JP2019087179A - Analyzer, analysis method and program - Google Patents

Abstract

To improve the accuracy of behavior analysis.SOLUTION: An analyzer 1 includes a control unit 1a and a storage unit 1b. The control unit 1a generates feature information in time series that represents a behavior state per unit time from behavior information derived by sensing the behavior of a user. When non-continuous feature information exists in which feature information differs before and after the unit time, the control unit 1a classifies the behavior information included in the non-continuous feature information into first behavior information and second behavior information differing in nature from each other. Then, the control unit 1a determines whether the continuity of feature information is appropriate from the increment/decrement tendencies of both the first behavior information and the second behavior information, and corrects the continuity of the feature information on the basis of the result of determination. The storage unit 1b stores the behavior information and feature information, and further stores other control information associated with the operation of the analyzer 1.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an analyzer, an analysis method, and a program.

BACKGROUND In recent years, behavior support systems that analyze user behavior from information acquired by sensing and provide various services according to behavior states have attracted attention.
Examples of sensing information used for user behavior analysis include motion data such as the number of steps and acceleration, and vital data such as heart beat and pulse, which can be acquired through a portable terminal or a wearable terminal.

Also, there are position information that can be acquired through an access point of a wireless local area network (LAN), scheduling information that can be acquired through a cloud network, and the like.
In the action support system, the user's action state (stand, walk, sit, talk, etc.) is analyzed using these sensing information, and services are provided individually for the user according to the analyzed action state.

  For example, it is possible to provide a service close to the user in various fields, such as guidance of a moving route, presentation of business information in an office environment, medical assistance in a hospital, recommendation of risk behavior in construction and maintenance sites.

  As a prior art, for example, a technique has been proposed for recognizing the behavior of a subject from the feature amount in the observation result and the feature amount stored. In addition, there has been proposed a technique for correcting the sequence of position information included in the action log based on the reference position. Furthermore, a technique has been proposed that corrects the result of the action estimation so that the action at the correction unit time that is longer than the basic unit time of the measurement value of the acceleration sensor continues.

JP 10-113343 A International Publication No. 2015/194269 Patent No. 5382719 gazette

  However, if the acquisition of sensing information is disturbed due to noise or the like, there is a possibility that the user's action state is erroneously recognized as being different from the original action state. For this reason, it is required to detect the erroneously recognized behavior state and correct it to the original behavior state.

  In one aspect, the present invention aims to provide an analysis device, an analysis method, and a program in which the accuracy of behavior analysis is improved.

  In order to solve the above-mentioned subject, an analyzer is provided. The analyzer includes a controller. The control unit generates feature information representing an action state for each unit time in time series from the action information in which the user's action is sensed, and the feature information before the unit time is discontinuous for non-continuous feature information If there is feature information, the action information included in the non-consecutive feature information is classified into first action information and second action information having different properties, and both of the first action information and the second action information The propriety of the continuity of the feature information is determined from the increase / decrease tendency, and the continuity of the feature information is corrected based on the determination result.

Moreover, in order to solve the said subject, the analysis method in which a computer performs control similar to the said analyzer is provided.
Furthermore, in order to solve the above-mentioned subject, a program which makes a computer perform control similar to the above-mentioned analysis device is provided.

  According to one aspect, it is possible to improve the accuracy of behavioral analysis.

It is a figure which shows an example of a structure of an analyzer. It is a figure showing an example of the composition of an information processing system. It is a figure showing an example of a model of action support. It is a figure which shows an example of the hardware constitutions of a server. It is a figure which shows an example of the functional block of a server. It is a figure which shows an example of production | generation operation | movement of a feature-value. It is a figure which shows an example of grouping of a feature-value. It is a figure which shows an example of labeling of a feature-value. It is a figure which shows the state in which grouping was performed based on the granularity of a log. It is a figure which shows the state in which grouping was performed based on the granularity of a log. It is a figure for demonstrating the operation | movement of correction of a feature-value. It is a figure for demonstrating the operation | movement of correction of a feature-value. It is a figure which shows an example of a noise determination table. It is a figure which shows the state after feature-value correction | amendment. It is a flowchart which shows operation | movement of a server.

Hereinafter, the present embodiment will be described with reference to the drawings.
First Embodiment
The analyzer according to the first embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of the configuration of an analyzer. The analyzer 1 includes a control unit 1a and a storage unit 1b. The control unit 1a generates feature information representing an action state for each unit time in time series from the action information in which the action of the user is sensed.

  In addition, when there is non-consecutive non-consecutive feature information in which the subsequent feature information is non-consecutive with respect to the feature information before the unit time, the control unit 1a is different from the first in the behavior information Classified into action information and second action information.

  Then, the control unit 1a determines the appropriateness of the continuity of the feature information from the increase / decrease tendency of both the first action information and the second action information, and corrects the continuity of the feature information based on the determination result. . The storage unit 1 b stores sensed behavior information and feature information, and further stores other control information related to the operation of the analyzer 1.

The operation will be described using the example shown in FIG.
[Step S1] The control unit 1a receives action information (corresponding to sensing information). The action information is information generated by the action of the user, which is sensed by a sensor device such as a mobile terminal. In the example of FIG. 1, “heart rate”, “step count”, “position information”, and “e-mail transmission / reception” are shown. The behavior information is periodically transmitted from the sensor device to the analysis device 1 and received by the control unit 1a.

  [Step S2] When receiving the action information, the control unit 1a generates feature information from the action information. In the example of FIG. 1, the feature information A is generated at time t0, t1, and t3, the feature information B is generated at time t2, and the feature information C is generated at time t4.

  In addition, the control unit 1a detects non-continuous feature information whose feature information differs before and after a unit time. In the example of FIG. 1, a set α of feature information A at time t1 and feature information B at time t2 and a set β of feature information A at time t3 and feature information C at time t4 are detected as non-continuous feature information. ing.

  [Step S3] The control unit 1a classifies the action information included in the non-continuous feature information into first action information and second action information having mutually different properties. In the example of FIG. 1, the action information of the feature information A is classified into a first action information d1a and a second action information d2a. Further, the action information of the feature information B is classified into the first action information d1 b and the second action information d2 b, and the action information of the feature information C is classified into the first action information d1 c and the second action information d2 c .

Further, the control unit 1a determines the appropriateness of the continuity of the feature information from the increase and decrease tendency of both the first action information and the second action information.
In this case, the control unit 1a changes the first behavior information d1a to the first behavior information d1b in the transition from the feature information A to the feature information B, and the second behavior information d2a to the second behavior. A first increase / decrease tendency is determined by the combination with the increase / decrease tendency to the information d2b. Then, the control unit 1a determines whether the continuity of the feature information A and B in the set α is appropriate or not based on the first increase / decrease tendency.

  In addition, the control unit 1a controls the increase / decrease tendency from the first action information d1a to the first action information d1c in transition of the feature information A to the feature information C, and the second action information d2c to the second action information d2c. A second increase / decrease tendency is determined in combination with the increase / decrease tendency. Then, the control unit 1a determines the appropriateness of the continuity of the feature information A and C in the set β based on the second increase / decrease tendency.

[Step S4] The controller 1a corrects the continuity of the feature information based on the determination result.
[Step S4a] It is assumed that the controller 1a determines that the continuity of the feature information is inappropriate from the first increase / decrease tendency. In this case, the control unit 1a replaces the feature information B with the feature information A and converts the non-continuous feature information into continuous feature information (arranges the feature information A and the feature information B into the feature information A and the feature information A).

  [Step S4b] It is assumed that the controller 1a determines that the continuity of the feature information is appropriate from the second increase / decrease tendency. In this case, the control unit 1a does not replace the feature information C with the feature information A, but maintains the state of the non-continuous feature information (leaves the feature information A and the feature information C in the same order).

  As described above, the analysis device 1 generates feature information representing the action state from the sensed user action information, and determines the appropriateness of the continuity of the feature information from the increase / decrease tendency of the action information classified according to the property, The continuity of feature information is corrected based on the determination result.

  Thereby, even if the analysis device 1 erroneously recognizes the action state of the user as something different from the original action state due to disturbance due to noise or the like, the analysis device 1 detects the erroneously recognized action state and corrects it to the original action state. Be able to improve the accuracy of behavioral analysis.

Second Embodiment
Next, a second embodiment will be described. First, the system configuration will be described. FIG. 2 is a diagram showing an example of the configuration of the information processing system. The information processing system 1-1 includes a server 10, a network 2, a radio base station 3, and a mobile terminal 4.

The server 10 connects to the network 2. The radio base station 3 is, for example, an access point, and connects to the network 2 and the mobile terminal 4.
The mobile terminal 4 transmits, to the server 10, behavior information obtained by sensing a user's behavior. In this case, the mobile terminal 4 may transmit based on a polling instruction from the server 10 or may transmit at a preset cycle. In addition, if it is an apparatus which can sense a user's action, things other than the mobile terminal 4 may be used. The device is not limited to wireless communication, and may be connected to the server 10 via the network 2 by wired communication.

  The server 10 has the function of the analysis device 1 shown in FIG. 1, analyzes the behavior of the user based on the acquired behavior information, and performs behavior support for each user. In addition, it is also possible to set it as the structure provided with the function of the analyzer 1 shown in FIG. 1 with respect to the mobile terminal 4 which is a sensor apparatus.

<Model for Behavior Support>
FIG. 3 is a diagram showing an example of an action support model. In the following description, behavior information may be referred to as a log.

  [Step S11] The server 10 acquires logs from each user. The server 10 acquires, for example, a log related to the behavior of the digital space. The log related to the behavior of the digital space corresponds to, for example, information on the application or network usage status.

  In addition, the server 10 acquires, for example, a log related to an action of the physical space. The log related to the action of the physical space corresponds to, for example, information on the whereabouts of the user and the operation state (standing, sitting, etc.).

  [Step S12] The server 10 labels the user's behavior and the environment in which the user is located. Specifically, the server 10 converts the acquired log into an action state, groups a plurality of action states, and assigns a label to each group (group unit labeling).

[Step S13] The server 10 generates a behavior model of the user based on the group to which the label is attached.
[Step S14] The server 10 stores the generated behavior model.

  [Step S15] The server 10 estimates the next action based on the current action model of the user. Note that the estimated next action is fed back to the process of step S13 to generate a new action model. Then, based on the generated new behavior model, the next behavior is estimated by the process of step S15.

[Step S16] The server 10 performs action support for instructing each user to perform an action (guidance) based on the action estimation (personalization of the action estimation) of each user.
<Hardware configuration>
FIG. 4 is a diagram showing an example of the hardware configuration of the server. The server 10 is controlled by the processor 100 as a whole. That is, the processor 100 functions as a control unit of the server 10.

  A memory 101 and a plurality of peripheral devices are connected to the processor 100 via a bus 103. The processor 100 may be a multiprocessor. The processor 100 is, for example, a central processing unit (CPU), a micro processing unit (MPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or a programmable logic device (PLD). The processor 100 may also be a combination of two or more elements of a CPU, an MPU, a DSP, an ASIC, and a PLD.

  The memory 101 is used as a main storage device of the server 10. The memory 101 temporarily stores at least part of an OS (Operating System) program and application programs to be executed by the processor 100. The memory 101 also stores various data required for processing by the processor 100.

  The memory 101 is also used as an auxiliary storage device of the server 10, and stores an OS program, an application program, and various data. The memory 101 may include, as an auxiliary storage device, a semiconductor storage device such as a flash memory or a solid state drive (SSD), or a magnetic recording medium such as a hard disk drive (HDD).

  Peripheral devices connected to the bus 103 include an input / output interface 102 and a network interface 104. The input / output interface 102 is connected with a monitor (for example, an LED (Light Emitting Diode), an LCD (Liquid Crystal Display), etc.) functioning as a display device for displaying the state of the server 10 according to an instruction from the processor 100.

The input / output interface 102 is connectable to an information input device such as a keyboard and a mouse, and transmits a signal sent from the information input device to the processor 100.
Furthermore, the input / output interface 102 also functions as a communication interface for connecting peripheral devices. For example, the input / output interface 102 can connect an optical drive device that reads data recorded on an optical disk using a laser beam or the like. Examples of the optical disc include Blu-ray Disc (registered trademark), CD-ROM (Compact Disc Read Only Memory), CD-R (Recordable) / RW (Rewritable), and the like.

  In addition, the input / output interface 102 can connect a memory device or a memory reader / writer. The memory device is a recording medium having a communication function with the input / output interface 102. The memory reader / writer is a device that writes data to the memory card or reads data from the memory card. The memory card is a card type recording medium.

  The network interface 104 performs interface control with an external network, and can use, for example, a network interface card (NIC), a wireless LAN card, or the like. The data received by the network interface 104 is output to the memory 101 or the processor 100.

  By the hardware configuration as described above, the processing function of the server 10 can be realized. For example, the server 10 can perform control of the present invention as the processor 100 executes a predetermined program.

  The server 10 implements the processing function of the present invention, for example, by executing a program recorded on a computer readable recording medium. The program describing the processing content to be executed by the server 10 can be recorded in various recording media.

  For example, a program to be executed by the server 10 can be stored in the auxiliary storage device. The processor 100 loads at least a part of the program in the auxiliary storage into the main storage and executes the program.

  Moreover, it can also record on portable recording media, such as an optical disk, a memory apparatus, and a memory card. The program stored in the portable recording medium can be executed after being installed in the auxiliary storage device under the control of the processor 100, for example. The processor 100 can also read and execute the program directly from the portable recording medium.

<Function block>
FIG. 5 is a diagram showing an example of a functional block of the server. The server 10 includes a behavior analysis unit 10a and a data management unit 10b. The behavior analysis unit 10a implements the function of the control unit 1a of FIG. 1, and the data management unit 10b implements the function of the storage unit 1b of FIG. In the following description, feature information is called a feature amount.

The behavior analysis unit 10 a includes a communication interface unit 11, a feature amount generation unit 12, a feature amount correction unit 13, a feature amount group processing unit 14, and an action support processing unit 15.
The communication interface unit 11 performs communication interface control with the mobile terminal 4 via the network 2. The feature amount generation unit 12 receives the log sensed and transmitted by the mobile terminal 4 and converts the received log into a feature amount.

  When the feature amount correction unit 13 detects the presence of non-continuous feature amounts whose feature amounts are different before and after time, logs included in the non-continuous feature amount are used as the first log and the second log whose properties are different from each other. Classify. Then, the feature amount correction unit 13 determines the appropriateness of the continuity of the feature amount from the increase and decrease tendency of both the first log and the second log, and corrects the continuity of the feature amount based on the determination result. .

The feature amount group processing unit 14 adds a label to a plurality of feature amounts whose appearance frequency is high (above the threshold) in a predetermined time zone, and groups the feature amounts.
The action support processing unit 15 performs action estimation from the grouped feature amounts, and outputs service information for performing action support to the user based on the action estimation.

  The data management unit 10 b stores and manages data required for the operation of the behavior analysis unit 10 a. Further, the data management unit 10b stores and manages table information of a feature amount management table Tb1, a feature amount correction table Tb2 and a noise determination table Tb3 described later.

The behavior analysis unit 10a is realized by the processor 100 illustrated in FIG. Also, the data management unit 10 b is realized by the memory 101 shown in FIG. 4.
<Generation of feature quantities>
FIG. 6 is a diagram showing an example of the feature amount generation operation. The feature amount is a primitive action state of the user and is also called a micro activity. For example, features such as still (still), walking (walking), and speaking (during conversation) are such feature amounts. A plurality of feature quantities can appear simultaneously.

  [Step S21] The feature amount generation unit 12 performs reception processing of the log transmitted from the mobile terminal 4. The log includes, for example, a sensor type, a sensing time, and a sensing result as attributes.

  For example, when the mobile terminal 4 senses the user position by the GPS (Global Positioning System) function, as the attribute contents, the sensor type is GPS, the sensing time is the time when the user position is measured, and the sensing result is latitude and longitude . In the log reception process, such attribute detection or the like is performed.

  [Step S22] The feature quantity generation unit 12 converts the log into feature quantities based on a predetermined condition (this conversion process is performed periodically). For example, when the condition that all the plurality of predetermined parameter values of the log a1 exceed the threshold value is satisfied in the past fixed period, the log a1 is converted to the feature amount A1.

  As a specific example, if the sensor type attribute of the log a1 is a microphone, the sensing result attribute is audio data, and the value of the audio data exceeds a threshold in a fixed period in the past, the log a1 is used as the feature amount A1 Converted to "speaking".

  [Step S23] The feature amount generation unit 12 registers and manages the history of feature amounts generated in the conversion process of step S22 in the feature amount management table Tb1. The feature amount management table Tb1 has time and feature amount as items. The time is a time when the mobile terminal 4 senses the log before being converted into the feature value.

  For example, in the first row of the feature amount management table Tb1, the sensing time is 2017-10-12 13:58:20, and the logs sensed at this time have the features [speaking], [strain] and It shows that it is converted to [walking].

<Grouping of feature quantities>
The feature amount group processing unit 14 groups the feature amounts whose appearance frequency is high (the appearance frequency exceeds the threshold) in the same time zone, and collects a plurality of feature amounts into one group. When grouping feature amounts, the feature amount group processing unit 14 calculates the co-occurrence frequency of the feature amounts, absorbs fluctuation (error) at a predetermined time (for example, 5 minutes to 10 minutes), and performs grouping.

Further, the feature amount group processing unit 14 corrects the number of appearances of the feature amount according to the following equation (1). Output correction is performed when O (pt _n ) in equation (1) is equal to or greater than the threshold. Note that CT is the current time, pT is the last action occurrence time, and C (pt) is the time when it is temporarily interrupted.

O (pt _n ) = C (pt _n ) / (pT _n −CT) (1)
Then, the feature amount group processing unit 14 calculates the similarity S (PS _t , PS _{t ′} ) by performing similarity calculation according to the following equation (2) using the Jaccard coefficient (0 ≦ S (PS _t , PS _{t '} ) <= 1). Here, pt is a feature amount, PS is a group (a set of feature amounts), n (G) is the number of elements of the group, and E () is an average.

S (PS _t , PS _{t ′} ) = {Σmin (pt _i , pt ′ _i ) / Σmax (pt _i , pt ′ _i )} × {E (n (G)) / E (n (PS _t ), n (PS _{t '} ))} (2)
For example, the feature amounts of the first and second lines of the feature amount management table Tb1 shown in FIG. 6 are the same. In this case, S (PS _t , PS _{t ′} ) = 1. When S (PS _t , PS _{t ′} ) is 1, the feature quantities to be compared are perfect match, and the similarity becomes larger as 1 is closer to 1. In addition, when S (PS _t , PS _{t ′} ) is 0, the feature quantities to be compared are completely different, and the similarity becomes smaller as it approaches 0. Therefore, the feature amount group processing unit 14 compares the value calculated by the equation (2) with a preset threshold, determines whether or not to make the same group, and groups those exceeding the threshold.

FIG. 7 is a diagram showing an example of grouping of feature amounts. The horizontal axis is time. Further, a specific example of the feature amount is given on the vertical axis.
[Time t1] When the feature amount group processing unit 14 detects that the occurrence frequency of [sitting] and [fj_kawasaki_lab] among the feature amounts shown on the vertical axis at time t1 exceeds the threshold value, it means Make [fj_kawasaki_lab] one group g1.

  [Time t2] If the feature amount group processing unit 14 detects that the frequency of appearance of [walking] and [strain] exceeds the threshold at time t2, among the feature amounts shown on the vertical axis, [walking] Make [strain] one group g2.

  [Time t3] When the feature amount group processing unit 14 detects that the occurrence frequency of [sitting] and [meeting] of the feature amounts shown on the vertical axis at time t3 exceeds the threshold value, it means Make [meeting] one group g3.

  [Time t4] At time t4, if the feature quantity group processing unit 14 detects that the occurrence frequency of [speaking], [meeting] and [strain] exceeds the threshold among the feature quantities shown on the vertical axis, Make [speaking], [meeting] and [strain] one group g4.

  [Time t5] If the feature amount group processing unit 14 detects that the frequency of appearance of [walking] and [office] exceeds the threshold among the feature amounts shown on the vertical axis at time t5, [walking] Make [office] one group g5.

  [Time t6] When the feature amount group processing unit 14 detects that the occurrence frequency of [walking], [no_wifi] and [normal] among the feature amounts shown on the vertical axis at time t6 exceeds a threshold value, Make [walking], [no_wifi] and [normal] one group g6.

<Labeling of feature quantities>
A plurality of feature quantities (micro activities) are grouped, and those to which a label is attached are called macro activities (Macro Activity). For example, during a meeting (still, speaking), writing materials (writing).

  The macro activity represents a more specific action of the user, and is used as a feature for recognizing the action. Also, the names “in meeting” and “in preparation of material” correspond to labels.

  FIG. 8 is a diagram showing an example of feature amount labeling. It shows a state in which a label is given to the group shown in FIG. The group g1 is labeled "Office", and the group g2 is labeled "moving".

  The label "in meeting" is given to the group g3, and the label "in material explaining at the meeting" is given to the group g4. The label "moving" is given to the group g5, and the label "moving to the place P" is given to the group g6.

  The feature amount group processing unit 14 associates the feature amount group with the user's execution motion by performing the above-described labeling process. Although the label is held in advance by the server 10, it is also possible that the feature amount group processing unit 14 requests the user for feedback, and a label arbitrarily set by the user can be added.

<Log granularity and analysis accuracy>
As described above, in the system that provides behavior support, logs are periodically collected from the mobile terminal 4 and the collected logs are analyzed to analyze the behavior of the user who is using the mobile terminal 4.

  However, since the mobile terminal 4 is battery-powered, it is required to suppress power consumption. Therefore, it is desirable that the frequency of log acquisition in the mobile terminal 4 be reduced. Therefore, the power consumption of the battery can be suppressed by increasing the sampling period of the log. However, increasing the log sampling period may degrade the accuracy of behavioral analysis.

  In the following description, the sampling period of the log may be referred to as the granularity. In addition, increasing the sampling period may be expressed as coarsening the particle size, and decreasing the sampling period may be expressed as fineness of the particle size.

  FIGS. 9 and 10 are diagrams showing grouping performed based on log granularity. It is assumed that the feature amount A in FIG. 9 and FIG. 10 is labeled as “meeting”, the feature amount B as “business trip”, and the feature amount C as intra-office movement.

FIG. 9 shows a grouping state based on a log of fine granularity.
[State St1] The feature amounts A, B, C converted from the fine grained log are shown. Times t1, t2, t3, t5, t6, t7, t11 and t12 are feature amounts A, time t4 is feature amounts B, and times t8, t9 and t10 are feature amounts C. Note that the feature amount at time t4 is temporarily disturbed in log acquisition due to noise or the like, and should originally be the feature amount A, but the feature amount B is assumed to be erroneously appearing.

  [State St2] The feature quantities A and B are included in the feature quantities from time t1 to time t7, but the number of feature quantities B is small, so the feature quantity B is considered to be erroneously appearing. In this case, the noise removal processing functions to delete the feature amount B, and the feature amounts from time t1 to time t7 are grouped as a group of the feature amount A.

[State St3] Since only the feature amount C appears from time t8 to time t10, the feature amounts from time t8 to time t10 are grouped as a group of feature amounts C.
[State St4] Since only feature amount A appears at time t11 and t12, the feature amount at time t11 and t12 is grouped as a group of feature amount A.

FIG. 10 shows the grouping state based on the coarse granularity log.
[State St11] The feature quantities A, B, and C converted from the coarse grain size log are shown. Times T1, T2 and T4 are feature amounts A, time T3 is feature amounts B, and time T5 is feature amounts C. It is assumed that the feature quantities recognized at each time do not appear erroneously due to noise, and all appear normally.

  [State St12] There are feature amounts A and B as feature amounts from time T1 to time T4. In this case, the noise removal processing performed on the fine-grained log as shown in the above-mentioned state St2 functions, and the feature amount B is considered to be erroneously appearing because the number is small. Therefore, the feature amounts from time T1 to time T4 are grouped as a group of feature amounts A.

  As described above, in the case of a fine-grained log, even if a temporary false occurrence occurs, the noise removal processing functions so that the feature quantity appearing relatively infrequently is regarded as the falsely appearing feature quantity, It is excluded at the time of grouping. However, in the case of a coarse-grained log, the feature amount detected as a normal appearance may be regarded as an erroneous appearance by the noise removal processing.

  In the example of FIG. 10, the characteristic quantity B is supposed to be originally at time T3, but since the number of characteristic quantities A is larger before and after time T3, the characteristic quantity B at time T3 is regarded as an erroneous appearance due to noise. Then, grouping is performed as feature amount A from time T1 to time T4, the break of the user's action state is not accurately detected, and the action analysis accuracy is degraded.

As described above, when a coarse-grained log is used, the other feature quantities that are erroneously recognized are mixed in the generated feature quantities of the group, and the accuracy of the behavior analysis may be degraded.
The present invention has been made in view of such a point, and aims to improve the accuracy of behavior analysis by accurately detecting the delimitation of the feature amount even when the log is acquired with the coarse granularity.

<Classification of log>
The server 10 according to the present invention classifies the log into two types: human dependent log (human dependent behavior information) and environment dependent log (environment dependent behavior information). The human dependency log can be defined as a log whose value changes in accordance with an operation actively performed by the user. The human dependency log defined in this way includes, for example, the number of steps, heart rate, gyro, acceleration, display on / off and the like. Furthermore, application and network usage (mail transmission / reception, addition of schedule, etc.) are also included in this classification.

  The environment dependent log can be defined as a log having a property in which the value changes without the user's awareness in the environment around the user. The environment-dependent log defined in this way corresponds to, for example, identifier information of the access point when the user enters the wireless area of the access point, or information received by broadcast communication.

  In addition to the above definitions, human dependent logs and environment dependent logs can be classified according to the following definitions. Depending on the way of definition, the log included in the human dependency log may be included in the environment dependent log, or the log included in the environment dependent log may be included in the human dependency log.

In the above, the human / environmental dependency is separated depending on whether or not to make a change according to the movement of the user, but the following criteria may be used.
A) Judgment based on whether or not there is a sensor change based on a result actively input by the user.

  In the case of a sensor with active input, it is relatively likely to be a change in behavior (no noise). On the other hand, the possibility that the input / changed information is a change in behavior (no noise) is relatively low (it is likely to be noise).

  As judged in the example A), the human dependency log includes display on / off, mail transmission, schedule, to-do input, and the like. The environment dependent log includes the number of steps, heart rate, gyro, acceleration, access point information, mail reception, broadcast communication and the like.

B) Judgment based on whether the frequency of the output of the value from the sensor is intermittent.
When the user acts, the values output from the sensor tend to be intermittent. Therefore, human / environment-dependent separation is possible depending on whether or not there is an intermittent output. That is, logs with intermittent output are human dependent, and periodic output logs are environmental dependent.

  When judged in the example of B), the human dependency log includes the number of steps, gyro, acceleration, display on / off, mail transmission, schedule, to-do input, and the like. The environment-dependent log includes heart rate, mail reception, access point information, schedule information reception, broadcast communication, and the like.

C) Judgment based on whether it is a sensor device intended for an individual.
There are sensor devices that are targeted for individual movements and those targeted for movements of multiple users. This property can be used to make human / environment dependent distinctions.

  When judged in the example of C), the human dependency log includes display on / off, step number, heart rate, gyro, acceleration, e-mail transmission, schedule and to-do input, access point information, broadcast communication and the like. Further, the environment dependent log includes a fixed arrangement human sensor, indoor temperature / humidity sensor, and the like.

<Correction of feature amount>
Next, the operation of the feature amount correction unit 13 will be described using FIG. 11 to FIG. 11 and 12 are diagrams for explaining the operation of correcting the feature amount. First, the feature amount generation unit 12 outputs the feature amount A at times t0 and t1, the feature amount B at time t2, the feature amount A at time t3, and the feature amount C at time t4.

  The feature quantity correction unit 13 receives the feature quantity output from the feature quantity generation unit 12 and detects a set of feature quantities that differ before and after time. In this example, it is assumed that the set α of the feature A at time t1 and the feature B at time t2 and the set β of the feature A at time t3 and the feature C at time t4 are detected.

  Also, the feature amount A includes human dependent logs h1 and h2, and environment dependent logs e1 to e10. Also, the feature amount B includes human dependent logs h1 and h2, and environment dependent logs e1 and e2. Furthermore, it is assumed that the feature amount C includes a human dependency log h3 and an environment dependency log e11 to e15.

  The feature amount correction tables Tb2-1 and Tb2-2 are tables generated by the feature amount correction unit 13. The feature amount correction tables Tb2-1 and Tb2-2 are items of the feature amount of the previous time, the feature amount of the later time, the number of log changes, the amount of change, the increase / decrease result, and the determination result for each of the human dependency log and the environment dependency log Have.

Note that the minus sign numerical value on the left side in the log change number box indicates the reduction number of the log, and the plus sign numerical value on the right side indicates the increase number of the log.
The feature quantity count at the previous time, the feature quantity count at the later time, and the log change count in the feature quantity correction table Tb2-1 corresponding to the set α in FIG. 11 will be described. The feature amount at the previous time is the number of human / environment dependent logs included in the feature amount A at time t1. Since the feature amount A includes two human dependent logs h1 and h2 and ten environment dependent logs e1 to e10, {characteristics of previous time: (number of human dependent logs, environment dependent The number of logs)} = (2, 10).

  The feature value of the later time is the number of human / environment dependent logs included in the feature value B at time t2. Since the feature amount B includes two human dependent logs h1 and h2 and two environment dependent logs e1 and e2, {feature amount of later time: (number of human dependent logs, number of environment dependent logs)} = (2, 2).

  Here, the human dependency log includes the logs h1 and h2 for both of the feature amounts A and B, and there is no decrease in log and no increase in new log in the transition from the feature amount A to the feature amount B. Therefore, since there is no change in the human dependency log, the number of log changes = (− 0, +0) is registered.

  In the environment-dependent log, since the feature amount A includes the logs e1 to e10 and the feature amount B includes the logs e1 and e2, in the transition from the feature amount A to the feature amount B, the log e3.・ ・ ・, Eight of e10 decrease. There is no new log increase. Therefore, the environment dependent log is registered as the number of log changes = (− 8, +0) because eight logs decrease and there is no new log increase.

  The feature quantity of the previous time, the feature quantity of the later time, and the number of log changes in the feature quantity correction table Tb2-2 corresponding to the set β of FIG. 12 will be described. The feature amount at the previous time is the number of human / environment dependent logs included in the feature amount A at time t3. Similarly to the above, {feature amount of previous time: (number of human dependent logs, number of environment dependent logs)} = (2, 10).

  The feature value of the later time is the number of human / environment dependent logs included in the feature value C at time t4. Since the feature amount C includes one human dependent log h3 and five environment dependent logs e11,..., E15, {characteristics of later time: (number of human dependent logs, number of environment dependent logs )} = (1, 5).

  Here, in the human dependent log, the feature amount A includes the logs h1 and h2, and the feature amount C includes the log h3, so in the transition from the feature amount A to the feature amount C, the two logs h1 and h2 decrease And one new log h3 is increasing. Therefore, the human dependency log is registered as the number of log changes = (− 2, +1) because two logs decrease and one new log increases.

  Further, the environment dependent log includes feature amounts A including logs e1 to e10, and a feature amount C including logs e11 to e15. Therefore, in the transition from the feature amount A to the feature amount C, ten of the logs e1, ..., e10 decrease, and five new logs e11, ..., e15 increase. Therefore, the environment-dependent log is registered as the number of log changes = (− 10, +5) because ten logs are decreased and five new logs are increased.

Next, the amount of change, the increase / decrease result, and the determination result in the feature amount correction tables Tb2-1 and Tb2-2 will be described. The amount of change is calculated by the feature amount correction unit 13 using equation (3).
(Amount of change) = (number of decreases in log) + (number of increases in log) × k (3)
Here, k (> 1) is a preset coefficient. For example, k = 10.

  In the feature amount correction table Tb2-1, the amount of change corresponding to the human dependency log is 0 (= 0 + 0 × 10), so the increase / decrease result is registered as no change. The change amount corresponding to the environment dependent log is −8 (= −8 + 0 × 10), so the increase / decrease result is registered as a decrease.

  Therefore, (the increase / decrease result of the human dependent log, the increase / decrease result of the environment dependent log) = (no change, decrease), the determination result is registered as noise based on the action scene described later. ). That is, the feature amount B is considered to be erroneously appearing due to disturbance such as noise.

  In the feature amount correction table Tb 2-2, the amount of change corresponding to the human dependency log is 8 (= −2 + 1 × 10), so the increase / decrease result is registered as an increase. Since the amount of change corresponding to the environment dependent log is 40 (= −10 + 5 × 10), the increase / decrease result is registered as an increase. Therefore, (the increase / decrease result of the human dependent log, the increase / decrease result of the environment dependent log) = (increase, increase), the determination result is registered as no noise (the description of the determination result will be described later). That is, the feature amount C is considered to have appeared normally as the user's action.

  Here, in Equation (3), the reason for multiplying the number of log increases by a coefficient will be described. When the behavior changes, there are logs that newly appear according to the change, but there are also logs that do not appear.

  When calculating the amount of change, if the difference between the number of logs appearing before and after is taken as the amount of change, the decrease in the amount of change is falsely recognized as the presence of noise despite the fact that a new action has appeared There is a possibility of

  For example, when transitioning from in-meeting (still, speaking) to in-office walking (walking), the log corresponding to the feature value is reduced by one, but a new action is appearing. However, if it is determined simply by the difference in the number of logs, it may be erroneously recognized as having noise.

  Therefore, as in equation (3), the number of log changes is divided into the number of decreases and the number of increases, and the number of increases is multiplied by the coefficient k (> 1). As a result, new actions can be greatly evaluated, and the occurrence of false recognition as having noise is suppressed.

  FIG. 13 is a diagram showing an example of the noise determination table. The noise determination table Tb3 has, as items, changes in human dependent logs, changes in environment dependent logs, action scene switching, and determination results. In addition, that a user's action is performed in a specific physical range is defined as an action scene.

  Describing each column, if (change in human dependent log, change in environment dependent log) = (no change, increase), the feature amount correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the noise determination result is noise. An example of this case is the movement of surrounding people.

  When (change in human dependent log, change in environment dependent log) = (no change, decrease), the feature amount correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the determination result is noise. One example is the movement of surrounding people.

  In the case of (change in human dependent log, change in environment dependent log) = (increase, increase), the feature amount correction unit 13 determines that there is a switch of action scenes. Therefore, the feature amount correction unit 13 determines that the noise determination result is no noise. An example is intra-office movement (movement from a meeting to another meeting room).

  When (change in human dependent log, change in environment dependent log) = (increase, no change), the feature amount correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the determination result is noise. One example is the material explanation at the meeting.

  In the case of (change in human dependent log, change in environment dependent log) = (increase, decrease), the feature amount correction unit 13 determines that the action scene is switched. Therefore, the feature amount correction unit 13 determines that the determination result is no noise. An example is intra-office movement (movement from a meeting to another meeting room).

  In the case of (change in human dependent log, change in environment dependent log) = (decrease, increase), the feature amount correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the determination result is noise. One example is movement during business trips.

  In the case of (change in human dependent log, change in environment dependent log) = (decrease, no change), the feature quantity correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the determination result is noise. One example is movement during business trips.

  In the case of (change in human dependent log, change in environment dependent log) = (decrease, decrease), the feature quantity correction unit 13 determines that the action scene is not switched. Therefore, the feature amount correction unit 13 determines that the determination result is noise. One example is movement during business trips.

As described above, the feature quantity correction unit 13 determines the switching of the action scene from the change amount of the human dependency log and the environment dependency log, and performs the noise determination from the determination result.
FIG. 14 is a diagram showing the state after the feature amount correction. From the above, in the transition from the feature amount A to the feature amount B, it is determined that the feature amount B is generated due to noise. Therefore, the feature quantity correction unit 13 replaces the feature quantity B at time t2 with the feature quantity A at time t1.

  Further, in the transition from the feature amount A to the feature amount C, it is determined that the feature amount C is not generated due to noise but that a new action has appeared. Therefore, the feature amount correction unit 13 leaves the feature amount C at time t4 as it is.

<Flow chart>
FIG. 15 is a flowchart showing the operation of the server.
[Step S31] The feature quantity generation unit 12 periodically receives a log.

[Step S32] The feature quantity generation unit 12 generates a feature quantity from the received log.
[Step S33] The feature quantity correction unit 13 determines whether the feature quantity at the previous time (referred to as feature quantity fv1) differs from the feature quantity at subsequent time (referred to as feature quantity fv2) before and after the appearance time of the feature quantity It is determined whether or not. If they differ, the process proceeds to step S34. If they match, the process returns to step S31.

[Step S34] The feature quantity correction unit 13 refers to the logs included in the feature quantities fv1 and fv2 and classifies them into human dependent logs or environment dependent logs.
[Step S35] The feature quantity correction unit 13 calculates the number of log changes of the human dependency log included in the feature quantity fv1 of the previous time and the human dependency log included in the feature quantity fv2 of the subsequent time. In addition, the feature quantity correction unit 13 calculates the number of log changes of the environment dependent log included in the feature quantity fv1 of the previous time and the environment dependent log included in the feature quantity fv2 of the later time.

  [Step S36] The feature amount correction unit 13 obtains the amount of change from the calculated number of log changes, and obtains the increase / decrease tendency of the human dependent log and the environment dependent log. Then, the feature amount correction unit 13 obtains a noise determination result from the obtained increase / decrease tendency based on the noise determination table Tb3. If the noise determination result indicates noise, the process proceeds to step S37. If the noise determination result indicates no noise, the process proceeds to step S38.

[Step S37] The feature amount correction unit 13 determines that the feature amount fv2 is generated due to noise, and replaces the feature amount fv2 of the later time with the feature amount fv1 of the previous time.
[Step S38] The feature amount correction unit 13 determines that the feature amount fv2 is not generated due to noise but that a new action has appeared, and leaves the feature amount fv2 of the later time as it is.

[Step S39] The feature amount group processing unit 14 groups the corrected feature amounts.
As described above, in the server 10, the log is converted to the feature amount, and for the feature amounts not continuous before and after the time, the suitability of the continuity of the feature amounts is judged from the change amount of the log classified according to the property. Inappropriate feature quantities are regarded as being identical to the previous feature quantity and replacement is performed.

  Thereby, since the server 10 can correct only the erroneously recognized action information, it is possible to improve the accuracy of the action analysis even when the action analysis is performed using the coarse granularity log .

  The processing functions of the analyzer 1 and the server 10 of the present invention described above can be realized by a computer. In this case, a program is provided in which processing contents of functions to be possessed by the analyzer 1 and the server 10 are described. The above processing functions are realized on the computer by executing the program on the computer.

  The program in which the processing content is described can be recorded on a computer readable recording medium. Examples of the computer-readable recording medium include a magnetic storage device, an optical disc, a magneto-optical recording medium, a semiconductor memory, and the like. The magnetic storage device includes a hard disk drive (HDD), a flexible disk (FD), a magnetic tape and the like. The optical disc is, for example, a CD-ROM / RW. Magneto-optical recording media include MO (Magneto Optical disk) and the like.

  When the program is distributed, for example, a portable recording medium such as a CD-ROM in which the program is recorded is sold. Alternatively, the program may be stored in the storage device of the server computer, and the program may be transferred from the server computer to another computer via a network.

  The computer executing the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing in accordance with the program.

  The computer can also execute processing in accordance with the received program each time the program is transferred from the server computer connected via the network. In addition, at least a part of the above processing functions can be realized by an electronic circuit such as a DSP, an ASIC, or a PLD.

  As mentioned above, although embodiment was illustrated, the structure of each part shown by embodiment can be substituted to the other thing which has the same function. Also, any other components or steps may be added. Furthermore, any two or more configurations (features) of the above-described embodiments may be combined.

1 analyzer 1a control unit 1b storage unit A, B, C feature information d1a, d1b, d1c first action information d2a, d2b, d2c second action information α, β non-consecutive feature information set

Claims (9)

The feature information representing the action state for each unit time is generated in time series from the action information in which the action of the user is sensed,
When there is non-continuous non-continuous feature information in which non-continuous feature information after the unit time is non-continuous with respect to the feature information before the unit time, the action information included in the non-continuous feature information Classified into 2 action information,
A control unit that determines the appropriateness of the continuity of the feature information from the increase and decrease tendency of both the first action information and the second action information, and corrects the continuity of the feature information based on the determination result,
Analyzer with. The control unit
When the non-continuous feature information includes the first feature information at the previous time and the second feature information at the later time,
When it is determined that the continuity of the feature information is inappropriate, the second feature information is replaced with the first feature information, and the non-continuous feature information is corrected to continuous feature information.
When it is determined that the continuity of the feature information is appropriate, the state of the non-continuous feature information is maintained without replacing the second feature information with the first feature information.
The analyzer according to claim 1.   The first action information is human-dependent action information having a property of changing a value according to an action actively performed by the user, and the second action information is an environment around the user. The analyzer according to claim 1, wherein the behavior information is environment-dependent and has a property of changing value according to. The control unit
A first decrease number and a first increase number with respect to the number of the first action information included in the first feature information and the number of the first action information included in the second feature information To calculate the first change amount,
A second decrease number and a second increase number with respect to the number of the second action information included in the first feature information and the number of the second action information included in the second feature information To calculate the second change amount,
The analyzer according to claim 2.   The control unit calculates the first change amount by adding the first decrease number to the weighted first increase number, and calculates the second decrease number to the weighted second increase number. The analyzer according to claim 4, wherein the second change amount is calculated by addition. The control unit
A first increase / decrease tendency of the first action information is detected based on the first change amount, and a second increase / decrease tendency of the second action information is detected based on the second change amount.
When the first increase / decrease tendency is not changed, and when the second increase / decrease tendency is increased, when the first increase / decrease tendency is not changed and when the second increase / decrease tendency is decreased, the first The increase / decrease tendency increases, and the second increase / decrease tendency does not change, the first increase / decrease tendency decreases, and the second increase / decrease tendency increases, and the first increase / decrease tendency decreases. The continuity of the feature information is determined to be inappropriate between the case where the second increase / decrease tendency is not changed and the case where the first increase / decrease tendency is decreased and the second increase / decrease tendency is decreased.
The analyzer according to claim 4. The control unit
A first increase / decrease tendency of the first action information is detected based on the first change amount, and a second increase / decrease tendency of the second action information is detected based on the second change amount.
In the case where the first increase / decrease trend is increased and the second increase / decrease trend is increased, and in the case where the first increase / decrease trend is increased and the second increase / decrease trend is decreased, the characteristic information is continuous Determine the appropriateness of the nature,
The analyzer according to claim 4. The computer is
The feature information representing the action state for each unit time is generated in time series from the action information in which the action of the user is sensed,
When there is non-continuous non-continuous feature information in which non-continuous feature information after the unit time is non-continuous with respect to the feature information before the unit time, the action information included in the non-continuous feature information Classified into 2 action information,
The propriety of the continuity of the feature information is determined from the increase / decrease tendency of both the first action information and the second action information, and the continuity of the feature information is corrected based on the determination result.
Analysis method to control. On the computer
The feature information representing the action state for each unit time is generated in time series from the action information in which the action of the user is sensed,
When there is non-continuous non-continuous feature information in which non-continuous feature information after the unit time is non-continuous with respect to the feature information before the unit time, the action information included in the non-continuous feature information Classified into 2 action information,
The propriety of the continuity of the feature information is determined from the increase / decrease tendency of both the first action information and the second action information, and the continuity of the feature information is corrected based on the determination result.
A program that performs control.

Images