By this method, it is possible to consider the dependence of the state over a long period of time. Furthermore, the same method can be used to predict the destination based on the learning data that includes only a part of the new user's positioning data series, so that the conventional Markov model is expanded to the third or higher order. Therefore, it is sufficient to prepare learning data with a small number of movement patterns, thereby solving the problem of data sparseness. In addition, it is possible to avoid predicting a transition that is physically separated which cannot occur physically, and thereby it is possible to predict a destination with a smaller error. Based on the above effects, the possibility of visiting a destination candidate is calculated, so that the user's destination can be accurately predicted.

  In the present embodiment, the case where a set of latitude and longitude is used as the positioning data indicating the spatial position has been described. However, the present invention is not limited to this, and in the space where the destination is predicted as the positioning data indicating the spatial position. Information that can identify the position may be used.

  Further, in the present embodiment, the operation of the components of the function shown in FIG. 1 is constructed as a program and is installed and executed on a computer used as the destination prediction apparatus 10, but the present invention is not limited to this. It may be distributed.

  Further, the constructed program may be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and installed in a computer or distributed.

DESCRIPTION OF SYMBOLS 10 Destination prediction apparatus 12 Positioning data series history database 14 Grid ID extraction part 16 Destination prediction model learning part 18 Destination prediction model parameter database 20 Destination candidate database 22 Destination prediction part

Claims (8)

A destination prediction model parameter database for storing parameters of an RNN (recurrent neural network) model for predicting a movement route from a grid ID sequence;
A grid ID extraction unit that extracts a grid ID series indicating a grid that is a divided region obtained by dividing the spatial position from a series of positioning data indicating the spatial position;
The RNN model is acquired from the destination prediction model parameter database to construct the RNN model, and the RNN model and the distance between grids are used from the grid ID series extracted by the grid ID extraction unit. A destination prediction unit that calculates the possibility of visiting each destination candidate by sampling simulation, and predicts the destination based on the calculated possibility of visiting each destination candidate;
Destination prediction device with A destination prediction model learning unit that learns the RNN model from the grid ID sequence extracted from the history of the positioning data sequence, and stores the learned RNN model parameters in the destination prediction model parameter database; The destination prediction apparatus according to claim 1. The destination prediction unit uses the RNN model based on the grid ID sequence extracted by the grid ID extraction unit in the sampling simulation, and uses each of the grids corresponding to the final point of the positioning data series. Calculate the transition probability to the grid and the state vector,
The calculated transition probability is corrected with a weight according to the distance between the grids, and the grid that moves to the next step is sampled based on the corrected transition probability, and the transition probability and the state vector are stored in the RNN model. The destination prediction apparatus according to claim 1, wherein the destination prediction apparatus repeatedly inputs and calculates the transition probability and the state vector in the next step. The destination prediction unit according to claim 3, wherein the destination prediction unit corrects the transition probability using a Hadamard product of a weight corresponding to a distance between grids and the transition probability when correcting the transition probability. apparatus. The destination prediction unit uses the RNN model from the grid ID sequence extracted by the grid ID extraction unit, with the spatial position of the final point of each series in the history of the positioning data series as the destination candidate. 5. The possibility of visiting a destination candidate is calculated by a sampling simulation using the distance between the grids, and the destination is predicted based on the calculated possibility of visiting each destination candidate. The destination prediction apparatus according to item 1. A destination prediction method performed by a destination prediction apparatus including a grid ID extraction unit and a destination prediction unit,
The grid ID extracting unit extracting a grid ID series indicating a grid which is a divided region obtained by dividing a spatial position from a series of positioning data indicating a spatial position;
The destination prediction unit acquires the parameters of the RNN model from a destination prediction model parameter database that stores parameters of a recurrent neural network (RNN) model that predicts a travel route from a grid ID series, and constructs the RNN model. Then, the possibility of visiting each destination candidate is calculated from the grid ID series extracted by the grid ID extraction unit by sampling simulation using the RNN model and the distance between the grids. Predicting a destination based on potential visits to destination candidates,
Destination prediction method having The destination prediction method according to claim 6, which is performed by a destination prediction device including a grid ID extraction unit, a destination prediction model learning unit, and a destination prediction unit,
The destination prediction model learning unit learns the RNN model from the grid ID series extracted from the positioning data series history, and stores the learned parameters of the RNN model in the destination prediction model parameter database. The destination prediction method according to claim 6, further comprising a step.   The program for functioning a computer as each part of the destination prediction apparatus of any one of Claims 1-5.

Images