JP2010267207A - Unusualness evaluation device, unusualness evaluation method, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To evaluate how different an event data series obtained by a user's behavior is from a standard pattern showing a user's daily behavior. <P>SOLUTION: A data acquisition part 11 acquires the event data series by the user's behavior. A pattern storage part 21 stores a standard pattern series showing the user's daily behavior. A score calculation part 31 calculates a matching score corresponding to an optimal alignment where similarity between the event data series and the standard pattern series becomes largest on the basis of dynamic programming. An optimal alignment acquisition part 32 acquires the optimal alignment on the basis of the matching score calculated by the score calculation part 31. A behavior evaluation part 33 evaluates a difference between events included in the event data series and the standard pattern series in the optimal alignment. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an extraordinary evaluation apparatus, an extraordinary evaluation method, and a computer program that acquire an event data series indicating a user's action and evaluate how unusual the action is.

  Conventionally, a technique for detecting an abnormality of an observation target by applying a statistical method to a measurement value acquired by a sensor is known. For example, in the unique pattern detection system described in Patent Document 1, the model learning server extracts a parent sensor group for each sensor based on the received data, performs parameter learning for the parent sensor group, and performs optimal processing. Specific sensor groups and statistical parameters are obtained, and the unique pattern detection server determines the specificity of the observation patterns received sequentially from the database server using the parent sensor group information and statistical parameters. Thus, it is possible to determine whether or not the obtained data is non-stationary.

JP 2007-279887 (paragraphs 0035 and 0036, FIG. 5)

  By the way, when sensing the user's behavior, information on how the observation data deviates from the reference is an important clue to know the user's behavior. However, if an attempt is made to sense a user's behavior using the unique pattern detection system described in Patent Document 1, how far the observed data deviates from the reference pattern even though the specificity of the observed data can be determined. That is, it is impossible to know how much data is extraordinary.

  The present invention has been made in view of the above circumstances, and evaluates how much an event data series obtained by a user's behavior is not relative to a reference pattern representing the user's daily behavior. An object of the present invention is to provide an extraordinary evaluation apparatus, an extraordinary evaluation method, and a computer program.

The extraordinary evaluation apparatus according to the present invention has the following configuration.
(1) Data acquisition means for acquiring an event data series in which event data representing an event generated by a user's action and time information when the event occurs are arranged in the order of event occurrence, and generated in the daily action of the user Pattern storage means for storing a reference pattern sequence in which event data representing an event to be performed and time information at which the event occurs are arranged in the order of occurrence of the event, and comparing the event data sequence and the reference pattern sequence, A score calculation unit that calculates a matching score corresponding to the optimal alignment with the highest degree of similarity between them based on dynamic programming, and an optimal that acquires the optimal alignment based on the matching score calculated by the score calculation unit Alignment acquisition means and the optimum alignment Definitive, and aspects comprising evaluation means for evaluating the extraordinary property from differences between events included in the event data sequence and the reference pattern sequence.
According to this aspect, the event data series and the reference pattern series can be efficiently compared, and the degree of similarity can be evaluated. It is also possible to evaluate differences between events included in both series.

(2) In the configuration of (1), for each pair of the event of the event data series and the event of the reference pattern series, a score function is defined based on time information corresponding to each event, and the score for all pairs The alignment that maximizes the sum of the functions is set as the optimum alignment.
According to this aspect, the similarity between the event data series and the reference pattern series can be evaluated based on the time information of the event.

(3) In the configuration of (1), the event data series includes user movement history data.
According to this aspect, the movement history of the user can be compared with the reference pattern series.
Further, the extraordinary evaluation method according to the present invention has the following configuration.

  (4) A reference pattern sequence in which event data representing events that occur in a user's daily behavior and time information at which the event occurs is arranged in the order of event occurrence is stored, and the event that is generated by the user's behavior Event data series in which event data representing the event information and the time information at which the event occurred are arranged in the order of event occurrence, and the event data series and the reference pattern series are compared, and the similarity between the two series is the highest. A matching score corresponding to a higher optimal alignment is calculated based on dynamic programming, the optimal alignment is acquired based on the calculated matching score, and the event data series and the reference pattern in the optimal alignment Unusuality due to differences between events in the series Worth and aspects.

According to this aspect, the event data series and the reference pattern series can be efficiently compared, and the degree of similarity can be evaluated. It is also possible to evaluate differences between events included in both series.
(5) In the configuration of (4), for each pair of the event of the event data series and the event of the reference pattern series, a score function is defined based on time information corresponding to each event, and the score for all pairs The alignment that maximizes the sum of the functions is set as the optimum alignment.
According to this aspect, the similarity between the event data series and the reference pattern series can be evaluated based on the time information of the event.

(6) In the configuration of (4), the event data series includes user movement history data.
According to this aspect, the movement history of the user can be compared with the reference pattern series.
The computer program according to the present invention has the following configuration.

(7) A computer program for causing a computer to perform an evaluation of unusualness in a user's behavior, wherein event data representing events generated by the user's behavior and time information when the event occurred are in the order of event occurrence. A data acquisition step for acquiring an array of event data, and a pattern for storing a reference pattern sequence in which event data representing events that occur in the daily behavior of the user and time information at which the event occurs are arranged in the order of event occurrence A score calculating step of comparing the event data series with the reference pattern series and calculating a matching score corresponding to the optimum alignment with the highest similarity between the two series based on dynamic programming And calculated in the score calculation step An optimum alignment obtaining step for obtaining the optimum alignment based on the matching score, and an evaluation step for evaluating a difference between the event data series and the event included in the reference pattern series in the optimum alignment. .
Let it be an aspect.
According to this aspect, the event data series and the reference pattern series can be efficiently compared, and the degree of similarity can be evaluated. It is also possible to evaluate differences between events included in both series.

  According to the present invention, the score calculation means calculates the matching score corresponding to the optimal alignment that maximizes the similarity between the event data series and the reference pattern series based on the dynamic programming, and the calculated matching Based on the score, the optimal alignment acquisition unit acquires the optimal alignment, and the evaluation unit evaluates the difference between the event data series and the event included in the reference pattern series in the optimal alignment. Therefore, it is possible to compare the event data series with the reference pattern pattern series and evaluate the similarity. It is also possible to evaluate differences between events included in both series.

The block diagram of the action evaluation apparatus which concerns on one Embodiment of this invention. The figure which shows an example of the time series event data which an event data acquisition part acquires. The figure which shows an example of the daily action pattern memorize | stored in a pattern memory | storage part. The flowchart of the data series evaluation process performed in a process part. The flowchart which shows the detail of a matching score calculation process. It is a figure which shows an example of a sequence alignment. The figure which shows an example of the optimal alignment of the time series event data shown in FIG. 2, and the daily action pattern shown in FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a block diagram of an extraordinary evaluation apparatus according to an embodiment of the present invention.
This extraordinary evaluation apparatus includes an input unit 10, a storage unit 20, a processing unit 30, and an output unit 40.
The input unit 10 includes an event data acquisition unit 11 that acquires time-series event data based on user behavior, and outputs the time-series event data acquired by the event data acquisition unit 11 to the processing unit 30. The time-series event data is a series in which event data generated along with the occurrence of events indicating various user actions is arranged in the order in which the events have occurred, and represents an event occurrence history.

In the present embodiment, the event data acquisition unit 11 acquires user position data measured by the global positioning system (GPS), and the time series event data includes the movement history of the user detected by the GPS device. Shall be included.
The event data acquisition unit 11 includes a GPS positioning unit or receives measurement data from an external GPS positioning device (not shown). When the event data acquisition unit 11 includes a GPS positioning unit, the extraordinary evaluation device is configured as a portable device that can be carried by the user, and sequentially acquires the user's position data. Alternatively, the event data acquisition unit 11 may be configured to be removable from the extraordinary evaluation device, and the user may carry only the event data acquisition unit 11. When the user carries an external GPS device, the event data acquisition unit 11 receives position data from the GPS device. The event data acquisition unit 11 may receive position data from an external GPS device via a wired or wireless network.

  Hereinafter, the time series event data is represented as s. In addition, each event data included in the time series event data s is represented as s [1], s [2], s [3],... In the order of appearance of events, and the total number of events included in the time series event data s | s | The i-th to j-th subsequence of the time series event data s is represented as s [i... J] (where i <j).

  FIG. 2 shows an example of time-series event data acquired by the event data acquisition unit 11. In this example, the user visits a predetermined place as one event. The event data includes the place visited by the user, the time of arrival at the place, and the time of departure from the place.

  In the example shown in FIG. 2, since the time series event data s includes event data of four events, | s | = 4. For example, the event data s [1] represents an event that the user arrived at “home” at 0:00 and departed at 9:00. The event data s [2] represents an event that the user arrived at “XX station” at 10:00 and departed at 10:05. The event data s [3] represents the event that the user arrived at “△△ station” at 10:30 and departed at 10:35, and s [4] the user arrived at “workplace” at 10:50 Represents an event that departed at 17:00. Thus, the time series event data s is data in which event data including event occurrence (visit by a user) and time information (arrival time and departure time) are arranged in the order of event occurrence.

The input unit 10 may include input means such as a keyboard and operation switches in addition to the event data acquisition unit 11. Alternatively, an input terminal to which an external input device is connected may be provided.
The storage unit 20 includes a storage device such as a flash memory, for example, and includes a pattern storage unit 21. The pattern storage unit 21 stores in advance one or a plurality of behavior patterns that the user performs on a daily basis. Each daily behavior pattern includes event data of events that the user performs on a daily basis in chronological order.

  In the following, a set of daily behavior patterns is represented by Q, and each daily behavior pattern that is an element of the set Q is represented by q (= q1, q2...). In addition, the event data included in the daily behavior pattern q is expressed as q [1], q [2], q [3], ... in the order of appearance of events, and the total number of events included in the pattern q is | q | . The i-th to j-th subsequence of the daily behavior pattern q is represented as q [i ... j] (where i <j).

  Such daily behavior patterns, for example, may be manually registered by the user, or time series event data s input in the past is accumulated, and a series pattern mining technique (Rakesh Agrawal and Ramakrishnan Srikant, “Mining Sequential Patterns”) , 11th International Conference on Data Engineering, pp 3-14, 1995)). When an event pattern is extracted from accumulated data, an event start time, an end time, and an average and variance of these times can be stored as time information.

  FIG. 3 is a diagram illustrating an example of the daily behavior pattern q stored in the pattern storage unit 21. Similar to the time-series event data s shown in FIG. 2, the daily behavior pattern q is expressed in a form in which event data of discrete events are arranged in time-series order. The event data included in the daily behavior pattern q includes a place where the user normally visits and time information related to the visit.

  In the example shown in FIG. 3, the daily behavior pattern q includes event data of three events, and | q | = 3. For example, the event data q [1] represents an event in which the user arrives at “home” at 0:00 and departs at 9:00. The event data q [2] represents an event that the user arrives at “XX station” at 10:00 and departs at 10:05, and q [3] indicates that the user goes to “workplace” at 10:50. Represents an event that arrives and departs at 19:00. The time information included in these event data may be an average value of the arrival time and departure time of each place. In the example of FIG. 3, the distribution of arrival time and departure time of each place is also included in the event data. As described above, the daily behavior pattern q is data in which event data that is normally performed by the user and event data including time information thereof are arranged in the order in which the events occur.

The processing unit 30 has a CPU (not shown) and controls the operation of the entire apparatus. The processing unit 30 includes a ROM in which a control program and the like are written, a RAM for providing a work area necessary for program execution, and the like.
The processing unit 30 further includes a score calculation unit 31, an optimal alignment acquisition unit 32, and an action evaluation unit 33. The processing unit 30 evaluates the similarity between the time series event data s obtained from the event data acquisition unit 11 and each daily action pattern q stored in the pattern storage unit 20, and most closely Process to select a pattern. Also, the difference between the daily behavior pattern q that is most similar to the time series event data s is evaluated. In the present embodiment, an optimal sequence alignment is acquired between the time-series input data s and the daily behavior pattern q by using dynamic programming used in the field of bioinformatics for similarity evaluation. .

  Alignment refers to an operation of arranging two sequences in association with each other. Assume that a score function is defined that represents the degree of coincidence of a pair of elements of each series, and the sum of the score function values represents the similarity between the two series. At the time of alignment, a gap symbol (−) is inserted in order to compensate for a missing element in any series.

  FIG. 6 is a diagram illustrating an example of a sequence alignment. In FIG. 6, two types of alignment are shown for two character strings X: “ACGT” and Y: “ATCGT” having different lengths. Although the number of elements (number of characters) included in the two character strings is different, a gap (-) is inserted to associate both character strings.

  In FIG. 6A, a gap (−) is inserted between A and C of the character string X. Then, “A” of the character string X and “A” of the character string Y are paired, and the gap of X and “T” of Y are paired. Further, “C” of the character string X and “C” of the character string Y, “G” of Y and “G” of Y, “T” of X and “T” of Y form a pair.

  In FIG. 6B, a gap (−) is inserted at the beginning of the character string Y, and two gaps are inserted at the end of the character string X. For this reason, the gap between the character string “A” and the character string Y is a pair. Further, “C” of character string X and “A” of character string Y, “G” of Y and “T” of Y, and “T” of X and “C” of Y are paired. The two gaps at the end of the character string X are paired with “G” and “T” of Y, respectively.

  A score function is defined for each such pair, and the sum of the scores of all pairs in the alignment is used as the score of the alignment. The alignment with the maximum score is the optimal alignment between the two sequences, and hereinafter, this maximum score is referred to as a matching score. The similarity between the two sequences is the highest in the optimal alignment.

Details of the alignment process are detailed in bioinformatics textbooks (eg Tatsuya Akutsu, “Mathematics and Algorithms of Bioinformatics”, Kyoritsu Shuppan, 2007).
In order to select the optimal alignment, it is necessary to define the score function w (x, y) for the pair of the element x of the series X and the element y of the series Y. The score function w (x, y) is assumed to be a real value. For example, the value of the score function w (x, x) for the same element is larger than the value of the score function w (x, −) for the gap. Define as follows.

  In applying the bioinformatics technique to the definition of the score function w, in this embodiment, several extensions are added. In the field of bioinformatics, the two series to be compared are usually the same, but in this embodiment, different objects of the daily action pattern registered in advance and the observed time series event data are handled. Therefore, the score function may be non-commutative. That is, w (x, y) ≠ w (y, x) may be satisfied.

  In this embodiment, the daily action patterns as shown in FIGS. 2 and 3 are compared with the time-series event data, so that the score function for the event pair is determined using the time information regarding the event. . For example, a score function is defined by the following likelihood.

ここで、ｘは時系列イベントデータs中のイベントを表し、ｙは日常行動パターンq中のイベントを表す。t_xbは時系列イベントデータｓ中のイベントxの開始時刻、t_xeはイベントxの終了時刻を表す。また、μ_ybは日常行動パターンqにおけるイベントyの（平均）開始時刻を、μ_yeは（平均）終了時刻を表す。イベントyの開始時刻の分散をσ_yb ² で表し、イベントyの終了時刻の分散をσ_ye ²とする。また、ギャップについてのスコア関数ｗ（ｘ，−）は、例えば所定の定数値をとるものとする。

Here, x represents an event in the time series event data s, and y represents an event in the daily behavior pattern q. t _xb represents the start time of the event x in the time series event data s, and t _xe represents the end time of the event x. Further, μ _yb represents the (average) start time of the event y in the daily behavior pattern q, and μ _ye represents the (average) end time. The variance of the start time of event y is represented by σ _yb ² , and the variance of the end time of event y is σ _ye ² . Further, the score function w (x, −) for the gap is assumed to take a predetermined constant value, for example.

  When the score function w is determined, the matching score of the optimal alignment can be obtained using dynamic programming. If the optimal alignment score for the subsequence s [1 ... i] of the time series event data and the subsequence q [1 ... j] of the daily behavior pattern is D [i, j], D [i, j].

この再帰式をD[i, j]の添字ｉ，ｊが小さいほうから順に、結果を記憶しておきながら計算する。このようにして得られるD[|s|, |q|]が時系列イベントデータsと日常行動パターンqとの最適アラインメントにおけるマッチングスコアに相当する。このため、ｉ，ｊのすべての組み合わせについてスコアD[i, j]を、O(|s||q|)の時間計算量、O(|s||q|)の空間計算量で計算することができる。

This recursive expression is calculated in order from the smallest subscript i, j of D [i, j] while storing the result. D [| s |, | q |] obtained in this way corresponds to the matching score in the optimal alignment between the time series event data s and the daily behavior pattern q. Therefore, the score D [i, j] is calculated for all combinations of i and j with the time complexity of O (| s || q |) and the space complexity of O (| s || q |). be able to.

The score calculation unit 31 of the processing unit 30 calculates a matching score for optimal alignment between all the patterns q included in the set Q and the time series event data s (matching score calculation process in FIG. 5).
The optimal alignment acquisition unit 32 acquires the optimal alignment for the daily action pattern q that gives the maximum matching score. The daily action pattern q giving the maximum matching score is the most similar pattern to the time series event data s. The optimum alignment can be calculated using the score D [i, j] used when calculating the matching score.

  That is, starting from the score D [| s |, | q |], D [i−1, j], D [i, j−1], D [i appearing in the max operation of Expression (2) therefrom. -1, j-1], the one with the maximum value is selected. Similarly, for the selected score, the maximum score that appears in the max operation is selected, and a path that reaches this point by repeating this operation to D [0, 0] gives the optimal alignment.

The behavior evaluation unit 33 evaluates how the event pattern q that is most similar to the time series event data s is different. The evaluation result is transmitted to the output unit 40.
The output unit 40 includes an evaluation result output unit 41. The evaluation result output unit 41 controls the output of the evaluation result by the behavior evaluation unit 33. When the output unit 40 includes a display device such as an LCD (not shown), the evaluation result output unit 41 may display the evaluation result on the display device. When the output unit 40 has an output terminal connected to an external device, the evaluation result output unit 41 may output the evaluation result to the external device via the output terminal. The evaluation result output unit 41 may output the evaluation result to an external device via a wired or wireless network.

Next, the operation of the extraordinary evaluation apparatus having the above configuration will be described.
FIG. 4 is a flowchart of the data series evaluation process performed in the processing unit 30.
When this process is started, the score calculation unit 31 acquires time-series event data s from the event data acquisition unit 11 (step S02). The score calculation unit 31 performs a matching score calculation process between the time-series event data s and all the daily action patterns q stored in the pattern storage unit 21 (step S03).

FIG. 5 is a flowchart showing details of the matching score calculation process. In this matching score calculation process, as described above, an optimal sequence alignment discovery method using dynamic programming is applied.
In the process of FIG. 5, the score calculation unit 31 reads any one pattern q from the set Q of daily activity patterns stored in the pattern storage unit 21 (step S22). The score calculation unit 31 calculates the matching score of the optimal alignment for the read pattern q and time series event data s using the dynamic programming described above (step S23).

  Then, it is determined whether or not the pattern q for which the matching score with the time series event data s has not been calculated exists in the set Q (step S24). If there is a pattern q for which a matching score is not calculated (Yes in step S24), the process returns to step S22.

  When the matching score is calculated for all patterns (No in step S24), the matching score calculation process is terminated, and the flow returns to the flow of FIG. Among all the patterns, for the pattern having the maximum matching score, the score and the calculated D [i, j] are stored.

  Returning to the flow of FIG. 4, whether the matching score is equal to or higher than a predetermined threshold for the daily action pattern q that maximizes the matching score, that is, the daily action pattern q that has the highest similarity to the time-series data s. Is determined (step S04). This threshold value may be a constant value or may be varied depending on the total number of events included in the daily behavior pattern and the time series event data.

  If the matching score is equal to or greater than the threshold (Yes in Step S04), the optimal alignment acquisition unit 32 acquires the optimal alignment that maximizes the score for the daily behavior pattern q that maximizes the matching score (Step S05). . As described above, the optimum alignment is obtained using the score D [i, j] used when calculating the matching score.

The behavior evaluation unit 33 evaluates the difference between the time-series input event data s and the daily behavior pattern q based on the optimum alignment obtained in Step S05 (Step S06).
For the evaluation of the time series input event data s, for example, the following criteria are given.
(A) If an event in the time series event data s corresponds to a gap (−), it is assumed that the event is an added event that does not appear in the daily behavior pattern q. That is, it means that the user took an action that is not included in the daily action. When the user's movement is evaluated based on the data as shown in FIGS. 2 and 3, this corresponds to the user detouring.
(B) If an event in the daily behavior pattern q corresponds to the gap (-), it is assumed that the event has not appeared in the time-series event data s. That is, it means that the user did not perform daily actions. When the user's movement is evaluated based on the data as shown in FIG. 2 and FIG. 3, this corresponds to the fact that the user does not normally visit the visited place.
(C) Even if the same events are associated with each other, when the time information shift between them is large, for example, when the likelihood calculated by the equation (1) is small, the time series event data s It is assumed that the event occurrence timing is different from the daily behavior pattern q. That is, it means that the user performed an event at a time different from that of daily life. When evaluating the movement of the user based on the data shown in FIGS. 2 and 3, this corresponds to the user's arrival time or departure time being earlier or later than usual.

FIG. 7 is a diagram showing an example of the optimal alignment of the time series event data s shown in FIG. 2 and the daily behavior pattern q shown in FIG.
In FIG. 7, the event data s [3] corresponds to a gap inserted between q [2] and q [3] of the daily behavior pattern q. This indicates that the user performed an action different from the daily action as in the criterion (a), that is, the user stopped by “ΔΔ station”.

  In FIG. 7, event data s [4] and q [3] correspond to each other. Event data q [3] indicates that the user typically arrives at “workplace” at 10:50 and departs at 19:00, while event data s [4] indicates that the user has 10 : 50 indicates that he arrived at "workplace" and departed at 17:00. That is, the user leaves the “workplace” two hours earlier than usual. Thus, even if the same event corresponds, it is understood that the user's action is earlier or later than usual when there is a difference in time information.

  The evaluation result by the behavior evaluation unit 33 is output to the evaluation result output unit 41 (step S07). The evaluation result by the behavior evaluation unit 33 shows that the input time series event data s is most similar to which daily action pattern q and how the time series event data s is compared with the daily action pattern q. It includes the degree of extraordinaryness that is different.

  On the other hand, when the maximum matching score is smaller than a predetermined threshold (No in step S04), the time series event data s is evaluated as extraordinary data different from any daily behavior pattern, and the evaluation result is evaluated. The result is output to the result output unit 41 (step S09).

When the evaluation result is output to the evaluation result output unit 41, the behavior evaluation process ends. The evaluation result output unit 41 displays the evaluation result on a display device such as an LCD or outputs the evaluation result to an external device. The user's behavior can be known from this output result.
As described above, according to the extraordinaryness evaluation apparatus according to the present embodiment, an event data series arranged in order of time when an event has occurred and a pre-stored action pattern are compared, and the generated event and the action pattern are compared. It can be determined how much is different.

  Specifically, the event data acquisition unit 11 acquires time series event data in which event data representing events generated by user behavior and event data representing the time information at which the event occurred are arranged in the order of event occurrence. A daily behavior pattern in which event data representing events that occur in the behavior and time information at which the events occur is arranged in the order in which the events occurred is stored in the pattern storage unit 21 in advance. The score calculation unit 31 compares the time-series event data with the daily action pattern sequence and calculates a matching score corresponding to the optimal alignment. At this time, the pairwise alignment method based on the dynamic programming method used for comparing two sequences in the field of bioinformatics is extended and applied. The similarity between the time series event data and the daily action pattern series is considered to be the highest in the optimal alignment. The optimal alignment acquisition unit 32 acquires the optimal alignment based on the matching score calculated by the score calculation unit 31. The behavior evaluation unit 33 evaluates unusualness from the difference between the elements of the time-series event data and the daily behavior pattern in the optimal alignment. Therefore, it is possible to evaluate which daily behavior pattern the time-series event data is similar to, or which daily behavior pattern is not similar to.

  Therefore, according to the extraordinary evaluation apparatus having the configuration of the above-described embodiment, it is possible to efficiently compare the observed action history with the previously stored pattern. Further, it is possible to detect extraordinary data having a large difference from a previously stored pattern and to evaluate how the difference differs. For example, it is possible to detect that an event has been added to the event data series or that an event has disappeared from the event data series, as compared with a previously stored pattern. In addition, it is possible to evaluate that the occurrence timing of a certain event is early or late as compared with a previously stored pattern.

  In the present embodiment, the event data acquisition unit 11 uses a GPS device to acquire a user's movement history to obtain time series event data, but the present invention is not limited to this. In a factory or the like, the detection history by the sensor, the operation history of the apparatus, the progress of the process, and the like may be compared with a predetermined daily pattern.

  In the above description, the pattern storage unit 21 stores in advance a set Q of one or a plurality of action patterns that are routinely performed by the user. The daily action pattern may be, for example, a day of the action normally performed by the user, a monthly pattern, or the like. Further, when a plurality of users use this apparatus, a daily behavior pattern may be stored for each user. When a plurality of users use this apparatus, an event data acquisition unit 11 that can be detached from the apparatus may be provided for each user, and each user carries a respective event data acquisition unit 11 to obtain time-series event data. get.

According to the extraordinary evaluation apparatus according to the present embodiment, it becomes possible to manage and supervise the user's behavior in detail. For example, an elderly person or a child can carry this apparatus, so that the family can supervise the behavior.
Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some configurations may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different example embodiments may be combined as appropriate.

  The present invention can also be implemented to cause a computer to execute predetermined means, to cause a computer to function as predetermined means, to cause a computer to realize predetermined functions, or as a computer-readable recording medium storing a program. You can also

  DESCRIPTION OF SYMBOLS 10 ... Input part, 11 ... Event data acquisition part, 20 ... Memory | storage part, 21 ... Pattern storage part, 30 ... Processing part, 31 ... Score calculation part, 32 ... Optimal alignment acquisition part, 33 ... Action evaluation part, 40 ... Output Part, 41 ... evaluation result output part.

Claims (7)

A data acquisition means for acquiring an event data series in which event data representing an event generated by a user's behavior and time information at which the event occurs are arranged in the order of event occurrence;
Pattern storage means for storing a reference pattern sequence in which event data representing events that occur in the daily activities of the user and time information at which the events occur is arranged in the order of event occurrence;
A score calculation means for comparing the event data series with the reference pattern series, and calculating a matching score corresponding to the optimum alignment with the highest similarity between the two series based on dynamic programming;
Optimal alignment acquisition means for acquiring the optimal alignment based on the matching score calculated by the score calculation means;
In the optimum alignment, an evaluation means for evaluating unusualness from a difference between events included in the event data series and the reference pattern series,
An extraordinary evaluation device. For a pair of the event data series event and the reference pattern series event, a score function is defined based on time information corresponding to each event,
The extraordinaryness evaluation apparatus according to claim 1, wherein an alignment that maximizes a sum of score functions for all pairs is determined as an optimal alignment. The non-daily life evaluation apparatus according to claim 1, wherein the event data series includes movement history data of the user. Store a reference pattern sequence in which event data representing events that occur in daily actions of users and time information at which the events occur are arranged in the order of event occurrence,
Obtain an event data series in which event data representing events generated by the user's behavior and time information when the event occurred are arranged in the order of event occurrence,
The event data series and the reference pattern series are compared, and a matching score corresponding to the optimal alignment with the highest similarity between the two series is calculated based on dynamic programming,
Obtaining the optimal alignment based on the calculated matching score;
An extraordinary evaluation method for evaluating extraordinaryness from a difference between events included in the event data series and the reference pattern series in the optimal alignment. For a pair of the event data series event and the reference pattern series event, a score function is defined based on time information corresponding to each event,
The extraordinaryness evaluation method according to claim 4, wherein an alignment that maximizes a sum of score functions for all pairs is defined as an optimal alignment. The non-daily life evaluation method according to claim 4, wherein the event data series includes movement history data of the user. A computer program for causing a computer to perform an evaluation of unusualness in user behavior,
A data acquisition step of acquiring an event data series in which event data representing the event generated by the user's behavior and event data representing the time information at which the event occurred are arranged in the order of event occurrence;
A pattern storage step for storing a reference pattern sequence in which event data representing events that occur in the daily behavior of the user and event data representing the time information at which the event occurs is arranged in the order of event occurrence;
A score calculation step of comparing the event data series with the reference pattern series, and calculating a matching score corresponding to an optimal alignment with the highest similarity between the two series based on dynamic programming;
An optimal alignment acquisition step of acquiring the optimal alignment based on the matching score calculated in the score calculation step;
An evaluation step for evaluating a difference between the events included in the event data series and the reference pattern series in the optimum alignment;
A computer program comprising:

Images