JP5225149B2 - Information processing apparatus, information processing system, and information processing method - Google Patents

Description

  The present invention relates to an information processing apparatus, an information processing system, and an information processing method for estimating a user's current state using sensing information, and in particular, probabilistically determines a user's current state from observation values transmitted from sensors asynchronously. It relates to an inference device.

  As a technique for estimating a user's situation / state using observation values notified from a plurality of sensors in a context-aware application, Patent Document 1 discloses that “body temperature” and “ It is described that observation values such as “sweat” and “room temperature” are acquired, and occurrence probability of a psychological state such as “sleepy” and “bad physical condition” corresponding to the combination of the acquired data is acquired from a database. Further, Patent Document 2 acquires an observation value from a sensor belonging to a sensor network, learns a singular pattern from the acquired observation value, identifies a singular pattern from the observation value based on the learning result, and gives a warning. It describes what will be done. In Patent Document 2, for these identification functions, inference using a Bayesian network is performed.

JP 2006-326184 A JP 2008-269215 A

  In Patent Document 1, in order to estimate a psychological state, the corresponding psychological state is acquired from the psychological state estimation table using an event identification information group that specifies an observation value as a key. However, when this method is used, when the probability of occurrence of an event is estimated based on a combination of observation values obtained from various sensors rather than being estimated in cooperation with a specific sensor, When there is a sensor that cannot be acquired, it is necessary to set table records corresponding to a large number of combinations, which is not a realistic estimation method.

  On the other hand, in Patent Document 2, a Bayesian network is used for estimation. In a Bayesian network, if one Bayesian network is constructed, it can be estimated using the same network regardless of which node (sensor) in the Bayesian network has acquired the observation value. For this reason, the problem that a setting item is required for the number of combinations of sensors as in Patent Document 1 does not occur.

  However, when estimation is performed in cooperation with various sensors, it is not only the problem of whether or not information can be acquired from the sensors, but also the asynchrony. Since the sensors normally operate independently of the estimation function and there are many more, it is difficult to install a device that controls these sensors in an integrated manner. Therefore, the observation value from the sensor cannot be obtained at a time, and the information is notified arbitrarily by the sensor.

  On the other hand, the Bayesian network assumes that observations are given at the same time, so if observations are obtained, applying inference by the Bayesian network as it is is obtained unless it happens to be obtained at the same time. The probability of occurrence of an event is estimated from a single observation value, and the estimation accuracy is greatly reduced.

  In view of the above problems, in the present invention, an information processing device that realizes a current state estimation that aggregates sensing information acquired asynchronously by appropriately changing an observation value acquired asynchronously as time elapses. It is an object to provide a processing system and an information processing method.

In order to solve the above-described problem, the present invention provides an observation value acquisition function unit that acquires an observation value notified from a sensor corresponding to a predetermined event, an observation value, and a reception time at which the observation value is acquired. an observation value storage unit for association with each other, based on the difference between the reception time and the current time of the observations, the observation value editing unit for leaving calculate the probability of an event at the current time, the observed value editing unit An information processing apparatus comprising: a probability reasoning function unit that estimates an occurrence probability of an event different from an event based on the calculated occurrence probability .

Further, according to the present invention, the probability inference function unit performs probability inference using a Bayesian network based on the occurrence probability calculated by the observation value editing function unit, and is different from the event in which the observation value acquisition function unit acquires the observation value. It is characterized by estimating the occurrence probability of an event .

The present invention also includes a prior probability storage unit that stores a prior probability indicating an occurrence probability of an event when an observation value is not observed, and an observation value duration storage unit that holds an average time during which the event continues to occur. the provided obs editing function unit includes a difference between the reception time and the current time of the observations, the prior probability, based on the average time, occurrence as over time approach the probability of occurrence of events prior probability Probability is calculated.

In addition, the present invention includes an inference result providing function unit that receives information indicating the notification destination of the occurrence probability estimated by the probability reasoning function unit and notifies the notification destination of the occurrence probability.

  In addition, the present invention includes an inference control unit that notifies the occurrence probability estimated by the probability reasoning function unit to a predetermined notification destination at predetermined intervals.

The present invention is also an information processing system including an application, a sensor, and an information processing device provided between the application and the sensor, wherein the sensor generates an observation value corresponding to a predetermined event. The information processing apparatus, and the information processing apparatus obtains an observation value acquisition function unit that acquires an observation value notified from a sensor corresponding to a predetermined event, an observation value, and a reception time at which the observation value is acquired. an observation value storage unit for association with each other, based on the difference between the reception time and the current time of the observations, the observation value editing unit for leaving calculate the probability of an event at the current time, the observed value editing unit based on the calculated occurrence probability, and a probability inference function section for estimating the occurrence probability of the different events and event, application an occurrence probability estimated by the probabilistic inference function section of the information processing apparatus An information processing system, characterized by Shin.

The present invention is also an information processing method for an information processing apparatus including an observation value storage unit that stores an observation value received from a sensor and a reception time at which the observation value is acquired in association with each other, and a predetermined event The observation value notified from the sensor corresponding to is acquired, and the observation value acquired from the sensor and the reception time at which the observation value was acquired are associated with each other and stored in the observation value storage unit. based on the difference between, calculating the occurrence probability of the event at the current time, based on the calculated occurrence probability is an information processing method and estimates the probability of occurrence of different events and events.

  According to the present invention, in the case of estimating the current state of the user by collecting observation values from a large number of sensors, the observation values are notified even when each sensor generates and notifies an observation value at each opportunity. By generating an appropriate pseudo-observation value from the time and the current time and inferring the occurrence probability using the generated pseudo-observation value, it is possible to estimate the current state with high accuracy using the observation values from sensors that are not synchronized. Thereby, even when the sensor provider exists independently of the information processing apparatus, a context-aware application can be realized.

It is a block diagram which shows the structure of the information processing system of embodiment of this invention. It is a block diagram used for description of embodiment of an information processing system. It is a block diagram used for description of embodiment of an information processing system. It is explanatory drawing of the structural example of a Bayesian network. It is explanatory drawing of a prior probability and an observation value continuation time table. It is explanatory drawing of an acquisition space | interval table. It is a sequence diagram used for operation | movement description of the information processing system of embodiment of this invention. It is explanatory drawing of an observation value table. It is a graph used for description of a time-dependent change of an observed value. It is explanatory drawing of an estimation result table.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. As shown in FIG. 1, the information processing system according to the embodiment of the present invention includes an application 200 and a sensor 300, and a probability inference device 100 between them. Between the application 200 and the sensor 300, the probability reasoning apparatus 100 shares information about what kind of information is used as an observation value (provided information). The observation value is notified to the probability reasoning apparatus 100 at an arbitrary timing. Here, the observed value includes, for example, an event ID to be observed, numerical data observed from the observed object, a user ID to be observed, and the like.

  The probability inference device 100 that has received these observation values stores the reception data together with the reception time in the observation value table (TB) 150. The probabilistic reasoning apparatus 100 infers the non-observed value at an appropriate timing as follows.

  First, the observation value is extracted from the observation value table 150 together with its reception time, and the observation value editing function unit 160 uses the elapsed time from the reception time and the occurrence probability when there is no prior information on the observation target event. Generate a pseudo-observation value indicating the probability. The probability observation function unit 170 receives the pseudo-observation value, probabilistically infers the non-observation value by using a method such as a Bayesian network, and other events (not having the pseudo-observation value) different from the event that generated the pseudo-observation value. Event) occurrence probability, and the estimated occurrence probability is calculated. Accordingly, the application 200 can receive inference results obtained from various sensors 300. The application 200 can also receive notification periodically by setting the acquisition interval by the acquisition interval table (TB) 120.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to FIG. Here, as an example of a service provided by an application, on the premise of the presence of the sensor 300 that identifies “location” and “person who is with”, the service that the user would like to use now is “watch a video”, “game” "I want to play music", "I want to listen to music", and recommends.

  In addition, as shown in FIG. 2, the form with which the several probability inference apparatus 100 exists is also considered. In FIG. 1, the probability reasoning device 100, the application 200, and the sensor 300 are written outside, but as shown in FIG. 3, each of the probability reasoning device 100, the application 200, and the sensor 300 includes other devices. Possible forms are also conceivable. In the present embodiment, the case of FIG. 1 will be described for simplification of description.

  In addition, the system of the present invention originally accommodates a plurality of users, but for the sake of simplification of description, it will be described hereinafter that only one user is accommodated. In an actual system construction, a plurality of users can be accommodated if a user ID is also given to the observation values transmitted and received.

  First, the setting of the probability reasoning apparatus 100 will be described. The probability reasoning apparatus 100 holds a probability reasoning function unit 170, and the probability reasoning function unit 170 constructs a Bayesian network. In this embodiment, in order to estimate which service you want to use from “location” and “people who are with you”, nodes as shown in FIG. 4 are defined, and they are structured as shown in FIG. The link indicates that there is a direct correlation between the nodes). It is assumed that each node only accepts binary values of occurrence / non-occurrence of corresponding events. In other words, the state of each node is expressed only by the probability that the corresponding event occurs.

  These nodes (or corresponding events) are assigned event IDs, and in this embodiment, the event ID is the corresponding event itself. For example, it is assumed that “I want to play a game” and “I want to listen to music” are event IDs. After constructing the Bayesian network, the node list defined in the network is notified to the application 200 and the sensor 300, so that the observed values and the estimated values (pseudo-observed values) are defined as an interface with them.

  Further, the probability inference device 100, after constructing the Bayesian network, in order to cope with asynchronous sensing information acquisition, prior probabilities of each node calculated from the Bayesian network (corresponding events when all node values of the Bayesian network are not observed). (Probability of occurrence) is set in the prior probability table (TB) 161 in the observation value editing function unit 160, and the average time (average duration) during which the corresponding event of each node continues to occur is set as the observation value duration table (TB) 162. Hold on. In FIG. 1, the prior probability table 161 and the observation value duration table 162 are separated, but both are held using the event ID as a key. Therefore, for example, as shown in FIG. It may be held in one table.

  Next, advance preparation of the sensor 300 will be described. The sensor 300 assumed in this embodiment is a GPS and a sensor of a nearby terminal, and these generate observation values such as “home”, “company”, and “with friends”. Assuming that both sensors are built in the mobile phone, the range of latitude and longitude that is “home” or “company” is registered in advance in the application that notifies the observation value based on GPS. In the observation value notification application corresponding to the sensor of the nearby terminal, which terminal ID corresponds to “with friends” is registered. By using such registration information, the mobile phone can generate and notify the observation value from the sensor information.

  Next, the application 200 will be described. In the application 200, in order to determine a recommended service, the inference result providing function unit in the probability inference device 100 makes subscription reservations for three nodes “I want to listen to music”, “I want to play a game”, and “I want to watch a video”. Register via H.110. The subscription reservation includes a subscriber ID, a notification URL, an event ID to be subscribed, and a period (acquisition interval) for confirming a change in status. These information are stored in the acquisition interval table 120 as shown in FIG. Retained. In this example, the subscriber ID, notification URL, subscription event ID, and acquisition interval are configured in the same table, but the database may be normalized and the notification URL column may be in a separate table. .

  The above is the presetting, and the operation at the time of inference will be described below. The outline of the operation sequence of the probability inference apparatus 100 is shown in FIG.

  First, it is assumed that a “company” is identified from a GPS response in a mobile phone that is one of the sensors 300. Upon detecting this, the sensor application in the mobile phone transmits an observation value indicating “company” to the observation value acquisition function unit 140 in the probability inference device 100.

  The observation value acquisition function unit 140 that has received “company” associates the received observation value with the reception time (for example, 10:22:45) (step S1), and as shown in FIG. (Step S2). The observation value acquisition function unit 140 further notifies the inference control function unit 130 that the observation value has been received (step S3).

  Upon receiving the observation value reception notification, the inference control function unit 130 requests the observation value editing function unit 160 for probability inference (step S4) and waits for a response.

  Upon receiving the request, the observation value editing function unit 160 obtains “company”, “10:22:45”, etc. from the observation value table 150 (step S5), and until the current time (21:02:45). Using the elapsed time 38400 seconds and the values of the prior probability table 161 and the observed value duration table 162 (prior probability 0.6, average duration 43200 seconds), the observed value is edited (step S6), and the obtained observation is obtained. The value is transmitted to the probability inference function unit 170 to request inference (step S7).

  In editing the observation value, the observation value received by the observation value acquisition function unit 140 is approximated to the prior probability according to the elapsed time. For example, in the case of “company”, assuming that both occurrence and end of an event follow an exponential distribution, the observed value after t seconds may be calculated as follows.

Then, the following formula (1) is established.

  Therefore, the above equation (1) is solved, and the observed value p (t) after t seconds may be calculated by the following equation (2).

  At this time, if a value calculated from the above equation at an arbitrary time is taken as an observed value, a numerical value almost the same as the prior probability is handled as the observed value. The asymptotic approach of the above equation to the prior probability indicates that the situation will not be observed in the long run, but if it is treated as an observed value, the estimation accuracy decreases. Therefore, the observed value may be generated according to the above equation until the average duration, and thereafter, the observed value may be deleted and treated as having no observed value. The change with time of these observed values is shown in FIG.

  The probability inference function unit 170 that receives the edited observation value calculates the received value as the observation value of the corresponding node (the information in FIG. 8 and the current time (21:02:45)). The occurrence probability is 0.64 and the occurrence probability of the “home” node is 0.96), and the occurrence probability of the remaining nodes is calculated on the Bayesian network by a normal method such as a probability propagation method (step S8). The calculation result is held in the inference result table (TB) 180 as shown in FIG. 10 (step S9), and then the inference completion is notified to the inference control function unit 130 via the observation value editing function unit 160 (step S10). (Step S11).

  Upon receiving the inference completion notification, the inference control function unit 130 requests the inference result providing function unit 110 to provide the inference result to the application 200 (step S12). Upon receiving this request, the inference result providing function unit 110 acquires a notification URL and event ID pair from the acquisition interval table 120 (step S13), and notifies the notification URL of the status of the node (step S14). Here, the recommendation application is notified of “I want to watch a video” with a high probability of occurrence in FIG.

  In response to the notification “I want to watch a video”, if the recommendation of the video service is made to the user as an application for performing the recommendation service in the application 200, the recommendation according to the current situation is realized. In addition, when “friends are together” is notified from the sensor 300, “I want to hear music” is notified to the application through the same process as described above, and the recommended service changes.

  In the above, inference is performed by notification of the observation value from the sensor 300. However, since the observation value used in the inference actually changes with the passage of time, the inference control function unit 130 infers other than when the observation value is acquired. To implement. For example, based on the minimum value of “acquisition interval” in the acquisition interval table 120, inference may be performed if the time elapses from the previous inference, and the application 200 may be notified in the same manner as described above. Even when the event ID is not acquired by the observation value acquisition function unit 140, the observation value editing function unit 160 and the probability inference function unit 170 are operated according to the acquisition interval.

  Moreover, although the observation value received from the sensor 300 is processed assuming that the occurrence probability is 1, the occurrence probability may actually be received together with the event ID. Even in this case, the observation value editing function is not changed.

  Finally, a special value (for example, 0) is prepared for the average duration value in the observed value duration table 162, and when that value is set, the observation value editing function unit 160 is editing the observed value. The value may be determined as a value that does not need to be stochastically changed, and the value in the observation value table may be input to the probability inference function unit 170 as it is.

  As described above, in the embodiment of the present invention, in the case of estimating the current state of a user by collecting observation values from a large number of sensors, each sensor generates and notifies an observation value at each opportunity. However, by appropriately editing observation values from the time when the observation values were reported and the current time, and inferring using the edited ones, it is possible to estimate the current situation with high accuracy using observation values from non-synchronized sensors. To do.

  That is, in the existing probability inference, the observation value itself is not edited for the observation value used for the probability inference of the non-observed value, and only the selection of whether to use or discard the observation value is performed. If the sensor is operated asynchronously, the previous observation value is applied or the non-observation value is inferred. However, since the sensing data changes every moment, if it is handled as it is, the estimation accuracy decreases with time fluctuation.

  Therefore, in the embodiment of the present invention, before inputting the observation value to the inference function, a function is provided in which the observation value is changed stochastically and approximated to the value at the time of non-observation as time changes. Specifically, when an event with an occurrence probability of 0.4 is observed, it is input to the inference function at the time of observation, assuming that the event occurred with a probability of 1. However, as time passes, it is closer to 0.4 The probability of occurrence is input to the inference function. As a result, it is possible to realize an inference that is closer to the current state by aggregating sensing data acquired asynchronously and further emphasizing the latest sensing information.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made without departing from the gist of the present invention.

100 Probability Reasoning Device 110 Inference Result Providing Function 120 Acquisition Interval Table 130 Inference Control Function 140 Observation Value Acquisition Function Unit 150 Observation Value Table 160 Observation Value Editing Function Unit 161 Prior Probability Table 162 Observation Value Duration Table 170 Probability Inference Function 180 Inference Result Table 200 Application 300 Sensor

Claims (7)

An observation value acquisition function unit for acquiring an observation value notified from a sensor corresponding to a predetermined event;
An observation value storage unit that stores the observation value and the reception time at which the observation value is acquired in association with each other;
Based on the difference between the reception time and the current time of the observations, the observation value editing unit for leaving calculate the occurrence probability of the event at the current time,
Based on the occurrence probability calculated by the observation value editing function unit, a probability reasoning function unit that estimates the occurrence probability of an event different from the event;
An information processing apparatus comprising: The probability inference function unit performs probability inference using a Bayesian network based on the occurrence probability calculated by the observation value editing function unit, and the event in which the observation value acquisition function unit acquires the observation value. The information processing apparatus according to claim 1, wherein occurrence probabilities of different events are estimated . A prior probability storage unit for storing a prior probability indicating the occurrence probability of the event when the observed value is unobserved;
An observation value duration storage unit that holds an average time during which the event continues to occur, and
The observed value editing function unit sets the occurrence probability of the event to the prior probability over time based on the difference between the reception time of the observed value and the current time, the prior probability, and the average time. The information processing apparatus according to claim 1, wherein the occurrence probability is calculated so as to be close to. The inference result providing function part which receives the information which shows the notification destination of the said occurrence probability estimated by the said probability reasoning function part , and notifies the said occurrence probability to the said notification destination is provided, The any one of Claim 1 to 3 characterized by the above-mentioned. An information processing apparatus according to claim 1. The inference control part which notifies the said occurrence probability estimated by the said probability reasoning function part to a predetermined notification destination for every predetermined space | interval is provided, The any one of Claim 1 to 4 characterized by the above-mentioned. Information processing device. An information processing system comprising an application, a sensor, and an information processing device provided between the application and the sensor,
The sensor is
Generate an observation value corresponding to the specified event and notify the information processing device,
The information processing apparatus includes:
An observation value acquisition function unit for acquiring an observation value notified from a sensor corresponding to a predetermined event;
An observation value storage unit that stores the observation value and the reception time at which the observation value is acquired in association with each other;
Based on the difference between the reception time and the current time of the observations, the observation value editing unit for leaving calculate the occurrence probability of the event at the current time,
A probability reasoning function unit that estimates an occurrence probability of an event different from the event based on the occurrence probability calculated by the observation value editing function unit;
The application is
An information processing system characterized by receiving the occurrence probability estimated by the probability reasoning function unit of the information processing apparatus. An information processing method of an information processing apparatus including an observation value storage unit that stores an observation value received from a sensor and a reception time at which the observation value is acquired in association with each other,
Obtain the observation value notified from the sensor corresponding to the specified event,
The observation value acquired from the sensor and the reception time at which the observation value was acquired are associated with each other and stored in the observation value storage unit,
Based on the difference between the reception time and the current time of the observed value, to calculate the occurrence probability of the events in the current time,
An information processing method, wherein an occurrence probability of an event different from the event is estimated based on the calculated occurrence probability .

Images