JP7520792B2 - POI series data extension device, POI series data extension method, and program - Google Patents

Description

The present invention relates to a POI series data expansion device, a POI series data expansion method, and a program.

People's movement patterns tend to show regularities, such as going back and forth between home and work, but movement patterns are not necessarily always the same. For example, even if a person rarely visits a store, he or she may stop by while passing by out of curiosity or other reasons. Therefore, it is believed that there is probabilistic fluctuation in the POI (Point of Interest) series for each user.

As a technique for expanding location history data, a method has been disclosed that uses accurate ground-truth data of latitude and longitude and raw data with noise superimposed thereon to estimate the variance of Gaussian noise, and uses this to expand the data by superimposing noise on the ground-truth data (see, for example, Non-Patent Document 1). This technique targets latitude and longitude data, and requires ground-truth data.

As a data expansion method for location history data, a method has been shown in which POI history is divided into regular time intervals, a machine learning model is trained to estimate POIs for time periods when no POIs are recorded, and data is expanded by making estimates and filling in the gaps (see, for example, Non-Patent Document 2). The technology disclosed in Non-Patent Document 2 mentions, as a comparison target for the proposed method, a method of selecting the closest POI or the most popular POI in the vicinity, based on the center of the line segment connecting two POIs.

Feng, J., Li, Y., Zhao, K., Xu, Z., Xia, T., Zhang, J., Jin, D.: DeepMM: Deep Learning Based Map Matching with Data Augmentation. IEEE Trans. Mob. . Comput. 3-6 (2020). https://doi.org/10.1109/TMC.2020.3043500.

Li, Y., Luo, Y., Zhang, Z., Sadiq, S., Cui, P.: Context-aware attention-based data augmentation for POI recommendation. Proc. - 2019 IEEE 35th Int. Conf. Data Eng. Work. ICDEW 2019. 177-184 (2019). https://doi.org/10.1109/ICDEW.2019.00-14

The technology disclosed in the above-mentioned non-patent document 2 requires training a machine learning model, which requires high preparation costs, and the comparison method is deterministic, so it is unable to express the probabilistic fluctuations seen in people's movement patterns.
The present invention has been made to solve the above problems, and has an object to provide a POI sequence data expansion device, a POI sequence data expansion method, and a program that can provide a POI sequence with appropriate fluctuations at low cost.

(1) One aspect of the present invention is a POI sequence data expansion device that includes a user POI sequence acquisition unit that acquires a user POI sequence, which is a history of POIs for each user; a neighborhood POI information acquisition unit that uses a random number to determine a POI from the user POI sequence and acquires neighborhood POI information indicating neighborhood POIs that exist in the neighborhood of the selected POI; a selection unit that selects one neighborhood POI from the neighborhood POIs included in the neighborhood POI information; a replacement insertion unit that replaces the selected POI with the neighborhood POI selected by the selection unit, or inserts the neighborhood POI selected by the selection unit into the user POI sequence; and a user POI sequence output unit that outputs the user POI sequence including the neighborhood POI replaced or inserted by the replacement insertion unit.
(2) In one aspect of the present invention, in the POI series data expansion device described in (1) above, the neighboring POI information includes objective neighboring POI information in which the neighboring POIs are defined based on the physical distance between multiple POIs.
(3) In one aspect of the present invention, in the POI series data expansion device described in (2) above, the neighboring POI information includes subjective neighboring POI information in which the neighboring POI is determined based on the degree of association between multiple POIs based on user behavior.
(4) In one aspect of the present invention, in the POI series data expansion device described in (3) above, the selection unit selects the neighboring POI from either the objective neighboring POI information or the subjective neighboring POI information by selecting the objective neighboring POI information and the subjective neighboring POI information in a predetermined manner .
(5) In one aspect of the present invention, in the POI series data expansion device described in (4) above, the specified method is a method of selection based on a transition probability indicating the degree of association between multiple POIs based on user behavior .
(6) One aspect of the present invention is that in the POI series data expansion device described in (5) above, the method of selection based on the transition probability is such that the greater the effectiveness of the transition probability, the greater the probability that the subjective nearby POI information will be selected.
(7) In one aspect of the present invention, in the POI series data expansion device described in (4) above, the predetermined method is a random selection method .
(8) In one aspect of the present invention, in the POI series data expansion device described in any one of (3) to (7) above, the subjective nearby POI information is provided for each administrative district.
(9) In one aspect of the present invention, in the POI series data expansion device described in any one of (1) to (8) above, the selection unit selects nearby POIs depending on a time period .

( 10 ) One aspect of the present invention is a POI series data expansion method performed by a POI series data expansion device, the POI series data expansion method including: a user POI series acquisition step of acquiring a user POI series, which is a history of POIs for each user; a neighborhood POI information acquisition step of determining one POI from the user POI series using a random number and acquiring neighborhood POI information indicating neighborhood POIs existing in the neighborhood of the one POI; a selection step of selecting one neighborhood POI from the neighborhood POIs included in the neighborhood POI information; a replacement insertion step of replacing the neighborhood POI selected in the selection step with the one POI, or inserting the neighborhood POI selected in the selection step into the user POI series; and a user POI series output step of outputting the user POI series including the neighborhood POI replaced or inserted in the replacement insertion step.

( 11 ) One aspect of the present invention is a program that causes a computer of a POI series data expansion device to execute the following steps: a user POI series acquisition step of acquiring a user POI series, which is a history of POIs for each user; a neighborhood POI information acquisition step of determining one POI from the user POI series using a random number and acquiring neighborhood POI information indicating neighborhood POIs existing in the neighborhood of the one POI; a selection step of selecting one neighborhood POI from the neighborhood POIs included in the neighborhood POI information; a replacement insertion step of replacing the neighborhood POI selected in the selection step with the one POI or inserting the neighborhood POI selected in the selection step into the user POI series; and a user POI series output step of outputting the user POI series including the neighborhood POI replaced or inserted in the replacement insertion step.

The present invention provides a POI sequence data expansion device, a POI sequence data expansion method, and a program that can provide a reasonable fluctuation to a POI sequence at low cost.

FIG. 1 is a diagram illustrating an example of a POI series data expansion device according to an embodiment of the present invention. FIG. 13 is a diagram illustrating an example of substitution. FIG. 13 is a diagram showing an example of insertion. FIG. 11 is a flowchart showing the flow of a POI series data extension process. FIG. 1 is a diagram showing point candidates N1 to Nj. 13 is a diagram showing point candidates acquired by a neighborhood information acquisition unit. FIG. FIG. 13 is a diagram illustrating an example of transition probability. 13 is a diagram showing point candidates acquired by a neighborhood information acquisition unit. FIG. FIG. 1 is a diagram illustrating a configuration example of a POI sequence data expansion device equipped with a learning unit.

The POI sequence data extension device, the POI sequence data extension method, and the program of the present embodiment will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.
In addition, in all the drawings for explaining the embodiment, the same reference numerals are used for the parts having the same functions, and the repeated explanation is omitted. In this embodiment, the POI is categorical data indicating the places that the user visits, such as stores, public facilities, stations, and educational facilities. In addition, the POI is assigned information such as latitude and longitude that allows the comparison of the relative distance between the POIs. The POI series is acquired by an application of a mobile terminal such as a smartphone. The acquired POI series is stored on a user basis, for example, in a server managed by a communication company.

Figure 1 is a diagram showing an example of a POI sequence data extension device according to an embodiment of the present invention. The POI sequence data extension device 100 of this embodiment receives a user POI sequence, which is a history of POIs for each user, and outputs a perturbed user POI sequence by perturbing the input user POI sequence. In this embodiment, the input user POI sequence is acquired from another device. The POI sequence data extension device 100 may store the user POI sequence and acquire the user POI sequence from among the stored user POI sequence, or may acquire the user POI sequence from a medium (such as a flash memory or optical disk) on which the user POI sequence is stored.

The POI sequence data expansion device 100 includes a sequence acquisition unit 110, a neighborhood information acquisition unit 120, a selection unit 130, a replacement and insertion unit 140, and an output unit 150. The sequence acquisition unit 110 acquires an input user POI sequence. The neighborhood information acquisition unit 120 acquires neighborhood POI information indicating neighborhood POIs that exist in the neighborhood of one POI included in the user POI sequence. The selection unit 130 selects one neighborhood POI from the neighborhood POIs included in the neighborhood POI information.

The replacement insertion unit 140 replaces the nearby POI selected by the selection unit 130 with one POI, or inserts the nearby POI selected by the POI selection unit into the user POI series. Details of the insertion position will be described later.

This generates a user POI sequence with fluctuations. The output unit 150 outputs the user POI sequence including the nearby POIs replaced or inserted by the replacement insertion unit 140.

In the above configuration, the sequence acquisition unit 110 is an example of a user POI sequence acquisition unit. The neighborhood information acquisition unit 120 is an example of a neighborhood POI information acquisition unit. The output unit 150 is an example of a user POI sequence output unit.

Next, examples of replacement and insertion by the replacement and insertion unit 140 will be described. In the following description, "series" refers to a user POI series. Also, "point" refers to a POI included in the series.

Figure 2 shows an example of replacement. In Figure 2, the input sequence is connected in the order of points P1, P2, and P3. Note that point P2 corresponds to "one POI" included in the user POI sequence.

Suppose that the replacement and insertion unit 140 replaces point P2 with point PA in the input sequence. In this case, the POI sequence data expansion device 100 outputs a sequence in which points P1, PA, and P3 are connected in that order. This results in a sequence that is given a fluctuation, such as the user moving from point P1 to point P3 via point PA.

Figure 3 shows an example of insertion. In Figure 3, the input sequence is connected in the order of points P1, P2, and P3. Note that point P2 corresponds to "one POI" included in the user POI sequence.

As shown in FIG. 3, the replacement insertion unit 140 inserts PB before or after the target point P2. For example, assume that point PB is inserted after point P2. In this case, the POI sequence data expansion device 100 outputs a sequence in which points P1, P2, PB, and P3 are connected in that order. This results in a sequence that is given a fluctuation, such as the user moving from point P2 to point P3 via point PB.

In the following explanation, the replacement point PA in Figure 2 and the insertion point PB in Figure 3 are referred to as "nearby points."

In this embodiment, there are three selection methods for selecting nearby points. The first is an objective nearby selection method. The second is a subjective nearby selection method. The third is a hybrid selection method that uses both the objective nearby selection method and the subjective nearby selection method. These selection methods will be explained in order. In the following explanation, the above-mentioned "one POI" will be referred to as a "target point."

Prior to the description of the selection method, the overall process flow common to any of the selection methods will be described. FIG. 4 is a diagram showing a flowchart showing the flow of the POI series data extension process executed by the POI series data extension device 100. In the flowchart shown in FIG. 4, the number of target points present in one series is n. For example, it is assumed that n points P 1 , P 2 , ..., P _n of a series configured as starting point P ₀ →P ₁ →P ₂ →... →P _n → ending point P _n+1 are set as target points. In this case, the process is performed to give a fluctuation so that the starting point and the ending _point are not moved. In this embodiment, an example in which the starting _point and the ending point are not moved will be described, but at least one of the starting point and the ending point may be set as the target point.

In FIG. 4, the sequence acquisition unit 110 of the POI sequence data extension device 100 acquires a sequence (step S101). The POI sequence data extension device 100 initializes a loop counter k to 1 (step S102). The POI sequence data extension device 100 determines whether the kth point is a target point for replacement or insertion (step S103).

This step S103 is a process to prevent all points in the series from being target points for replacement insertion. If all points were to be target points for replacement insertion, a series that is significantly changed from the input series would be output, so a determination is made, for example using random numbers, to determine which points are to be target points for replacement insertion with a low probability (for example, a probability of 10% or less).

If it is determined that the kth point is not a target point for replacement insertion (step S103: NO), the process proceeds to step S113. If it is determined that the kth point is a target point for replacement insertion (step S103: YES), the POI series data expansion device 100 determines whether to replace the kth point with the target point or to insert it after the target point. Here, for example, a random number is used to determine whether to replace or insert.

If it is determined in step S104 that replacement is to be performed, the neighborhood information acquisition unit 120 acquires neighborhood POI information for the kth point (step S105). The selection unit 130 selects one neighborhood point from the neighborhood points included in the neighborhood POI information (step S106). The replacement insertion unit 140 replaces the target point with the neighborhood point selected by the selection unit 130 (step S107) and proceeds to step S111.

If it is determined in step S104 that insertion is to be performed, the neighborhood information acquisition unit 120 acquires neighborhood POI information for the kth point (step S108). The selection unit 130 selects one neighborhood point from the neighborhood points included in the neighborhood POI information (step S109). The replacement insertion unit 140 inserts the neighborhood point selected by the selection unit 130 into the series as a point before or after the target point (step S110), and proceeds to step S111.

The POI sequence data expansion device 100 determines whether k=n (step S111). If k=n (step S111: YES), processing has been performed for all target points, so the output unit 150 outputs the sequence (step S112) and ends processing. If k≠n (step S111: NO), there are unprocessed target points, so the POI sequence data expansion device 100 increments k (step S113) and ends processing.

By the POI sequence data extension process described above, the POI sequence data extension device 100 can output a sequence with fluctuations. In light of this POI sequence data extension process, each selection method will be explained below. In the following explanation, nearby point candidates will be referred to as "point candidates."

[Objective Neighborhood Selection Method]
The objective neighborhood selection method is a method based on the physical distance between multiple POIs. A specific explanation will be given with reference to Fig. 5. Fig. 5 is a diagram showing point candidates N1 to Nj of a target point P. In the objective neighborhood selection method, points located within a distance d from the target point P (i.e., located within a radius d) are selected as point candidates.

The neighborhood information acquisition unit 120 acquires neighborhood information including these point candidates. FIG. 6 is a diagram showing the point candidates acquired by the neighborhood information acquisition unit 120. The selection unit 130 selects a neighborhood point from among the point candidates N1 to Nj. The distance d may be fixed at, for example, 10,000 meters, but may also be set for each user according to the typical travel distance of each individual user (for example, the average value of the distance between two consecutive POIs in a series).

The neighborhood information acquisition unit 120 periodically (e.g., once a day) acquires point candidates from the latitude and longitude assigned to the POI, for example, using the Huynhve formula. The selection unit 130 selects nearby points from the point candidates using non-probabilistic criteria (randomly). Note that the neighborhood information acquisition unit 120 may acquire point candidates prepared by another device.

The objective neighborhood selection method selects neighboring points based on their distance from the target point, making it possible to impart reasonable fluctuations to the POI sequence at a very low cost compared to training a machine learning model.

[Subjective Neighborhood Selection Method]
The subjective neighborhood selection method is a method based on the degree of association between multiple POIs based on the user's behavior. Specifically, the actual series is tallied, and the transition probability from point P to other points Nk (k = 1 to r) is calculated. This transition probability indicates the value of the degree of association between POIs. FIG. 7 is a diagram showing an example of the transition probability. In FIG. 7, the transition probability from point P to Nk is indicated by Pk. As a method of calculating the transition probability, there is a method of Pk = S(k)/Q, where Q is the total number of transitions from P to any point, and S(k) is the total number of transitions from P to Nk. In other words, the transition probability is calculated by counting the number of occurrences of the pair by treating the front and rear of the POI in the series as a pair, and dividing by the total number of occurrences of all pairs.

The neighborhood information acquisition unit 120 sets the Nk points up to the Rth place in order from the highest transition probability as point candidates, and acquires neighborhood information including these. Figure 8 is a diagram showing point candidates acquired by the neighborhood information acquisition unit 120. Figure 8 shows the top 20 point candidates N'1 to N'20 with the highest transition probability, where R=20, as an example.

The neighborhood information acquisition unit 120 calculates the transition probability and acquires point candidates, for example, when a specified time arrives or when a POI is updated. The selection unit 130 selects a neighborhood point from the point candidates according to the transition probability or based on a non-probabilistic criterion (random). Note that the neighborhood information acquisition unit 120 may acquire point candidates prepared by another device.

The subjective neighborhood selection method allows for realistic fluctuations in the POI sequence in that neighborhood points are selected based on actual POIs.

Hybrid selection method
The hybrid selection method is a method of selecting a point candidate obtained in the objective neighborhood selection method (hereinafter referred to as the "first point candidate") and a point candidate obtained in the subjective neighborhood selection method (hereinafter referred to as the "second point candidate") based on a predetermined criterion.

In the hybrid selection method, the neighborhood information acquisition unit 120 prepares in advance a point candidate acquired in the objective neighborhood selection method (referred to as the "first point candidate") and a point candidate acquired in the subjective neighborhood selection method (referred to as the "second point candidate").

In addition, the hybrid selection method has three types of selection methods (α, β, γ) used by the selection unit 130. Selection method α is a method in which the predetermined criterion is a non-probabilistic criterion. Specifically, the selection unit 130 randomly selects either the first point candidate or the second point candidate (non-probabilistic criterion), and selects a nearby point from the selected point candidates.

Selection method β is a method in which the predetermined criterion is a probabilistic criterion. Specifically, selection method β is a method in which the transition probability used when obtaining the second point candidate is used. Some transition probabilities are probabilistically meaningful (significant) and some are not. Therefore, whether the transition pattern of interest is significant is determined by comparing it with the transition probability when all POI transition patterns occur with equal probability, by testing the difference in ratios with a significance level α of 0.05, for example.

Among those judged to be significant as a result of the judgment, the points with the highest transition probabilities (up to the top 10) are re-selected as second point candidates. The selection unit 130 selects nearby points from among the second point candidates. On the other hand, when there are only a few series to be counted, there may be no transition probabilities that are in the top 10. In such a case, the selection unit 130 selects nearby points from among the first point candidates. According to selection method β, it is possible to select nearby points based on the actual POI, and in the case of a transition probability that is not significant, a point is selected from the first point candidates, so that a reasonable fluctuation can be given to the POI series at low cost.

Selection method γ is a method that uses the transition probability used when obtaining the second point candidate. In this method, the selection unit 130 selects either the first point candidate or the second point candidate based on probability. For example, if there are a sufficient number of target points in the series in which the second point candidate is tallied, the second point candidate is considered to be more effective than if there are fewer target points.

Therefore, e (0≦e≦1) is set as an index of sufficiency of target points in the series to be aggregated. The calculation method of e is e=M/N, where M is the number of second point candidates corresponding to the target point and N is the maximum number of second point candidates. The larger this e is, the more effective the second point candidates are considered to be compared to the first point candidates. In other words, the degree of effectiveness of the association is considered to be high. Note that the "maximum number of second point candidates" is, for example, N=10 when the top 10 are the second point candidates as described above. Therefore, the index e for a target point where there are 10 second point candidates is 1.

The selection unit 130 selects a nearby point from the second point candidates with a probability e, and selects a nearby point from the first point candidates with a probability 1-e. In this way, it is possible to impart a reasonable fluctuation to the POI sequence at low cost.

As described above, the POI sequence data extension device 100 according to this embodiment outputs a POI sequence with appropriate fluctuations. Therefore, the POI sequence data extension device 100 may be provided with a learning unit that outputs user location information based on the user POI sequence by learning the POI sequence with appropriate fluctuations as the learning subject.

Figure 9 is a diagram showing an example of the configuration of a POI sequence data expansion device 200 equipped with a learning unit. The configuration shown in Figure 9 includes a user POI sequence storage unit 160 and a learning unit 170 in addition to the configuration shown in Figure 1.

In FIG. 9, the output unit 150 successively adds the series output to the user POI series storage unit 160. The learning unit 170 performs learning by referring to the series stored in the user POI series storage unit 160. The user POI series storage unit 160 stores a POI series (in this embodiment, a series of POIs indicated by latitude and longitude) for each user.

When the learning unit 170 constructs a POI visit prediction model, it may do so in the same way as when constructing a model for a series before the POI series data expansion device imparts fluctuations. For example, if the POI visit prediction model is a neural network consisting of an LSTM, a fully connected layer, and a softmax function, the POI series data excluding the latest POI is input in order from oldest to newest, and the cross entropy loss is calculated from the output and the latest POI that has been 1-hot encoded, and the model parameters are updated using the backpropagation method.

If the operation of replacing or inserting a nearby point selected from the first point candidate or the second point candidate is considered as a transformation, the machine learning model trained using the POI series data extension device 200 according to this embodiment becomes robust against probabilistic fluctuations in the series. Therefore, the POI series data extension device 200 improves the performance of the POI visit prediction model that uses a series as input.

In FIG. 9, the learning unit is provided in the POI series data extension device 200, but this is not limited to the above. The learning unit may be provided in a device different from the POI series data extension device 200. For example, the learning unit may be provided in a device that can communicate with the POI series data extension device 100.

In the embodiment described above, the second point candidates may be provided for each administrative district (prefecture or city/town/village). This is because the degree of association between multiple POIs based on user behavior is considered to differ for each administrative district. In this way, the POI series data expansion device can output a series with fluctuations using nearby points that are more in line with reality.

The selection unit 130 may also select nearby points according to the time of day. For example, restaurants that serve alcoholic beverages are mainly open only at night, and department stores are not open late at night, so there are points that have a 0% chance of being visited during certain times of day. Furthermore, there are POIs that some users do not visit during certain times of day, but visit during other times of day. Therefore, the selection unit 130 may select nearby points according to the time of day. In this way, the POI series data expansion device can output a series that has been given fluctuations using nearby points that are more in line with reality.

Furthermore, this makes it possible to provide a POI sequence data expansion device, a POI sequence data expansion method, and a program that can provide appropriate fluctuations to a POI sequence at low cost, thereby contributing to the achievement of Goal 9 of the United Nations-led Sustainable Development Goals (SDGs), which is to "Develop resilient infrastructure, promote sustainable industrialization and foster innovation."

Although the embodiments have been described above, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations can be made without departing from the spirit of the invention. These embodiments are included in the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents described in the claims.
The above-mentioned POI series data extension devices 100 and 200 may be realized by a computer. In this case, a program for realizing the functions of each functional block is recorded on a computer-readable recording medium. The program recorded on the recording medium may be read into a computer system and executed by a CPU to realize the functions. The "computer system" here includes hardware such as an OS (Operating System) and peripheral devices.
In addition, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, CD-ROMs, etc. In addition, "computer-readable recording medium" includes storage devices such as hard disks built into computer systems.
Furthermore, the term "computer-readable recording medium" may include a medium that dynamically stores a program for a short period of time, such as a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line.
Furthermore, the "computer-readable recording medium" may also include a storage medium that holds a program for a certain period of time, such as a volatile memory in a computer system that is a server or a client. The program may be for implementing part of the above-mentioned functions. The program may be capable of implementing the above-mentioned functions in combination with a program already recorded in the computer system. The program may be implemented using a programmable logic device. The programmable logic device is, for example, an FPGA (Field Programmable Gate Array).

100, 200...POI sequence data expansion device, 110...sequence acquisition unit, 120...neighborhood information acquisition unit, 130...selection unit, 140...substitution and insertion unit, 150...output unit, 160...sequence storage unit, 170...learning unit

Claims (11)

A user POI sequence acquisition unit that acquires a user POI sequence that is a history of POIs for each user;
a neighboring POI information acquisition unit that determines one POI from the user POI sequence using a random number and acquires neighboring POI information indicating neighboring POIs existing in the vicinity of the one POI;
A selection unit that selects one nearby POI from the nearby POIs included in the nearby POI information;
a replacement/insertion unit that replaces the neighboring POI selected by the selection unit with the one POI, or inserts the neighboring POI selected by the selection unit into the user POI sequence;
a user POI sequence output unit that outputs the user POI sequence including the neighboring POI replaced or inserted by the replacement insertion unit;
A POI series data expansion device comprising: The POI series data expansion device according to claim 1, wherein the neighboring POI information includes objective neighboring POI information in which the neighboring POIs are determined based on the physical distance between multiple POIs. The POI series data expansion device according to claim 2, wherein the neighboring POI information includes subjective neighboring POI information in which the neighboring POIs are determined based on the degree of association between multiple POIs based on user behavior. The POI series data expansion device according to claim 3 , wherein the selection unit selects the neighboring POI from either the objective neighboring POI information or the subjective neighboring POI information by selecting the objective neighboring POI information and the subjective neighboring POI information using a predetermined method . 5. The POI series data expansion device according to claim 4, wherein the predetermined method is a selection method based on a transition probability indicating a degree of association between a plurality of POIs based on a user's behavior . 6. The POI series data expansion device according to claim 5, wherein the method of selection based on the transition probability is a method in which the greater the effectiveness of the transition probability, the greater the probability that the subjective neighboring POI information is selected. 5. The POI series data expansion device according to claim 4, wherein the predetermined method is a random selection method . The POI series data expansion device according to any one of claims 3 to 7, wherein the subjective nearby POI information is provided for each administrative district. The POI series data expansion device according to any one of claims 1 to 8, wherein the selection unit selects nearby POIs according to a time period. A POI series data extension performed by a POI series data extension device, comprising:
A user POI sequence acquisition step of acquiring a user POI sequence which is a history of POIs for each user;
a neighboring POI information acquisition step of determining one POI from the user POI sequence using a random number and acquiring neighboring POI information indicating neighboring POIs existing in the vicinity of the one POI;
a selection step of selecting one nearby POI from the nearby POIs included in the nearby POI information;
a replacement/insertion step of replacing the one POI with the neighboring POI selected in the selection step, or inserting the neighboring POI selected in the selection step into the user POI sequence;
a user POI sequence output step of outputting the user POI sequence including the neighboring POI replaced or inserted in the replacement and insertion step;
The POI sequence data extension method includes: A computer of the POI series data expansion device
A user POI sequence acquisition step of acquiring a user POI sequence which is a history of POIs for each user;
a neighboring POI information acquisition step of determining one POI from the user POI sequence using a random number and acquiring neighboring POI information indicating neighboring POIs existing in the vicinity of the one POI;
a selection step of selecting one nearby POI from the nearby POIs included in the nearby POI information;
a replacement/insertion step of replacing the one POI with the neighboring POI selected in the selection step, or inserting the neighboring POI selected in the selection step into the user POI sequence;
a user POI sequence output step of outputting the user POI sequence including the neighboring POI replaced or inserted in the replacement and insertion step;
A program that executes the following.

Images