JP2020071858A - Parameter estimation program, device, and method - Google Patents

Abstract

To estimate parameters for evaluating interest level in real time with respect to terminal posture changes.SOLUTION: An information processing terminal 16 calculates the cumulative distribution of the terminal momentum observed in the first period regarding the terminal momentum of the terminal and determines a temporary value of the parameter for evaluating the degree of interest of the terminal operator from the default rank in the cumulative distribution. A new distribution of the terminal momentum observed during the second period is calculated. Comparing the centroid of the cumulative distribution with the centroid of the new distribution, and depending on the result of the comparison, the default rank in the cumulative distribution is corrected, and the parameter is updated.SELECTED DRAWING: Figure 11

Description

  The present invention relates to a parameter estimation program, device and method.

  Conventionally, there is a technique of calculating a desired estimated value using a distribution of momentum such as speed.

  For example, there is a technique relating to an ultrasonic diagnostic apparatus using the distribution of velocity information. This ultrasonic diagnostic apparatus detects a distribution of velocity information over a predetermined period, calculates a feature amount based on a flow velocity in the predetermined period using the velocity information, and displays the feature amount in a predetermined form.

  Further, there is a technique related to a method for estimating the inlet coolant flow rate using the heat output distribution obtained by the core performance calculation. In this method, the weight value is corrected in consideration of the horizontal distribution of the bubble velocity in the fuel channel.

  Further, there is a technology that assists the customer in actual purchasing behavior when purchasing what he / she wants at an EC site using a mobile terminal such as a smartphone or a tablet.

  For example, there is a technique of evaluating the degree of interest in the content being displayed by the operator who operates the terminal. In this technique, the amount of exercise of the information processing terminal is acquired, and the degree of interest in the content is evaluated based on the length of the period during which the amount of exercise is less than or equal to the average amount of exercise in the display period of the content.

  In addition, there is a technique for setting an operation target item of an electric device according to the detected posture. In this technique, a posture change is detected, and it is determined that the posture has changed when a person stands for a certain period of time.

  In addition, there is a technique for detecting the tilt of a mobile terminal using a tilt sensor or the like. In this technique, the object displayed on the display unit of the mobile terminal is processed according to the displacement of the tilt, and an erroneous operation at the time of operation is suppressed.

  In addition, there is a technique of calculating the degree of interest and the degree of confusion of a user who uses the information processing terminal for the content, based on information acquired from the information processing terminal that accesses the content. In this technique, the degree of interest is calculated by the staying time of the website and the terminal operation. Also, the degree of confusion is calculated based on the operation pattern.

JP2012-176232A JP 2002-267787 A JP, 2008-036912, A JP, 2008-64849, A JP, 2016-57461, A JP, 2007-309803, A International publication number WO2014 / 185444 JP, 2016-103269, A JP, 2009-100366, A JP, 2018-41327, A JP, 2018-67158, A JP, 2007-226371, A JP, 2005-149051, A JP, 2008-126374, A

  However, in the conventional technology, when the posture of the terminal changes due to a change in the way the user holds the terminal, it is necessary to re-estimate the parameters required to evaluate the degree of interest. There was a problem that it was difficult to evaluate the degree of interest.

  One aspect of the present invention is to estimate a parameter for evaluating the degree of interest in real time with respect to a change in the attitude of a terminal.

  As one aspect, a first distribution of terminal momentum of the terminal observed in the first period is obtained, and the degree of interest of the operator of the terminal is evaluated from the terminal momentum corresponding to a predetermined rank in the first distribution. The provisional values of the parameters for A second distribution of the terminal momentum observed during a second period, which is a period after the first period, is obtained. The center of gravity of the first distribution and the center of gravity of the second distribution are compared, the predetermined rank in the first distribution is corrected, and the parameters are updated according to the result of the comparison.

  As one aspect, there is an effect that it is possible to estimate a parameter for evaluating the degree of interest in real time with respect to the attitude change of the terminal.

It is an example of a graph of reliability in interest level evaluation. It is a block diagram which shows schematic structure of the interest degree evaluation system which concerns on the 1st Embodiment of this invention. It is a figure which shows an example of how to determine the parameter from a terminal momentum. It is a figure which shows an example of how to determine a parameter when a distribution centroid exists in the area | region with low terminal momentum. It is a figure which shows an example at the time of comparing using the gravity center distance of distributions. It is a figure which shows the example of a data structure of frequency distribution of terminal momentum. It is a block diagram which shows schematic structure of the computer which functions as an information processing terminal of 1st Embodiment. It is a block diagram which shows schematic structure of the computer which functions as a content server. It is a flow chart which shows an example of interest degree evaluation processing concerning a 1st embodiment of the present invention. It is a flow chart which shows an example of processing of a parameter estimating part. It is a graph which shows an example of the experimental result of the method of the 1st Embodiment of this invention. It is a figure which shows an example at the time of clustering the terminal momentum and posture when a posture change has not occurred. It is a figure which shows an example in the case of clustering the terminal momentum and posture when a posture change occurs. It is a figure which shows an example at the time of quantifying a trunk movement by the magnitude of the fluctuation of a posture vector. It is a figure which shows an example at the time of calculating | requiring conversion momentum from each of posture vector. It is a figure which shows an example of the time chart of the parameter used when not considering a posture change. FIG. 17 is a diagram showing an example of a time chart of parameters used when the posture change is considered. It is a block diagram which shows schematic structure of the interest degree evaluation system which concerns on the 2nd Embodiment of this invention. It is a block diagram which shows an example of a structure of a short term parameter estimation part. It is a block diagram which shows an example of a structure of a long term parameter estimation part. It is a block diagram which shows schematic structure of the computer which functions as an information processing terminal of 2nd Embodiment. It is a flow chart which shows an example of interest degree evaluation processing concerning a 2nd embodiment of the present invention. It is a flow chart which shows an example of processing of a long-term parameter estimating part. It is a flow chart which shows an example of processing of a posture change judgment part. It is a flow chart which shows an example of processing of a parameter integration part. It is a graph which shows an example of the experimental result of the method of the 2nd Embodiment of this invention.

  Before describing each embodiment of the present invention, a background of the premise will be described first.

  In recent years, not only personal computers but also smartphones (hereinafter referred to as smartphones) and smart device terminals such as tablets are used to search for information and obtain the desired information. If the degree of interest of the user can be detected in such a usage scene, not only the marketing effect but also a new service becomes possible. Taking Web service on EC sites as a typical example, it is drawing attention to implement measures by One To One according to individual tastes. In this case, if it is possible to know not only the information that the Web page was accessed but also the actual interest in the content and where on the page it was interested, it becomes important information as customer information. sell. And when such data and purchasing data are gathered, it will suggest what kind of person should buy and what kind of measures should be taken when assuming a persona.

  Against this background, there is a method of evaluating the degree of interest in consideration of the movement of the terminal when gazing at the terminal (see Patent Document 3).

  In this method, information is provided in real time based on interest level evaluation. In this method, when the user is browsing the content of interest, the user fixes the terminal so as to gaze at the screen, so that the terminal movement is suppressed. In this method, it is assumed that the user is interested when the terminal motion per unit time is below a certain level and continues for a predetermined time, and a model formula for weighting the reliability is set so as to reduce the interest in a large terminal motion. There is. Since the baseline value of the terminal exercise depends on the habit of individual operation of the user, by controlling the weighting of the reliability for each individual, it is possible to evaluate the interest level according to the individual.

  In addition, there is a method of detecting the habit of individual operation in the evaluation of interest level. In this method, in order to estimate the parameters related to terminal momentum, not only the terminal momentum for a certain period but also dimensions such as posture information and operation information are added to obtain information that becomes noise, and the degree of interest is accurately determined. Seeking However, this method focuses on evaluating individual characteristics through analysis later. Real-time interest level evaluation is necessary to realize a service that appropriately works according to the psychology of a user who is visiting a website at the present time. However, in this method, in order to reduce the influence of the terminal momentum during posture change, it is necessary to perform clustering of the terminal momentums detected in a predetermined time (for example, a relatively long period such as 40 seconds), and the real-time interest It was not able to correspond to the degree evaluation. Therefore, this method has a problem that there is no parameter estimation value in the first learning period, and it is difficult to follow the interest degree evaluation when a posture change occurs.

  Regarding the parameter estimation method necessary for performing such real-time evaluation, there are mainly two methods in the disclosure of the related art. One is parametric estimation, and the maximum likelihood estimation method is a typical method. Maximum-likelihood estimation is an estimation method that maximizes posterior probability, and for this purpose a probabilistic model must be known. Patent Document 4 and Patent Document 5 are examples in which parametric estimation is used for speech recognition and speaker separation, respectively.

  The other is a nonparametric estimation method, in which the shape of the probability density function to be estimated is not predetermined but is determined depending on the data. The method using the histogram is one of the simplest methods for nonparametric estimation. As a technique using non-parametric estimation, for example, in autonomous navigation of a moving body using data of a gyro sensor, it is possible to obtain a displacement amount in a rotation direction or a translation direction by matching histograms (Patent Documents 6 to 8). ..

  However, in the parametric estimation methods as disclosed in Patent Documents 4 and 5, it is difficult to model a probability density function that represents the motion of a terminal when held in a hand by some signal sources. It cannot be applied to evaluation. Further, the methods of Patent Documents 6 to 8 also require distributions to be compared and are not suitable for use in interest degree evaluation. In other words, in any case, the data to be referred to is required, and it has been difficult to grasp the degree of interest according to the habit of individual operation in the first section when browsing the first visited Web page.

  In view of the above points, each embodiment of the present invention provides a method for sequentially updating the estimated value of the parameter used for evaluation of interest regarding the habit of individual operation.

  The principle of interest level evaluation, which is the premise of each embodiment of the present invention, will be briefly described.

  The interest level is calculated by the following equation (1), for example.

Degree of interest = Σ (operation type coefficient × operation time × reliability) (1)

  The operation type coefficient is a coefficient that is determined based on the scroll speed calculated by analyzing the touch state on the operation surface of the terminal, the sensor value detected in the detection of the terminal momentum, and the like It may be determined according to The operation time is the operation time for each operation type of the user within the unit time. The reliability is a value determined based on the gaze state and the non-gaze state. The gaze state and the non-gaze state are based on the principle that the movement of the terminal becomes small when the user is gazing at the screen of the smartphone with interest. Note that the non-operation time may be used in addition to the operation time for calculating the interest level.

As for the reliability, as shown in FIG. 1, consider a reliability curve that decreases as the terminal momentum increases. In this reliability curve, the individual characteristic parameter can be represented by two parameters. As shown in the reliability graph of FIG. 1, one is a terminal momentum parameter P _th that starts to decrease the reliability from the maximum value, and the other is a terminal that sets the reliability to the minimum value (for example, 0). The momentum parameter P ₀ .

In the method using clustering described above, the parameter P _th represents the average terminal momentum in the gaze state, and is considered to be, for example, the average terminal momentum of the cluster in which the distribution is most concentrated. Therefore, it can be treated as having high reliability. P ₀ represents the average terminal momentum in the non-gaze state.

In the clustering method, in order to classify the terminal momentum during the gazing and the non-gazing, by adding not only the momentum but also information such as the posture and the terminal operation to increase the dimension and perform clustering, the estimation accuracy of the parameter P _th is improved. There is. By evaluating the degree of interest by applying the parameter P _th obtained by the clustering, it becomes possible to understand where the degree of interest has increased in the time series by compensating for the difference between the users and to know the upper and lower sides of the degree of interest. However, even if this method can be used for later analysis, it was difficult to evaluate the degree of interest in real time and give feedback to the user accordingly.

On the other hand, by using the clustering method, it was found that when a task (reading a sentence, looking at an image, etc.), which is a specific purpose of the user's terminal operation, is performed, the results show that the parameters change in a certain section. There is. Cluster analysis parameters P _th estimated in those to those changes by the nature of the task, also a parameter P _th by the timing analyzing the cluster among the same task changes. With respect to such characteristics of the parameter P _th , the parameters are determined so that it is clear whether or not there is a degree of interest.

Even when the degree of interest is evaluated in real time, there is a difference in the length of the section to be evaluated, a parameter is estimated from the terminal momentum observed in a certain section of a certain task, and the parameter P _th is set as a threshold value. Distinguishing whether or not there is a gaze is similar. Regardless of whether the section is a short term or a long term, it is possible to divide the data of the terminal momentum observed in a certain section, and if the interest level rises or falls based on the divided data, the interest level can be evaluated in real time. It will be possible. Therefore, the real-time, if the parameter P _th regarding gaze is determined, if a large momentum than parameter P _th is observed, it lowered the interest that each time calculated. Since it is not essential to obtain the parameter P ₀ regarding non-gaze, it is sufficient to obtain the parameter P _th in real time.

Therefore, in the first embodiment of the present invention, the parameter P _th for evaluating the interest level in real time is estimated based on the distribution of the terminal momentum.

  Hereinafter, an example of the first embodiment of the present invention will be described in detail with reference to the drawings.

<First Embodiment>
As shown in FIG. 2, the interest level evaluation system 10 according to the first embodiment includes a content server 12 and an information processing terminal 16 (hereinafter, sometimes simply referred to as “terminal”). The content server 12 and the information processing terminal 16 are connected via a network 14 such as the Internet. The information processing terminal 16 is an example of the parameter estimation device.

  The content server 12 transmits the content to the information processing terminal 16 in response to the content request signal from the information processing terminal 16.

  The information processing terminal 16 includes a communication unit 18, a control unit 20, a display unit 22, a user operation detection unit 24, a terminal exercise amount detection unit 26, and a touch state analysis unit 28. In addition, the information processing terminal 16 includes a first distribution calculating unit 30, a second distribution calculating unit 32, a comparing unit 34, a parameter updating unit 36, and an interest degree evaluating unit 38.

  The communication unit 18 sends and receives information to and from the content server 12. For example, the communication unit 18 receives the content transmitted from the content server 12. The communication unit 18 also transmits a content request signal output by the control unit 20 described later to the content server 12.

  The control unit 20 controls the display unit 22 described below so that the content received by the communication unit 18 is displayed.

  The display unit 22 is realized by a display such as a Liquid Crystal Display (LCD) or an Organic Electroluminescence Display (OELD). On the display unit 22, a screen change occurs according to an input operation such as a touch of the user. The display unit 22 presents the content under the control of the control unit 20.

  The user operation detection unit 24 receives an operation input by the user from the touch panel superimposed on the display unit 22 while displaying the content on the information processing terminal 16, determines the type of the input operation by the user, and the presence / absence of the input operation. To detect. Specifically, the user operation detection unit 24 detects the type of input operation such as tap, flick, swipe, and pinch by the user. The user operation detection unit 24 also detects a scroll operation for scrolling the screen based on the type of input operation. Then, the user operation detection unit 24 measures the operation time of the operation in the unit time.

  The terminal momentum detection unit 26 detects a sensor value related to the movement of the terminal as the terminal momentum at predetermined sample intervals. In the first embodiment, a mode in which the terminal momentum detecting unit 26 is realized by a 9-axis sensor will be described. The 9-axis sensor is a gyro sensor including three types of sensors, namely, a 3-axis angular velocity sensor, a 3-axis acceleration sensor, and a 3-axis geomagnetic sensor. However, the gyro sensor of the terminal momentum detection unit 26 may be realized by one or more types of the above three types of sensors.

  The touch state analysis unit 28 detects a touch state indicating whether or not the touch panel is being touched and an operation state corresponding to the operation type of the touch. The touch operation type includes zooming and scrolling, and detects the operation state by each operation, for example, touch start, touch continuing, LEAVE (non-touch state such as touch end), scroll position, and scroll speed. ..

Here, the premise and principle of the estimation method of the parameter P _th in the first embodiment will be described.

First, a typical method of estimating the parameter P _th will be described. As shown in FIG. 3, the median value of the terminal momentum observed in a certain section of a certain task is obtained, the median value is set as a parameter P _th , the section in which the terminal momentum is smaller than the parameter P _th is a gaze section, and the terminal is set from the parameter P _th A section with a large amount of exercise is defined as a non-gaze section. In this way, estimating the parameter P _th can increase or decrease the degree of interest calculated each time according to the magnitude of the terminal momentum observed in a certain section.

This method is different from the previous clustering method in that the parameter P _th is estimated by using the rank when the terminal momentums are sorted in ascending or descending order in order to distinguish information whose dimension is one-dimensional. is there. Although there may be a case where it is difficult to classify by clustering with one-dimensional information, there is always a solution in the method of classifying by rank. Instead, this method is a method based on the assumption that the distribution of terminal momentum varies to some extent. In the case where the distribution is biased, a clustering method that has flexibility in the number of elements included in the cluster works well. And in reality such cases are often seen. If a distribution-based estimation method is applied to such a case and classification is performed in order of sorting, a problem may occur. For example, the distribution is as shown in FIG. If the distribution centroid is in a region with many gaze sections and low terminal momentum, the evaluation result becomes too strict if the median value is set to the parameter P _th , which may be inappropriate as a threshold for evaluation of interest. know. In such a case, it is necessary to loosen the estimated value by performing a correction to shift to a parameter P ′ _th having a larger terminal momentum than the median parameter P _th .

Further, when the estimated value of the parameter P _th regarded as in gaze is obtained, the parameter P _th which is obtained from a terminal momentum observed thereafter if distribution is unchanged, was estimated to terminal momentum in gaze parameter P _th We know that it will never fall below. On the contrary, if it falls below, the lowered terminal momentum is the original gaze threshold value, and it is necessary to update so as to correct the estimated value of the parameter P _th so far. In detecting this state, in the case of real-time time-series data, attention is paid to changes in the cumulative frequency distribution of the terminal momentums observed up to that time and the newly observed frequency distribution of the terminal momentums. If such a change is repeated, it means that the cumulative frequency distribution has changed. The case where the center of gravity of the distribution exists in the region where the terminal momentum is low occurs as a result.

Based on the above assumptions, the principle of estimating the user parameter P _th in real time will be described. First as a mechanism for updating the parameter estimates P _th in real time, estimated values of the parameters P _th in the default order (e.g., the median) when the terminal momentum observed so far, were arranged in the order of magnitude Determine. Then, when there is a change in the median value of the terminal momentum, which is the threshold value of whether or not the user is watching, in the frequency distribution of the terminal momentum, the predetermined rank is corrected according to the degree of change. As described above, if the distribution does not change, the parameter P _th determined by the predetermined rank can be regarded as the lower limit of the terminal momentum during gazing. That is, when the median value of the terminal momentum in the new frequency distribution is lower than the median value of the terminal momentum (parameter P _th ) in the cumulative frequency distribution up to that point, a new lower limit is set for the terminal momentum during gazing. Means that a change has been discovered. Specifically, the new lower limit change is, for example, between the cumulative frequency distribution of the terminal momentums observed up to that time and the newly observed frequency distribution of the terminal momentums, as shown in FIG. It is detected by comparing using the center of gravity. When the centroid in the frequency distribution of the newly observed terminal momentum is lower than the centroid of the cumulative frequency distribution of the terminal momentum observed up to that point, the estimated median parameter P _th is significantly corrected. Become. However, if the influence of the correction is too large, the change of the parameter P _th will be accelerated toward an inappropriate value. Therefore, the correction position Δi is introduced as a variable indicating how much the predetermined rank is shifted, and the larger the center of gravity of the distribution, the larger the correction is performed to determine the estimated value of the parameter P _th . The correction position Δi is designed so as to be within a small value if the change in distribution is not so large. In the above description, the default rank is the median value for ease of understanding, but this is an example, and the rank corresponding to a predetermined percentile may be defined as the default rank.

Each of the following processes will be described based on the principle of the method of estimating the parameter P _th .

The first distribution calculation unit 30 obtains the cumulative frequency distribution of the terminal momentums observed in the first period based on the terminal momentums detected by the terminal momentum detection unit 26. The first distribution calculation unit 30 determines a provisional value of the reliability parameter P _th for evaluating the degree of interest of the operator of the terminal from the predetermined rank in the cumulative frequency distribution. In the first embodiment, the default rank is the median value. The cumulative frequency distribution is an example of the first distribution. The cumulative period from the start of the process of the first distribution calculation unit 30 to the reset is an example of the first period. Further, here, the first period is a period immediately before a second period described later.

The first distribution calculating unit 30 first waits for the input of the terminal momentum from the terminal momentum detecting unit 26, and starts the process when a predetermined number of samples are collected. Since the frequency distributions are compared in the first embodiment, it is preferable to start processing after obtaining a sample of terminal momentum of at least about 5 samples. For example, if the terminal exercise amount is acquired every 100 ms, the process is designed to be performed every 500 ms. One sample takes 10 seconds. A delay occurs in such block processing, but if it is about 1 second, it is considered that the user's operational feeling does not feel awkward. It suffices to set an appropriate block length and sample interval within a range of fitting within 1 second and determine the required number of samples. Then, when the required number of samples are prepared, the cumulative frequency distribution of these terminal momentums is obtained. In the first embodiment, the cumulative frequency distribution is obtained by processing every 5 samples. That is, the unit time is 0.5 seconds, and the first period is accumulated every 0.5 seconds. If the process is the first time, the first distribution calculation unit 30 sets the median value of the first frequency distribution as the initial value of the parameter P _th . If the process is the second or later, the first distribution calculation unit 30 updates the cumulative frequency distribution with the terminal momentum observed in the first period, and the provisional value of the parameter P _{th is} the median of the updated cumulative frequency distribution. A value is set and the cumulative frequency distribution is output to the comparison unit 34. Note that the cumulative frequency distribution is updated after adding the newly observed terminal momentum, the provisional value of the parameter P _th is set by the median of the updated cumulative frequency distribution, and the cumulative frequency distribution is output to the comparison unit 34. You may do it. In this case, the latest unit time of the cumulative period in the first period is the second period. Further, since the new frequency distribution is calculated by the second distribution calculation unit 32, the one calculated by the second distribution calculation unit 32 may be used for the first time.

FIG. 6 shows an example of the data structure of the frequency distribution of terminal momentum. In this data structure, the frequency is counted when the observed terminal momentum (sum of squares of values of three axes of gyro sensor data) is within each class range. As for the provisional value of the parameter P _th, the correction position Δi is initially set to 0, and the class representative value of the class including the median of this frequency distribution is selected. Here, the class representative value is defined as, for example, the geometric mean of the lower boundary value and the upper boundary value indicating the range of the class. After that, as the parameter P _th , the histogram total number $ sumOfFreq is obtained, and the class representative value of the class including the observed value of the order of ($ sumOfFreq + Δi) / 2 is set as the provisional value of the parameter P _th . Note that the provisional value of the parameter P _th may be set in the order of percentiles when the terminal momentums are arranged in ascending or descending order. For example, if there are 20 samples in the cumulative frequency distribution and the 40th percentile when arranged in ascending order, the 8th smallest terminal momentum is set as the provisional value of the parameter P _th .

  The second distribution calculating unit 32, based on the terminal momentum detected by the terminal momentum detecting unit 26, the terminal momentum observed in the second period which is a period after the first period of the first distribution calculating unit 30. Find a new frequency distribution for. In the first embodiment, since the cumulative frequency distribution is updated every 5 samples, a new frequency distribution is obtained each time 5 most recently observed terminal momentums are obtained. The new frequency distribution is an example of the second distribution. The period during which the latest 5 samples are observed is an example of the second period.

  The comparison unit 34 detects a change that is a new lower limit of the terminal momentum in the cumulative frequency distribution by comparing the centroid of the cumulative frequency distribution with the center of gravity of the new frequency distribution. In the first embodiment, the center of gravity is the median value. The new lower limit change is detected as a new lower limit change when the center of gravity of the new frequency distribution falls below the center of gravity of the cumulative frequency distribution. Further, the comparison unit 34 obtains the center-of-gravity distance between the center of gravity of the cumulative frequency distribution and the center of gravity of the new distribution. The center-of-gravity distance may be a distance obtained by subtracting the center of gravity of a new frequency distribution from the center of gravity of a cumulative frequency distribution, but instead, a ratio may be used instead. When taking the ratio, the ratio of the centroid of the new frequency distribution to the centroid of the cumulative frequency distribution is used, and the centroid of the cumulative frequency distribution is taken as the denominator. The ratio of the center of gravity means that the closer the ratio obtained from the center of gravity is to 1, the closer the center of gravity distance is, and the smaller the ratio is, the farther the center of gravity distance is. If the ratio obtained from the center of gravity is greater than 1, it cannot be considered that the user is gazing from the terminal momentum of the new frequency distribution, and therefore need not be considered.

When the comparison unit 34 detects a new lower limit change, the parameter updating unit 36 corrects the median of the cumulative frequency distribution based on the distance between the center of gravity of the cumulative frequency distribution and the center of gravity of the new frequency distribution. Then, the parameter P _th is updated. When the comparison unit 34 has not detected a new lower limit change, the parameter update unit 36 updates the provisional value defined by the median value as the parameter P _th .

  Specific processing of the parameter updating unit 36 will be described. The parameter updating unit 36 calculates the correction position Δi: by the following equation (2) based on the distance of the center of gravity. The correction position Δi is a value obtained by adding the weight α according to the distance of the center of gravity to the correction position Δi.

Δi: = Δi + α (2)

The value of α may be set by providing a threshold value according to the distance of the center of gravity. For example, when the ratio obtained from the center-of-gravity distance is smaller than 0.25 when 0.25 is set as the threshold value, α = 2 is set, Δi: = Δi + 2 is calculated, and a correction of removing +2 as the correction position is performed. To do. Then, the terminal momentum in the order of the median + correction position Δi in the cumulative frequency distribution is used as the estimated value of the parameter P _th . If the value obtained from the distance of the center of gravity is 0.10, α = 3 may be set and Δi: = Δi + 3 may be corrected.

The interest level evaluation unit 38 updates the scroll speed calculated by the touch state analysis unit 28, the sensor value detected by the terminal momentum detection unit 26, the operation time detected by the user operation detection unit 24, and the parameter update unit 36. The degree of interest is calculated based on the calculated parameter P _th . The interest level evaluation unit 38 calculates the interest level per unit time according to the above equation (1). Since the calculation method is as described above, it is omitted here.

Although the above process is sequentially repeated, there is also a method of accumulating the cumulative frequency distribution in order to follow a situation change, and there is also a method of resetting after a certain amount of time and restarting the update from the beginning. When redoing, it is conceivable to use the estimated value of the parameter P _th updated up to that point, or use the value estimated by the clustering performed by the prior method instead and proceed.

  The information processing terminal 16 can be realized by, for example, the computer 50 shown in FIG. 7. The computer 50 includes a CPU 51, a memory 52 as a temporary storage area, and a non-volatile storage unit 53. Further, the computer 50 includes a display unit 22, a touch panel or the like which is the user operation detection unit 24 superimposed on the display unit 22, an input / output interface (I / F) 54 to which the terminal exercise amount detection unit 26 is connected, and a recording. A read / write (R / W) unit 55 that controls reading and writing of data with respect to the medium 59 is provided. The computer 50 also includes a network I / F 56 connected to a network such as the Internet. The CPU 51, the memory 52, the storage unit 53, the input / output I / F 54, the R / W unit 55, and the network I / F 56 are connected to each other via a bus 57.

  The storage unit 53 can be realized by a hard disk drive (HDD), a solid state drive (SSD), a flash memory, or the like. An interest level evaluation program 60 for causing the computer 50 to function as the information processing terminal 16 is stored in the storage unit 53 as a storage medium. The interest level evaluation program 60 includes a communication process 62, a control process 63, a user operation detection process 65, a touch state analysis process 66, a first distribution calculation process 67, a second distribution calculation process 68, and a comparison process 69. And a parameter update process 70 and an interest degree evaluation process 71.

  The CPU 51 reads the interest level evaluation program 60 from the storage unit 53, expands it in the memory 52, and sequentially executes the processes of the interest level evaluation program 60. The CPU 51 operates as the communication unit 18 illustrated in FIG. 2 by executing the communication process 62. Further, the CPU 51 operates as the control unit 20 shown in FIG. 2 by executing the control process 63. Further, the CPU 51 operates as the user operation detection unit 24 shown in FIG. 2 by executing the user operation detection process 65. Further, the CPU 51 operates as the touch state analysis unit 28 shown in FIG. 2 by executing the touch state analysis process 66. Further, the CPU 51 operates as the first distribution calculation unit 30 shown in FIG. 2 by executing the first distribution calculation process 67. Further, the CPU 51 operates as the second distribution calculation unit 32 shown in FIG. 2 by executing the second distribution calculation process 68. Further, the CPU 51 operates as the comparison unit 34 illustrated in FIG. 2 by executing the comparison process 69. Further, the CPU 51 operates as the parameter updating unit 36 shown in FIG. 2 by executing the parameter updating process 70. Further, the CPU 51 operates as the interest level evaluation unit 38 shown in FIG. 2 by executing the interest level evaluation process 71. As a result, the computer 50 that executes the interest level evaluation program 60 functions as the information processing terminal 16. The CPU 51 that executes the program is hardware. The interest level evaluation program 60 is an example of a parameter estimation program.

  The function realized by the interest level evaluation program 60 can be realized by, for example, a semiconductor integrated circuit, more specifically, an application specific integrated circuit (ASIC) or the like.

  The content server 12 can be realized by, for example, the computer 80 shown in FIG. The computer 80 includes a CPU 81, a memory 82 as a temporary storage area, and a non-volatile storage unit 83. The computer 80 also includes an input / output I / F 84 such as a display device and an input device, and an R / W unit 85 that controls reading and writing of data with respect to the recording medium 89. The computer 80 also includes a network I / F 86 connected to a network such as the Internet. The CPU 81, the memory 82, the storage unit 83, the input / output I / F 84, the R / W unit 85, and the network I / F 86 are connected to each other via a bus 87.

  The storage unit 83 can be realized by an HDD, SSD, flash memory, or the like. A storage unit 83 serving as a storage medium stores a content providing program 90 for causing the computer 80 to function as the content server 12. Further, in the content storage area 98, content that can be provided to the information processing terminal 16 is stored in advance.

  The function realized by the content providing program 90 can also be realized by, for example, a semiconductor integrated circuit, more specifically, an ASIC or the like.

  Next, the operation of the interest level evaluation system 10 according to the first embodiment will be described. In the interest level evaluation system 10, the information processing terminal 16 receives content from the content server 12. Then, when the received content is displayed on the display unit 22 of the information processing terminal 16 and the user operation detecting unit 24 receives the input of the operation by the user, the information processing terminal 16 performs the interest level evaluation process illustrated in FIG. 9. Executed. Hereinafter, each process will be described in detail.

  In step S100, the user operation detection unit 24 detects an input operation for each unit time in a fixed time (here, 30 seconds). Specifically, the operation information including the type of the input operation, the time of the input operation, the contact position, the operation time of each unit time, and the non-operation time is detected.

  In step S102, the touch state analysis unit 28 detects a touch state indicating whether or not the touch panel is being touched and an operation state corresponding to the operation type of the touch. As operation states, touch start, touch continuation, LEAVE (non-touch state such as touch end), scroll position, and scroll speed are detected.

  In step S104, the terminal momentum detection unit 26 detects the sensor value related to the terminal motion as the terminal momentum at predetermined sample intervals shorter than the unit time for executing the parameter estimation process. The sensor value is a sensor value of any one of a triaxial angular velocity sensor, a triaxial acceleration sensor, and a triaxial geomagnetic sensor, and the predetermined sample interval is, for example, 100 ms.

In step S106, the parameter updating unit 36 and the like update the reliability parameter P _th for evaluating the degree of interest of the operator of the terminal based on the terminal momentum detected in step S104.

  In step S108, the interest level evaluation unit 38 determines the interest based on the scroll speed detected in step S102, the sensor value detected in step S104, the operation time detected in step S100, and the parameter updated in step S106. Calculate the degree.

  The detailed flow of step S106 will be described with reference to FIG.

  In step S200, the first distribution calculation unit 30 determines whether or not the terminal momentum newly detected in step S104 is the required number of samples, and returns to step S104 and repeats the detection until necessary samples are collected. Here, the required number of samples is five.

  In step S202, the second distribution calculation unit 32 obtains a new frequency distribution of the terminal momentum observed in the second period based on the terminal momentum detected in step S102.

  In step S204, the first distribution calculation unit 30 determines whether it is the first process, and if it is the first process, the process proceeds to step S206, and if it is not the first process, the process proceeds to step S208.

In step S206, the first distribution calculation unit 30 sets the median value of the new frequency distribution obtained in step S202 as an initial value of the reliability parameter P _th for evaluating the degree of interest of the operator of the terminal. .. After the setting, the process returns to step S200 to repeat the process.

In step S208, the first distribution calculation unit 30 obtains the cumulative frequency distribution of the terminal momentums observed in the first period based on the terminal momentums detected in step S102. In addition, the first distribution calculation unit 30 determines the median value in the cumulative frequency distribution as the provisional value of the parameter P _th .

  In step S210, the comparison unit 34 determines whether the centroid of the new frequency distribution is lower than the centroid of the cumulative frequency distribution. If it is not below, the process proceeds to step S212, and if it is below, the process proceeds to step S214 on the assumption that a new lower limit change in the terminal momentum has been detected.

In step S212, the parameter updating unit 36 updates the provisional value determined by the median value as the parameter P _th . After the update, the process returns to step S200 to repeat the process.

  In step S214, the comparison unit 34 obtains the centroid distance between the centroid of the cumulative frequency distribution and the centroid of the new frequency distribution.

In step S216, the parameter updating unit 36 corrects the median value of the cumulative frequency distribution according to the above equation (2) based on the centroid distance between the centroid of the cumulative frequency distribution and the centroid of the new frequency distribution to obtain the parameter P _th . Update. After the update, the process returns to step S200 to repeat the process.

[Experimental Results of First Embodiment]

  FIG. 11 shows an example of the experimental result of the method of the first embodiment. The result of the interest degree graph by machine learning in FIG. 11 is a graph in which the degree of interest is evaluated by the parameters obtained by learning the individual characteristics by performing machine learning every 40 seconds by the conventional method. .. Black circles indicate scroll coordinates, and white circles indicate interest level. On the other hand, the result of the interest level graph according to the method of the first embodiment in FIG. 11 is the result of sequentially performing the interest level evaluation while estimating the parameters in real time. In FIG. 11, the results of two patterns of verification are shown on the left and right. When the result of the interest degree graph by machine learning is regarded as the correct answer, the results on the left are almost the same except that there is a difference in the vicinity of 50 seconds. Around 50 seconds is immediately after learning in units of 40 seconds and is considered to be different because learning is insufficient. In the result on the right, the results are almost the same at the place where the degree of interest increased near 34 seconds. This suggests that the learning results are reflected after this time.

  As described above, according to the interest level evaluation system according to the first embodiment, the cumulative frequency distribution of the terminal momentum observed in the first period is obtained based on the detected terminal momentum. A new frequency distribution of the terminal momentum observed in the second period is obtained. By comparing the centroid of the cumulative frequency distribution with the centroid of the new frequency distribution, a new lower limit change in terminal momentum in the cumulative frequency distribution is detected, and the centroid distance is obtained. When a new lower limit change is detected, the median of the cumulative frequency distribution is corrected based on the centroid distance between the centroid of the cumulative frequency distribution and the centroid of the new frequency distribution to update the parameters. The degree of interest is calculated using the updated parameters. This makes it possible to estimate the parameter for evaluating the interest level in real time with respect to the change in the attitude of the terminal and evaluate the interest level.

  A modified example of the first embodiment will be described.

  In the above-described first embodiment, the interest level evaluation system estimates the parameters and performs the interest level evaluation, but the present invention is not limited to this. For example, parameter estimation may be performed by another device. In that case, the parameter estimation system may be configured by a first distribution calculation unit, a second distribution calculation unit, a comparison unit, and a parameter update unit. In this case, the device estimates the parameters by receiving the terminal momentum from the information processing terminal including the internal terminal momentum detection unit or the external terminal momentum detection unit, and transmits the estimated parameters to the device for evaluating the degree of interest. You can do this.

  In the above-described first embodiment, the interest level evaluation system includes the content server and the information processing terminal, but the present invention is not limited to this. For example, a user information storage DB may be provided to store parameter estimated values and interest degrees for each user. In this case, the degree of interest of the user may be transmitted to the content server, and the content delivered from the content server to the information processing terminal may be changed based on the degree of interest of the user.

  In addition, in the first embodiment, the information processing terminal includes a touch state analysis unit, a first distribution calculation unit, a second distribution calculation unit, a comparison unit, a parameter update unit, and an interest level evaluation unit. The case has been described as an example, but the present invention is not limited to this. For example, a user information management server is provided, and the user information management server includes a touch state analysis unit, a first distribution calculation unit, a second distribution calculation unit, a comparison unit, a parameter update unit, and an interest level evaluation unit. You may do it. In this case, the information processing terminal detects the input operation by the user and the terminal exercise amount every unit time. The detected data may be transmitted from the information processing terminal to the user information management server via the network, and the user information management server may analyze the touch state, estimate the parameters, and calculate the degree of interest.

<Second Embodiment>
Next, a second embodiment of the present invention will be described. The above-described first embodiment is a method focused on estimating parameters in real time. In the second embodiment, the parameter estimation method used in the first embodiment is defined as a short-term parameter estimation method for estimating a short-term parameter. Further, the clustering method described in the above background is defined as a long-term parameter estimation method for estimating long-term parameters. The short-term parameter is an example of the first parameter, and the long-term parameter is an example of the second parameter.

  Hereinafter, the premise of the second embodiment will be described.

  In estimating parameters related to the habit of individual operation when obtaining the interest level, there is a problem of wanting to follow posture changes. Examples of scenes in which the posture changes include following posture changes such as sitting down and looking at the smartphone screen, leaning back, leaning forward, holding both elbows on the desk, holding in the hand. There is.

  In the short-term parameter estimation method, if the parameters are estimated before the number of data is sufficient, the estimation accuracy may not be sufficiently secured. In this method, it may happen that the estimation accuracy is not sufficiently secured with a time length of about 10 seconds, so that the parameters cannot be estimated accurately when the posture information frequently changes. Therefore, if an attempt is made to estimate parameters after increasing the time window and obtaining a sufficient number of data, there is a problem that when a posture change occurs, the change cannot be sufficiently tracked. Further, even if the time window for parameter estimation is made small and the posture change is followed, a sufficient number of data for estimation cannot be obtained, which may lead to deterioration of estimation accuracy.

  FIG. 12 is a diagram showing an example of a case where the terminal momentum and the posture are clustered when the posture has not changed. FIG. 13 is a diagram showing an example of clustering the terminal momentum and posture when a posture change occurs. The clustering in FIGS. 12 and 13 is the result of clustering performed by the long-term parameter estimation method. Comparing FIG. 12 and FIG. 13, it can be seen that the clusters determined to be gaze differ due to the change in posture. However, as shown in FIG. 13, it is possible that the two clusters are determined to be one cluster and the estimation accuracy is reduced. Even if there are two clusters, the long-term parameter estimation method determines that the cluster with smaller terminal momentum has high reliability, and the cluster of terminal momentum generated after the posture change is used for estimation. There is a fear that it will not be possible.

  Therefore, in the second embodiment, the method of short-term parameter estimation and the method of long-term parameter estimation are successfully combined in order to achieve both tracking of posture changes and robustness by parameter estimation for a certain length of time. Suggest a method.

  First, a method that is premised on a posture change will be described. A method for estimating the posture change with high accuracy is disclosed in Patent Document 7 and Patent Documents 12 to 14. The technique of Patent Document 7 is a method of autonomous navigation of a moving body using data of a gyro sensor, and a displacement amount in a rotation direction or a translation direction is obtained by matching histograms. In the technique of Patent Document 12, a pedometer is used as an application, and body movement detection that is robust in the wearing direction is performed using an acceleration vector. In the technology of Patent Document 12, for application to a game controller mounted on an arm, body motion and arm rotation motion are separated by using data of an acceleration sensor and a gyro sensor to improve posture estimation accuracy. Patent Document 14 discloses posture estimation in a case where a wearable device is mounted and a passenger is on a transportation facility. In Patent Document 14, the wearable device as a slave and the sensor of a smart device as a master are used in combination to separate the motion of the transportation facility from the human motion.

  It can be said that Patent Documents 12 to 14 are methods that use a sensor together to improve the accuracy of posture estimation. In the target use case, it is characterized by separating another motion that is superimposed on the posture change and becomes a noise component. To put it the other way around, it is an effective method for each use case, and it is difficult to apply it to the interest level evaluation which is the use case of the second embodiment of the present invention. If the methods of Patent Documents 12 to 14 are combined, the accuracy of posture change detection is improved, and the parameter estimation is reset each time the posture is estimated. As a result, the parameter estimation accuracy is reduced.

  In the second embodiment of the present invention, in view of the above points, by using a method for detecting a posture change that matches the purpose of estimating parameters relating to the habit of an individual operation, it is possible to robustly follow the posture change. I examined the method that would become.

  Next, with respect to the method of the second embodiment, the principle of posture change determination and parameter estimation according to the posture change will be described.

  First, the principle of posture change determination in the method of the present embodiment will be described.

  In the interest level evaluation, whether the user is paying attention to the content after the posture change, rather than detecting the change in the posture that is known by the acceleration sensor or the gyro sensor as in the known example. It is important to detect whether or not. Even if there is a change in posture, a transient change in posture without gaze can be ignored. This is because the terminal momentum of the transient posture change can be excluded by the clustering process in the long-term parameter estimation.

  Therefore, in the present embodiment, attention is paid to “whether or not the user is in the gaze state after the posture is changed”, and it is detected whether or not the posture is changed to the gaze state. Further, the trunk movement is quantified by the magnitude of the fluctuation of the posture vector, and the posture change candidate is detected. If the state changes to the gaze state after the posture change candidate is detected, it is determined that the posture change is confirmed.

FIG. 14 is a diagram showing an example of the case where the trunk movement is quantified by the magnitude of the fluctuation of the posture vector. In FIG. 14, the radius drawn by the continuous time series change of the posture vector is shown. The left is the case where the radius is large and the right is the case where the radius is small. The posture vector is a plurality of three-dimensional vectors (gravitational acceleration vector G _acc (n)) based on the acceleration of gravity obtained by the acceleration sensor. The three-dimensional coordinate obtained by averaging a plurality of vectors is defined as the posture center of the posture vector. In FIG. 14, the left posture vector indicates that there is a large fluctuation around the posture center, and the right posture vector indicates a smaller fluctuation than the left posture vector. In the present embodiment, it is determined that the user is gazing when this fluctuation is small, and it is determined that the posture change has been confirmed at the coordinates of the posture center at that time. Further, in the determination of the posture change, it is prevented to follow the posture change unnecessarily by referring to the previously known terminal momentum at the time of gaze.

FIG. 15 is a diagram showing an example of a case where the converted momentum is obtained from each of the posture vectors. The fluctuation of the posture vector can be explained as a converted momentum obtained from each of the posture vectors for each time series. As shown in FIG. 15, the time-series posture vectors are G _acc (n), G _acc (n + 1), G _acc (n + 2), and G _acc (n + 3). Since accelerations other than gravitational accelerations are superimposed on these posture vectors, it can be understood that such a trajectory is drawn. If the user operating the terminal is stationary, it can be said that the superimposed acceleration indicates the acceleration due to the motion held in the hand. Therefore, the acceleration can be obtained by the difference between the posture vectors at the adjacent times. For example, it can be expressed as Δ _acc (n) = G _acc (n) −G _acc (n−1).

  When a certain time is multiplied by the minute time Δt at the adjacent time, it can be regarded as the speed at the time, and therefore the following equation (3) is obtained as the converted momentum using the sum of squares.

Converted momentum = Σ (Δ _acc (k) Δt) ² (3)

  The data of the gravitational acceleration sensor is used as the posture vector for obtaining the converted momentum. Here, the data of the gravitational acceleration sensor is used, but the absolute position, that is, the posture center in FIG. 14 cannot be specified even if the relative change of the posture angle is known from the data of the gyro sensor. Therefore, the posture information cannot be obtained from the data of the gyro sensor, but the data of the gravitational acceleration sensor can be associated with the current posture information. The converted momentum obtained by time integration of gravitational acceleration is considered to be core movement. Therefore, the threshold value set for the converted exercise amount can be appropriately set by investigating the converted exercise amount by the equation (3) and the posture change in advance. On the other hand, in order to determine whether the user is gazing, the terminal momentum acquired by the gyro sensor may be used.

  Next, a method of parameter estimation according to a posture change will be described.

  When both the short-term parameter and the long-term parameter are used together, a section is divided in one cycle according to the cycle defined on the time axis, and which parameter is used for each section is determined. The short-term parameter is used in the section where the estimation accuracy of the short-term parameter is stable, and the long-term parameter is used or the interpolation parameter obtained by interpolating the short-term parameter and the long-term parameter is used in the section where the estimation accuracy of the short-term parameter is not stable. .. How to obtain the interpolation parameter will be described later.

  FIG. 16 is a diagram showing an example of a time chart of parameters used when the posture change is not considered. FIG. 17 is a diagram showing an example of a time chart of parameters used when the posture change is considered.

  FIG. 16 is a time chart showing a basic operation when a short-term parameter and a long-term parameter are used together. In FIG. 16, STE (Short-term estimation) represents a short-term parameter, and LTE (Long-term estimation) represents a long-term parameter. In the example of FIG. 16 and FIG. 17, as a premise, the short-term parameter and the long-term parameter are estimated in a cycle of 40 seconds. The short-term parameter estimation is performed at any time, but the cumulative frequency distribution data is reset every 40 seconds. In long-term parameter estimation, long-term parameters are estimated every 40 seconds. In 40 seconds of the first cycle, the parameter estimation is insufficient for the first 20 seconds, and thus the interest degree evaluation is not performed, and the interest degree evaluation is performed by the short-term parameter in 20-40. After the first cycle, the normal cycle is repeated at intervals of 80 seconds, 120 seconds, and 40 seconds. As shown in FIG. 17, during the normal cycle of 40 to 80 seconds, the parameters are divided into three sections to determine the parameters. When the short-term parameter is reset in 40 seconds, the estimation accuracy of the short-term parameter is not stable after the reset. Use the period parameter. In addition, the interpolation parameter is used because it is considered that the parameter is stable for a short period of time for 10 seconds in the 50-60 second period after the reset. Since the estimation accuracy of the short-term parameter becomes stable again in the section of 60-80 seconds after the reset, the short-term parameter is used.

  In the cycle of every 40 seconds shown in FIG. 16, it is not possible to follow the posture change because the posture change is not taken into consideration. Therefore, in order to adopt a robust estimation result with respect to the posture change, the parameter estimation is performed in a new cycle, triggered by the confirmation of the posture change.

In the time chart of FIG. 17, the behavior is the same as that when the delimiter of the long-term parameter update cycle, which is fixed every 40 seconds in FIG. 16, is replaced with the attitude change determination time. In FIG. 17, T _PC represents the time when the posture change is determined. The short-term parameter estimation is reset at the timing when the posture change is confirmed, and switched to use the long-term parameter estimated immediately before. In this way, the parameter update cycle is separated upon confirmation of the posture change.

  The above is the outline of the parameter estimation of the present embodiment.

  Hereinafter, an example of the configuration of the second embodiment will be described in detail with reference to the drawings. The same parts as those in the first embodiment are designated by the same reference numerals and the description thereof will be omitted.

  As illustrated in FIG. 18, the interest level evaluation system 210 according to the second embodiment includes a content server 12 and an information processing terminal 216.

  The information processing terminal 216 includes a communication unit 18, a control unit 20, a display unit 22, a user operation detection unit 24, a terminal exercise amount detection unit 226, a touch state analysis unit 28, and a terminal attitude detection unit 227. ing. Further, the information processing terminal 216 includes a short-term parameter estimation unit 230, a long-term parameter estimation unit 232, a posture change determination unit 234, a parameter integration unit 236, and an interest level evaluation unit 38.

  FIG. 19 is a block diagram showing an example of the configuration of the short-term parameter estimation unit 230. FIG. 20 is a block diagram showing an example of the configuration of the long-term parameter estimation unit 232. As shown in FIG. 19, the short-term parameter estimation unit 230 includes a first distribution calculation unit 30, a second distribution calculation unit 32, a comparison unit 34, and a parameter update unit 36. As shown in FIG. 20, the long-term parameter estimation unit 232 includes a first exercise amount calculation unit 240, a second exercise amount calculation unit 242, and a parameter determination unit 244.

  The terminal momentum detection unit 226 is the same as in the first embodiment in that the sensor value related to the movement of the terminal is detected as the terminal momentum at predetermined sample intervals. In addition to the gyro sensor used in the first embodiment, a gravitational acceleration sensor is also provided in the second embodiment. The terminal momentum detection unit 226 detects the gravitational acceleration by the gravitational acceleration sensor.

  The terminal attitude detection unit 227 detects the attitude of the terminal for each unit time. The terminal attitude detection unit 227 obtains the attitude from data acquired from, for example, a 9-axis gyro sensor. As the data, the value of the triaxial angular acceleration sensor or the value of the triaxial magnetic sensor is used as the gravity acceleration, the azimuth compass, and the inertial force data. The posture is obtained from the gravitational acceleration in the normal direction (z axis) to the screen of the terminal.

  The short-term parameter estimation unit 230 estimates short-term parameters at predetermined intervals. The predetermined interval is 0.5 seconds, which is the unit time of the first embodiment. The short-term parameter is obtained by the same processing as that of the first distribution calculating unit 30, the second distribution calculating unit 32, the comparing unit 34, and the parameter updating unit 36 of the first embodiment, and detailed description thereof will be omitted.

  The long term parameter estimation unit 232 estimates long term parameters at predetermined intervals. In this embodiment, the predetermined interval is 40 seconds. Here, as the long-term parameter, two long-term parameters, that is, a latest long-term parameter and a long-term parameter having a prescribed time length are estimated. The specified time length may be set as a multiple of the predetermined interval. If the predetermined interval is 40 seconds, it is set to a multiple of 40 seconds, such as 200 seconds. The latest long-term parameter is a long-term parameter estimated using the latest 40 seconds of data. The long-term parameter of the specified time length is a long-term parameter estimated using the data for a maximum of 200 seconds. In this way, not only the latest long-term parameter is estimated with the data for every 40 seconds, but also the long-term parameter of a longer span is estimated.

  Next, each unit of the long-term parameter estimation unit 232 will be described. The processes of the following units may be performed for the latest estimation of the long-term parameter and the estimation of the long-term parameter having a prescribed time length. The former shows the habit of the user who operates the terminal, and the latter shows the degree of the user's gaze.

  The first momentum calculation unit 240 classifies the terminal momentum into a plurality of clusters based on a combination of the terminal momentum regarding the posture of the terminal per unit time and the posture of the terminal per unit time. The first exercise amount calculation unit 240 identifies the cluster corresponding to the gaze state of the content of the operator of the terminal. The first momentum calculation unit 240 calculates the momentum in the identified cluster in the gaze state. In addition, the terminal momentum used by the first momentum calculation unit 240 is a sensor value detected by the terminal momentum detection unit 226 (a sensor of any one of a triaxial angular velocity sensor, a triaxial acceleration sensor, and a triaxial geomagnetic sensor). The sum of squares of (value) is calculated and integrated within a unit time to obtain a logarithmically converted value.

  The gaze state cluster is obtained from the average of the momentum of the terminal when the posture of the terminal is constant. Two-dimensional clustering processing of the posture of the terminal and the momentum of the terminal is performed. In the present embodiment, clusters in the gaze state are classified in consideration of clusters having a small cluster radius.

  Clustering is performed by, for example, cluster analysis based on Euclidean distance on a two-dimensional plane. At this time, the number of clusters is set to 3 or more (for example, 5) and executed, so that outliers (outliers) of the terminal momentum are not particularly included. From each cluster, the terminal momentums of the cluster centroids of the two relevant clusters are calculated in ascending order of the minimum terminal momentum. To classify the clusters that are being gazed, first, two clusters whose center of gravity of the terminal is small are selected as cluster candidates. Next, for the cluster having the smallest terminal momentum of the cluster center of gravity, it is checked whether the terminal momentum of the cluster center of gravity of the cluster is larger than a predetermined threshold value. The threshold value here corresponds to the minimum accuracy of the gyro sensor of the terminal itself. If it is larger than the threshold, the cluster centroid of the cluster is determined as the first momentum. When it is less than the threshold value, the cluster centroid of the second smallest cluster is determined as the first momentum. Since the terminal momentum is stable and the cluster radius is often small in the gaze state, the first momentum can be determined from the one having the smaller cluster radius by such threshold processing.

  The second momentum calculation unit 242 calculates the terminal momentum based on a predetermined operation state of the terminal, and classifies whether the calculated terminal momentum is a terminal momentum that belongs to a non-gaze state for the content of the operator of the terminal. To do. A second momentum is calculated from the classification result. Here, the predetermined operation state is a state in which the touch state of the terminal is non-touch (LEAVE) and the screen of the terminal is scrolling during inertial scrolling. Note that the terminal momentum used by the second momentum calculation unit 242 is a value obtained by integrating and logarithmically converting the sum of squares of the sensor values detected by the terminal momentum detection unit 226 during inertial scrolling.

  The inertial scroll in the second momentum calculation unit 242 is started after detecting the above state, and is ended when the change in the scroll position disappears and the data acquisition is interrupted. Based on the time stamp, the continuation length from the start to the end of inertial scroll is obtained. The second momentum calculation unit 242 calculates the average terminal momentum of the terminal momentum per unit time during the inertial scroll. The second exercise amount calculation unit 242 determines whether or not the average terminal exercise amount is larger than the first exercise amount obtained previously. When it is larger than the first exercise amount, the exercise amount is determined as the second exercise amount. If it is smaller than the first momentum, the value of the terminal momentum of the cluster centroid of the cluster having the larger momentum obtained by the cluster analysis in the first momentum calculation unit 240 is determined as the second momentum.

  The parameter determination unit 244 sets a long-term parameter for evaluating the degree of interest of the operator of the terminal based on the first momentum belonging to the cluster corresponding to the gaze state and the second momentum belonging to the non-gaze state. decide.

In the parameter determination process, the parameter determination unit 244 performs a process of performing exponential calculation and returning to a linear value when logarithmic conversion is performed by the first exercise amount calculation unit 240 and the second exercise amount calculation unit 242 to obtain the terminal exercise amount. To do. As a specific example, threshold parameters P _th and P ₀ for the first amount of exercise and the second amount of exercise are determined as shown in equations (4) and (5) below.

P _th = exp (first momentum * 2 + second momentum) / 3) (4)
P ₀ = exp (second momentum) (5)

Further, when the reliability curve is defined by P _th and the slope Δ, specific examples thereof are defined by the following expressions (6) and (7).

P _th = exp ((first momentum * 2 + second momentum) / 3) (6)
Δ = (exp (second momentum) −P _th ) / 9 (7)

The parameter determination unit 244 performs the conversion processing as described above and determines the thresholds P _th and P ₀ that are long-term parameters regarding the reliability used in the interest level evaluation formula.

  The above is a description of each processing unit of the long-term parameter estimation unit 232.

  The posture change determination unit 234 calculates the converted momentum according to the above equation (3) based on each of the posture vectors obtained from the gravitational acceleration at predetermined intervals. In this embodiment, the predetermined interval is 1 second. When the gravitational acceleration sensor acquires data at intervals of 200 ms, five attitude vectors are collected. The posture change determination unit 234 obtains the converted momentum of the above equation (3) by taking the difference between the five posture vectors in time order.

  Next, the posture change determination unit 234 determines the situation regarding the posture change. The situation regarding the posture change is a normal situation in which no posture change has occurred, a situation in the posture change candidate phase that can be regarded as a posture change, and a situation in which the posture change is confirmed from the posture change candidate phase. To determine the situation, a first threshold value for the converted momentum is set. In addition, the second threshold value regarding the terminal momentum is set. The posture change determination unit 234 determines that there is a situation in which the posture change may occur when the converted exercise amount is larger than a predetermined first threshold, and sets the posture change candidate phase. The posture change determination unit 234 determines that the posture change has been confirmed when the terminal movement amount is smaller than a predetermined second threshold in the posture change candidate phase. The terminal momentum used here may be calculated by the same method as the first momentum calculator 240 of the long-term parameter estimator 232.

  The parameter integration unit 236 determines a parameter used for the interest degree evaluation based on the situation regarding the posture change, and outputs the parameter to the interest degree evaluation unit 38. The parameter integration unit 236 performs processing according to each case depending on whether the posture change has been confirmed or not. When the posture change has not been confirmed, it is divided into three sections during a cycle of 40 seconds, and the parameters are determined in each section. The section of 0-10 seconds is the first section, the section of 10-20 seconds is the second section, and the section of 20-40 seconds is the third section. The parameters to be adopted are the same as the concept described with reference to FIG. The cumulative frequency distribution observed by the short-term parameter estimation unit 230 is reset every 40 seconds. For 10 seconds in the first section of 0-10 seconds after resetting, the long-term parameter estimated at the previous 40 seconds is used. Interpolation parameters are used for 10 seconds in the second section of 10-20 seconds after reset. The third period of 20-40 seconds after reset uses short-term parameters.

  When the posture change is confirmed, the cumulative frequency distribution observed by the short-term parameter estimation unit 230 is reset when the posture change is confirmed. The long-term parameters estimated at the time of the determination of the posture change are used for 10 seconds in the first section of 0-10 seconds from the time of the determination of the posture change. The interpolation parameter is used for 10 seconds in the second section of 10-20 seconds from the time when the posture change is determined. The short-term parameters are used for the third section of 20-40 seconds at the time when the posture change is determined.

  Here, a method of integrating interpolation parameters will be described.

  The interpolation parameter is a parameter obtained by linearly interpolating a long-term parameter and a short-term parameter, as shown in the following equation (8).

... (8)

^ _{P ave} is a short period of time _{parameters, P ave} is a long period of time parameter, _{t e} represents the elapsed time from the start of the most recent cycle.

  The information processing terminal 216 can be realized by, for example, the computer 280 shown in FIG. The computer 280 includes a CPU 91, a memory 92 as a temporary storage area, and a non-volatile storage unit 283. The computer 280 also includes an input / output I / F 94, an R / W unit 95, and a network I / F 96. The CPU 91, the memory 92, the input / output I / F 94, the R / W unit 95, and the network I / F 96 are connected to each other via a bus 97.

  The storage unit 283 can be realized by an HDD, SSD, flash memory, or the like. An interest level evaluation program 260 for causing the computer 280 to function as the information processing terminal 216 is stored in the storage unit 283 as a storage medium. The interest level evaluation program 260 has a communication process 262, a control process 263, a user operation detection process 264, a terminal attitude detection process 265, and a touch state analysis process 266. The interest level evaluation program 260 has a short-term parameter estimation process 267, a long-term parameter estimation process 268, a posture change determination process 269, a parameter integration process 270, and an interest level evaluation process 271.

  The CPU 91 reads the interest level evaluation program 260 from the storage unit 283, expands it in the memory 92, and sequentially executes the processes of the interest level evaluation program 260. The CPU 91 operates as the communication unit 18 illustrated in FIG. 18 by executing the communication process 262. Further, the CPU 91 operates as the control unit 20 shown in FIG. 18 by executing the control process 263. Further, the CPU 91 operates as the user operation detection unit 24 shown in FIG. 18 by executing the user operation detection process 264. Further, the CPU 91 operates as the terminal attitude detection unit 227 shown in FIG. 18 by executing the terminal attitude detection process 265. Further, the CPU 91 operates as the touch state analysis unit 28 shown in FIG. 18 by executing the touch state analysis process 266. Further, the CPU 91 operates as the short-term parameter estimation unit 230 shown in FIG. 18 by executing the short-term parameter estimation process 267. Further, the CPU 91 operates as the long-term parameter estimation unit 232 shown in FIG. 18 by executing the long-term parameter estimation process 268. Further, the CPU 91 operates as the posture change determination unit 234 shown in FIG. 18 by executing the posture change determination process 269. Further, the CPU 91 operates as the parameter integration unit 236 shown in FIG. 18 by executing the parameter integration process 270. The CPU 91 operates as the interest level evaluation unit 38 shown in FIG. 18 by executing the interest level evaluation process 271.

  As a result, the computer 280 that executes the interest level evaluation program 260 functions as the information processing terminal 216. The CPU 91 that executes the program is hardware.

  The function realized by the interest level evaluation program 260 can be realized by, for example, a semiconductor integrated circuit, more specifically, an ASIC or the like.

  Next, the operation of the interest level evaluation system 210 according to the second embodiment will be described. The same parts as those in the first embodiment are designated by the same reference numerals and the description thereof will be omitted. In the interest level evaluation system 210, the information processing terminal 216 receives content from the content server 12. Then, when the received content is displayed on the display unit 22 of the information processing terminal 216 and the user operation detection unit 24 receives the input of the operation by the user, the interest level evaluation process shown in FIG. Executed. Hereinafter, each process will be described in detail. Steps S100 to S104 and S108 are the same as those in the first embodiment. Also, the short-term parameter estimation and the long-term parameter estimation in steps S2102 and S2104 are executed in parallel.

  After step S102, in step S2100, the terminal attitude detection unit 227 detects the attitude of the terminal for each unit time.

  In step S2101, the terminal momentum detection unit 226 detects the gravitational acceleration by the gravitational acceleration sensor.

  In step S2102, the short-term parameter estimation unit 230 estimates short-term parameters at predetermined intervals. The predetermined interval is 0.5 seconds, which is the unit time of the first embodiment.

  In step S2104, the long-term parameter estimation unit 232 estimates long-term parameters at predetermined intervals. The predetermined interval is 40 seconds. The long-term parameters estimated here are two long-term parameters, the most recent long-term parameter and the long-term parameter having a prescribed time length. Note that the latest long-term parameter may be estimated when the posture change is confirmed.

  In step S2106, the posture change determination unit 234 determines the situation regarding the posture change.

  In step S2108, the parameter integration unit 236 determines the parameter used for the interest degree evaluation based on the situation regarding the posture change, and outputs the parameter to the interest degree evaluation unit 38.

  The above step S2102 may be executed similarly to the flow of FIG. 10 of the first embodiment.

  The detailed flow of step S2104 will be described with reference to FIG. As for the long-term parameter, the processing of each of the following steps is executed every 40 seconds so that the latest long-term parameter and the long-term parameter of the specified time length are obtained. The latest long-term parameter is a long-term parameter estimated using the latest 40 seconds of data. The long-term parameter of the specified time length is a long-term parameter estimated using the data for a maximum of 200 seconds.

  In step S2200, the first momentum calculation unit 240 classifies the terminal momentum into a plurality of clusters based on the combination of the terminal momentum regarding the posture of the terminal per unit time and the posture of the terminal per unit time, and gaze state Specify the cluster corresponding to. Then, the first momentum calculation unit 240 calculates the first momentum in the cluster in the gaze state.

  In step S2202, the second momentum calculation unit 242 does not gaze the terminal momentum based on the first momentum belonging to the cluster corresponding to the gaze state obtained in step S2200 and the terminal momentum in the predetermined operation state of the terminal. Categorize whether or not it is in a state. Then, the second exercise amount calculation unit 242 calculates the second exercise amount from the classification result.

  In step S2204, the parameter determining unit 244 evaluates the degree of interest of the operator of the terminal based on the first momentum belonging to the cluster corresponding to the gaze state and the second momentum belonging to the non-gaze state. Determine long-term parameters.

  The detailed flow of step S2106 will be described with reference to FIG. The flow of FIG. 24 is executed at 1 second intervals, which are predetermined intervals of the posture change determination unit 234. Further, the posture change candidate phase at the time of first execution is set to OFF.

  In step S2300, the posture change determination unit 234 determines whether or not it is the posture change candidate phase. If it is the posture change candidate phase, the process proceeds to step S2308, and if it is not the posture change candidate phase, the process proceeds to step S2302.

  In step S2302, the posture change determination unit 234 calculates the converted momentum of the posture vector according to the above equation (3) based on the gravitational acceleration of the terminal momentum.

  In step S2304, the posture change determination unit 234 determines whether the converted exercise amount is larger than a predetermined first threshold value. If it is larger than the first threshold value, the process proceeds to step S2306. If it is smaller than the first threshold value, the process ends. When the converted exercise amount is larger than the first threshold, the user operation detection unit 24 is assumed to be a non-operation section in which the scroll operation is not detected. This is because it is considered that the posture does not change during the gazing and the posture changes in the non-operation section.

  In step S2306, the posture change determination unit 234 sets the posture change candidate phase to ON, sets the value of the phase counter, and ends the process. The value of the phase counter is, for example, set to 10 and is a value that counts negatively and returns the posture change candidate phase to OFF when it becomes 0.

  In step S2308, the posture change determination unit 234 determines whether the terminal exercise amount is smaller than a predetermined second threshold value. If it is smaller than the second threshold value, the process proceeds to step S2310. If it is not smaller than the second threshold value, the process proceeds to step S2312.

  In step S2310, posture change determination section 234 determines that the posture change is confirmed, and proceeds to step S2316.

  In step S2312, the posture change determination unit 234 counts the value of the phase counter as -1.

  In step S2314, the posture change determination unit 234 determines whether or not the phase counter is greater than 0. If the phase counter is greater than 0, the process ends, and if the phase counter is 0, the process proceeds to step S2316.

  In step S2316, posture change determination section 234 sets the posture change candidate phase to OFF and ends the process.

  The detailed flow of step S2108 will be described with reference to FIG. The flow of FIG. 25 is executed after the processing of the posture change determination unit 234 of FIG. In addition, a cycle is set for the first execution, and the cycle is set to 0-40 seconds, 0-10 seconds is the first section, 10-20 seconds is the second section, and 20-40 seconds is the third section. Further, it is assumed that the short-term parameter is given at any time, and the long-term parameter having a prescribed time length is given every 40 seconds.

  In step S2400, the parameter integration unit 236 determines whether the posture change is confirmed in the determination of step S2106. If the posture change is confirmed, the process proceeds to step S2402, and if the posture change is not confirmed, the process proceeds to step S2406.

  In step S2402, the parameter integration unit 236 starts a new cycle triggered by the confirmation of the posture change.

  In step S2404, the cumulative frequency distribution of short-term parameter estimation is reset, the latest long-term parameter estimation result is acquired, and the process proceeds to step S2408. In each of the following steps, when the latest long-term parameter is acquired, the latest long-term parameter is used in the cycle. If not acquired, the long-term parameter with the specified length is used.

  In step S2406, the parameter integration unit 236 determines whether it is the first section of the cycle, the process proceeds to step S2408 if it is the first section, and the process proceeds to step S2410 if it is not the first section.

  In step S2408, the parameter integration unit 236 determines that the first section is a long-term parameter as a parameter used in the interest degree evaluation.

  In step S2410, the parameter integration unit 236 determines whether it is the second section of the cycle, the process proceeds to step S2412 if it is the second section, and the process proceeds to step S2416 assuming that it is the third section if it is not the second section. To do.

  In step S2412, the parameter integration unit 236 calculates the interpolation parameter according to the above equation (8) assuming that it is the second section.

  In step S2414, the parameter integration unit 236 determines the interpolation parameter as the parameter used in the interest degree evaluation.

  In step S2416, the short-term parameter as the third section is determined as the parameter used in the interest level evaluation.

  In step S2418, the parameter integration unit 236 determines whether to reset the cycle. When resetting, the process proceeds to step S2420 and the cycle is reset. If not reset, the process ends. Whether or not the cycle is reset is determined on the condition that 40 seconds have elapsed from the start of the cycle, and is reset when 40 seconds have elapsed, and is not reset when 40 seconds has not elapsed.

[Experimental Results of Second Embodiment]

  FIG. 26 shows an example of the experimental result of the method of the second embodiment. In FIG. 26, the correspondence relationship between the degree of interest and the scroll coordinates in time series is shown. In the result of the interest degree graph by machine learning in the upper part of FIG. 26, the section of the elapsed time of 200 to 250 seconds on the horizontal axis should be detected with high interest degree in consideration of the subject questionnaire, but only low interest degree is detected. The condition continues. In this experiment, the operator closes the browser due to an operation mistake in the section immediately before 200 to 250 seconds, and a situation arises in which the assistant assists. Therefore, the movement of the terminal and the change in the posture are large. In spite of such large terminal movements and posture changes in the learning data, the machine learning method tends to continue the previous parameters assuming that there is no posture change, and the mixed large terminal movements are likely to continue. Has been excluded as noise. On the other hand, the result of the interest level graph by the method of the present embodiment in the lower part of FIG. 26 shows that the interest is high and is correctly detected in the section of the elapsed time of 200 to 250 seconds.

  With respect to each of the above embodiments, the following supplementary notes are further disclosed.

(Appendix 1)
The first distribution of the terminal momentum of the terminals observed in the first period is calculated,
From a terminal momentum corresponding to a predetermined rank in the first distribution, a provisional value of a parameter for evaluating the degree of interest of the operator of the terminal is determined,
A second distribution of the terminal momentum observed in a second period, which is a period after the first period, is obtained,
Comparing the centroid of the first distribution with the centroid of the second distribution,
Correcting the predetermined rank in the first distribution and updating the parameter according to the result of the comparison;
A parameter estimation program for causing a computer to execute processing including the above.

(Appendix 2)
The parameter estimation program according to attachment 1, wherein the predetermined rank is a rank corresponding to a predetermined percentile when the terminal momentums in the first distribution are arranged in ascending or descending order according to the magnitude of the terminal momentum.

(Appendix 3)
In the comparison, a new lower limit change in the terminal momentum in the first distribution is detected,
The supplementary note 1 or the supplementary note 2 that corrects the predetermined rank and updates the parameter based on the center-of-gravity distance between the first distribution and the second distribution when a change that is the new lower limit is detected. Parameter estimation program.

    (Appendix 4)

The center of gravity is the median of the terminal momentum in each of the first distribution and the second distribution,
4. The parameter estimation program according to attachment 3, wherein the new lower limit change is detected when the center of gravity of the second distribution is lower than the center of gravity of the first distribution.

(Appendix 5)
The parameter estimation program according to any one of appendices 1 to 4, wherein the terminal momentum is detected at a sample interval shorter than a unit time within 1 second, and the first distribution and the second distribution are obtained for each unit time. ..

(Appendix 6)
6. The parameter estimation program according to any one of appendices 1 to 5, which evaluates the degree of interest of the operator of the terminal based on the scroll speed of the terminal, the momentum of the terminal, and the updated parameter.

(Appendix 7)
Estimating the updated parameter as the first parameter,
Estimating a second parameter as the parameter, based on a cluster obtained by classifying the terminal momentum based on a combination of information about the posture of the terminal and the terminal momentum,
Based on the converted momentum obtained from each of the posture vector which is a vector related to the gravitational acceleration detected for the terminal, it is determined whether or not a posture change candidate has occurred,
When it is determined to be a posture change candidate, based on the terminal momentum, it is determined whether the posture change is confirmed,
The degree of interest evaluation is performed on the first parameter, the second parameter, or the third parameter obtained by interpolating the first parameter and the second parameter by a predetermined method according to each section determined by the determination of the posture change. 7. The parameter estimation program according to any one of appendices 1 to 6, which is determined as a parameter used for.

(Appendix 8)
The parameters to be determined are
In the first section, which is a section of a predetermined time from the determination of the posture change, the second parameter is adopted,
In the second section which is a section of a predetermined time from the first section, the third parameter is adopted,
8. The parameter estimation program according to appendix 7, which employs the first parameter in a third section that is a section of a predetermined time from the second section.

(Appendix 9)
After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from the cluster that classifies the terminal momentum observed at the latest predetermined interval when the posture change is confirmed is adopted in the determination Note 7 or Note 8. The parameter estimation program described in 8.

(Appendix 10)
A first distribution of the terminal momentums of the terminals observed in the first period is obtained, and the provisional parameters for evaluating the degree of interest of the operator of the terminal from the terminal momentums corresponding to a predetermined rank in the first distribution. A first distribution calculator that determines the value,
A second distribution calculation unit that obtains a second distribution of the terminal momentum observed in a second period that is a period after the first period,
A comparison unit that compares the center of gravity of the first distribution with the center of gravity of the second distribution;
A parameter updating unit that corrects a predetermined rank in the first distribution according to a result of the comparison and updates the parameter;
A parameter estimation device including.

(Appendix 11)
11. The parameter estimation device according to attachment 10, wherein the predetermined rank is a rank corresponding to a predetermined percentile when the terminal momentums in the first distribution are arranged in ascending or descending order according to the magnitude of the terminal momentum.

(Appendix 12)
In the comparison, the comparison unit detects a change that is a new lower limit of the terminal momentum in the first distribution,
Note that, when the parameter update unit detects a change that is the new lower limit, the parameter update unit updates the parameter by correcting the predetermined rank based on a center-of-gravity distance between the first distribution and the second distribution. 10 or the parameter estimation device according to attachment 11.

(Appendix 13)
In the comparison unit, the center of gravity is a median value of the terminal momentum in each of the first distribution and the second distribution,
13. The parameter estimation device according to appendix 12, wherein the new lower limit change is detected when the center of gravity of the second distribution is lower than the center of gravity of the first distribution.

(Appendix 14)
The terminal momentum is detected by the terminal momentum detection unit at a sample interval shorter than a unit time within 1 second, and the first distribution and the second distribution are calculated for each unit time. The parameter estimation device described in 1.

(Appendix 15)
The parameter according to any one of appendices 10 to 14, further including an interest degree evaluation unit that evaluates an interest degree of an operator of the terminal based on the scroll speed of the terminal, the terminal momentum, and the updated parameter. Estimator.

(Appendix 16)
Estimating the updated parameter as the first parameter,
Estimating a second parameter as the parameter, based on a cluster obtained by classifying the terminal momentum based on a combination of information about the posture of the terminal and the terminal momentum,
Based on the converted momentum obtained from each of the posture vector which is a vector related to the gravitational acceleration detected for the terminal, it is determined whether or not a posture change candidate has occurred,
When it is determined to be a posture change candidate, based on the terminal momentum, it is determined whether the posture change is confirmed,
The degree of interest evaluation is performed on the first parameter, the second parameter, or the third parameter obtained by interpolating the first parameter and the second parameter by a predetermined method according to each section determined by the determination of the posture change. 16. The parameter estimation device according to any one of supplementary notes 10 to 15, which is determined as a parameter used for.

(Appendix 17)
After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from a cluster that classifies the terminal momentum observed at a predetermined interval closest to the time when the posture change is confirmed is adopted in the determination Note 16. Parameter estimation device.

(Appendix 18)
After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from a cluster that classifies the terminal momentum observed at a predetermined interval closest to the time when the posture change is confirmed is adopted in the determination Note 16 or Note 17. The parameter estimation device according to item 17.

(Appendix 19)
The first distribution of the terminal momentum of the terminals observed in the first period is calculated,
From a terminal momentum corresponding to a predetermined rank in the first distribution, a provisional value of a parameter for evaluating the degree of interest of the operator of the terminal is determined,
A second distribution of the terminal momentum observed in a second period, which is a period after the first period, is obtained,
Comparing the centroid of the first distribution with the centroid of the second distribution,
Correcting the predetermined rank in the first distribution and updating the parameter according to the result of the comparison;
A method for estimating parameters, characterized in that a computer executes processing including the above.

(Appendix 20)
20. The parameter estimation method according to Note 19, wherein the predetermined rank is a rank corresponding to a predetermined percentile when the terminal momentums in the first distribution are arranged in ascending or descending order according to the magnitude of the terminal momentum.

(Appendix 21)
In the comparison, a new lower limit change in the terminal momentum in the first distribution is detected,
The supplementary note 19 or the supplementary note 20 that corrects the predetermined rank and updates the parameter based on the center-of-gravity distance between the first distribution and the second distribution when a change that is the new lower limit is detected. Parameter estimation method.

(Appendix 22)
The center of gravity is the median of the terminal momentum in each of the first distribution and the second distribution,
22. The parameter estimating method according to appendix 21, wherein the new lower limit change is detected when the center of gravity of the second distribution falls below the center of gravity of the first distribution.

(Appendix 23)
23. The parameter estimation method according to any one of appendices 19 to 22, wherein the terminal momentum is detected at a sample interval shorter than a unit time within 1 second, and the first distribution and the second distribution are obtained for each unit time. ..

(Appendix 24)
24. The parameter estimation method according to any one of appendices 19 to 23, wherein the degree of interest of the operator of the terminal is evaluated based on the scroll speed of the terminal, the terminal momentum, and the updated parameter.

(Appendix 25)
Estimating the updated parameter as the first parameter,
Estimating a second parameter as the parameter, based on a cluster obtained by classifying the terminal momentum based on a combination of information about the posture of the terminal and the terminal momentum,
Based on the converted momentum obtained from each of the posture vector which is a vector related to the gravitational acceleration detected for the terminal, it is determined whether or not a posture change candidate has occurred,
When it is determined to be a posture change candidate, based on the terminal momentum, it is determined whether the posture change is confirmed,
The degree of interest evaluation is performed on the first parameter, the second parameter, or the third parameter obtained by interpolating the first parameter and the second parameter by a predetermined method according to each section determined by the determination of the posture change. 25. The parameter estimation method according to any one of appendixes 19 to 24, which is determined as a parameter used for.

(Appendix 26)
After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from a cluster that classifies the terminal momentum observed at a predetermined interval closest to the time when the posture change is confirmed is adopted in the determination Note 25. Parameter estimation method.

(Appendix 27)
After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from a cluster that classifies the terminal momentum observed at a predetermined interval immediately after the posture change is confirmed is adopted in the determination Note 25 or Note 26. The parameter estimation method according to 26.

10, 210 Interest level evaluation system 12 Content server 14 Network 16, 216 Information processing terminal 18 Communication unit 20 Control unit 22 Display unit 24 User operation detection unit 26, 226 Terminal momentum detection unit 28 Touch state analysis unit 30 First distribution calculation unit 32 second distribution calculating unit 34 comparing unit 36 parameter updating unit 38 interest level evaluating units 51, 81, 91 CPU
52, 82, 92 memory 53, 83, 93 storage unit 59, 89, 99 recording medium 60, 260 interest degree evaluation program 227 terminal posture detection unit 230 short-term parameter estimation unit 232 long-term parameter estimation unit 234 posture change determination unit 236 Parameter integration unit 240 First exercise amount calculation unit 242 Second exercise amount calculation unit 244 Parameter determination unit

Claims (11)

The first distribution of the terminal momentum of the terminals observed in the first period is calculated,
From a terminal momentum corresponding to a predetermined rank in the first distribution, a provisional value of a parameter for evaluating the degree of interest of the operator of the terminal is determined,
A second distribution of the terminal momentum observed in a second period, which is a period after the first period, is obtained,
Comparing the centroid of the first distribution with the centroid of the second distribution,
Correcting the predetermined rank in the first distribution and updating the parameter according to the result of the comparison;
A parameter estimation program for causing a computer to execute processing including the above.   The parameter estimation program according to claim 1, wherein the predetermined rank is a rank corresponding to a predetermined percentile when the terminal momentums in the first distribution are arranged in ascending or descending order according to the magnitude of the terminal momentum. In the comparison, a new lower limit change in the terminal momentum in the first distribution is detected,
3. When the change that is the new lower limit is detected, the predetermined order is corrected and the parameter is updated based on the distance of the center of gravity between the first distribution and the second distribution. Parameter estimation program described in. The center of gravity is the median of the terminal momentum in each of the first distribution and the second distribution,
The parameter estimation program according to claim 3, wherein the new lower limit change is detected when the center of gravity of the second distribution falls below the center of gravity of the first distribution.   The terminal momentum is detected at a sample interval shorter than a unit time within 1 second, and the first distribution and the second distribution are obtained for each unit time. Parameter estimation program.   The parameter estimation program according to claim 1, wherein the degree of interest of the operator of the terminal is evaluated based on the scroll speed of the terminal, the momentum of the terminal, and the updated parameter. Estimating the updated parameter as the first parameter,
Estimating a second parameter as the parameter, based on a cluster obtained by classifying the terminal momentum based on a combination of information about the posture of the terminal and the terminal momentum,
Based on the converted momentum obtained from each of the posture vector which is a vector related to the gravitational acceleration detected for the terminal, it is determined whether or not a posture change candidate has occurred,
When it is determined to be a posture change candidate, based on the terminal momentum, it is determined whether the posture change is confirmed,
According to each section determined by the determination of the posture change, the first parameter, the second parameter, or the third parameter obtained by interpolating the first parameter and the second parameter by a predetermined method is used for the interest degree evaluation. Determined as the parameter to be used,
The parameter estimation program according to any one of claims 1 to 6. The parameters to be determined are
In the first section, which is a section of a predetermined time from the determination of the posture change, the second parameter is adopted,
In the second section which is a section of a predetermined time from the first section, the third parameter is adopted,
The parameter estimation program according to claim 7, wherein the first parameter is adopted in a third section which is a section of a predetermined time from the second section. After the converted exercise amount exceeds a predetermined first threshold value, the posture change is confirmed when the terminal exercise amount falls below a predetermined second threshold value,
When the posture change is confirmed, the first distribution for obtaining the first parameter is reset, and the first parameter after the posture change is confirmed is updated.
When the posture change is confirmed, the second parameter obtained from a cluster that classifies the terminal momentum observed at a predetermined interval closest to the time when the posture change is confirmed is adopted in the determination. The parameter estimation program according to claim 8. A first distribution of the terminal momentums of the terminals observed in the first period is obtained, and the provisional parameters for evaluating the degree of interest of the operator of the terminal from the terminal momentums corresponding to a predetermined rank in the first distribution. A first distribution calculator that determines the value,
A second distribution calculation unit that obtains a second distribution of the terminal momentum observed in a second period that is a period after the first period,
A comparison unit that compares the center of gravity of the first distribution with the center of gravity of the second distribution;
A parameter updating unit that corrects a predetermined rank in the first distribution according to a result of the comparison and updates the parameter;
A parameter estimation device including. The first distribution of the terminal momentum of the terminals observed in the first period is calculated,
From a terminal momentum corresponding to a predetermined rank in the first distribution, a provisional value of a parameter for evaluating the degree of interest of the operator of the terminal is determined,
A second distribution of the terminal momentum observed in a second period, which is a period after the first period, is obtained,
Comparing the centroid of the first distribution with the centroid of the second distribution,
Correcting the predetermined rank in the first distribution and updating the parameter according to the result of the comparison;
A method for estimating parameters, characterized in that a computer executes processing including the above.