JP2006163911A - User status management device for notification, notification control device, information notifying device, information notifying system, and information notifying method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a user status management device for notification, a notification control device, information notifying device, information notifying system and information notifying method for reducing an unpleasant feeling and a load and prompting a user to acknowledge information. <P>SOLUTION: The information notifying system is achieved by being provided with a user status management device for notification having a user internal status estimation means to estimate a user internal status on the basis of an observable status element, and a user notification effect estimation means to estimate effect for a user who receives information when the information has been notified on the basis of the user internal status; and a notification control device having a notification execution means to determine whether the notification should be made or not on the basis of the effect and notify the information. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a notification user state management device, a notification control device, an information notification device, an information notification system, and an information notification method that perform control related to information notification.

  Currently, various information notification devices such as personal computers, portable terminals such as PDAs (Personal Digital Assistance) and mobile phones, televisions, radios, and car navigation systems are used. Under such circumstances, many methods for controlling information notification based on the type of information, user behavior, surrounding conditions, and the like have been proposed.

  For example, a certain information notification device has a means for adding priority to information and a means for determining importance according to the user's interest, and determines the order of displaying information and the detail level of individual information, Information is presented (see, for example, Patent Document 1). In addition, this information notification device has means for controlling the timing of collected information from the network.

  Another information notification device has means for acquiring user situations from various information sources, adjusting the amount of information, and presenting the information in a timely manner according to the user's behavior. In addition, as a technique for acquiring appropriate information from a large amount of information, a Push type information notification technique that sends information that matches user preferences from the server, or information that matches user preferences is selected from the sent information. There are information filtering technology to be presented, agent technology to collect information from the network according to user's preference, and the like.

  On the other hand, the peripheral vision information notification technology has been actively researched in the field of Attentive User Interface (AUI). In these studies, studies are being conducted with the aim of efficiently displaying messages on the screen edge on the monitor of a desktop PC regardless of the content of the information or depending on the content of the information (for example, Non-Patent Document 1).

  Moreover, examination about the recognition degree of the information at the time of presenting information to such a screen periphery location is also performed (for example, refer nonpatent literature 2).

  As described above, user preference information obtained from a history of explicit settings by the user, user information usage and reaction to information, etc., meta information explaining information contents set in advance, results obtained by analysis, and information By comparing the content of information to be provided candidates, the priority and importance of the provided candidate information can be set to some extent for the user.

  However, the importance of the information and the priority for the user are rarely determined by such static information to some extent, and depend on dynamically changing user information and surrounding information, or the user's emotions It is greatly influenced by. In other words, it may be said that priority and importance are determined only after all such situations are taken into account. Therefore, even if the importance and priority of the information can be determined to some extent, the importance felt by the user greatly varies depending on the information provision method, that is, the provision timing and the presentation method of the provision information.

  With respect to this point, there is an information presentation device that realizes scheduling of information notification by estimating a user's idle time (see, for example, Patent Document 2). However, the user is directed to spend time for acquiring information even when the user is free. Further, such information notification has the same element as that for acquiring attention by ringtone or vibration regardless of the current user situation in the current mobile phone call. In other words, presenting in the user's idle time is not necessarily the same as the current call to the mobile phone, but since it is generally not known what the user is paying attention to at each time, It is difficult to grasp the user's free time in a complete sense.

  In this state, when attention is forcibly acquired from a location different from the location of interest of the user, it is accepted when the information content is important for the user. However, since it is difficult to set the importance completely, it is highly expected that the result will cause discomfort and load on the user and the service will not be used.

  Furthermore, as with a complete push-type information notification service, not only the information notification method that is important to the user, but also to some extent, information that differs from the user's preference is less likely to cause discomfort and burden on the user regardless of the content of the information. The technology that can be presented is important. Such a request is particularly high as a request from the information provider side who distributes advertisements and the like.

Regarding such a problem, there is a peripheral vision information providing technique for providing information on a peripheral portion of a screen as a technique for providing information on a display of a personal computer while considering a load on a user.
Japanese Patent Laid-Open No. 9-81099 JP 2000-99441 A “Attentive User Interface”, ACM Communications of the ACM, March, 2003, Vol. 46, No. 3, pp. 31-72 D. Scott McCrickard, Richard Catrambone, C.C. M. Chewar, John T. Stasko, “Establishing tradeoffs that leverage attention for utility: empirically evaluating information display in notification systems,” International Journal of Human-Computer Studies, v. 58 n. 5, p. 547-582, May 2003 “Multi-Sensor Activity Context Detection for Wearable Computing”, Nicky Kern and Bernt Schiele, In European Symposium on Ambient Intelligence (EUSAI), Eindhoven, The Netherlands, November 2003

  However, the background art described above has the following problems.

  Many information notification methods for controlling the notification timing, the amount of information, the display method, and the like in consideration of the type of information, user behavior, surrounding conditions, and the like have been proposed. However, they do not take into account the human internal state elements that change dynamically and uncertainly, and the control rules based on these element elements are also static and lack flexibility.

  Peripheral vision information provision technology assumes information notification when the user is sitting in front of a personal computer, and considers information notification that takes into account the state that can change dynamically around the user or the user himself / herself There are problems that are not.

  Therefore, these techniques only control by a static notification method according to provided information in consideration of the degree of task interruption due to information notification.

  In addition, there is a problem that information notification to a user who is in an environment where the surrounding situation changes dynamically, such as a portable display terminal, has not been studied.

  On the other hand, for such an information notification method, an evaluation experiment is performed on the degree of recognition of information displayed in peripheral vision. However, considering how much information is recognized based on these information, a method for performing dynamic control has not been studied.

  Therefore, the consistency of the provided information content with the user preference is not emphasized, and the user status that changes dynamically is taken into consideration, and any type of content is surely recognized by the user without any discomfort or load on the user. An information notification method that can be urged is necessary.

  Accordingly, an object of the present invention is to provide a notification user state management device, a notification control device, an information notification device, an information notification system, and an information notification method that can reduce discomfort and load and can prompt the user to recognize information. It is to provide.

  In order to solve the above problem, the user state management device for notification according to the present invention provides a user internal state estimation means for estimating a user internal state based on an observable state element, and information based on the user internal state. User notification utility estimation means for estimating utility to a user who receives information in the case of notification.

  With this configuration, the internal state can be estimated from the observable state elements, for example, the user operation state and the task state, and the effect of information notification can be estimated based on this estimation.

  In addition, the notification control apparatus of the present invention determines whether to notify information based on the utility to the user who receives the information when notifying the information, and performs notification notification notifying the information based on the determination. Means.

  With this configuration, information can be effectively notified to the user.

  The information notification device of the present invention includes the above-described notification user state management device and notification control device.

  By configuring in this way, the internal state can be estimated from the observable state elements, for example, the user operation state and the task state, and the effect of the information notification can be estimated based on the estimation, so that the user can be effectively Information can be notified.

  The information notification system of the present invention includes the above-described notification user state management device and notification control device.

  By configuring in this way, the internal state can be estimated from the observable state elements, for example, the user operation state and the task state, and the effect of the information notification can be estimated based on the estimation, so that the user can be effectively Information can be notified.

  In addition, the information notification method according to the present invention includes a step of estimating a user internal state based on an observable state element, and a user receiving information when notifying the information based on the user internal state. There are a step of estimating the utility, a step of determining whether to notify information based on the previous utility, and a step of notifying information based on the determination.

  By doing so, the internal state is estimated from the observable state element, and the effect of the information notification can be estimated based on the estimation, so that the information can be effectively notified to the user.

  According to an embodiment of the present invention, a user state management device for notification, a notification control device, an information notification device, an information notification system, and an information notification method capable of reducing discomfort and load and prompting a user to recognize information Can be realized.

Next, embodiments of the present invention will be described with reference to the drawings.
In all the drawings for explaining the embodiments, the same reference numerals are used for those having the same function, and repeated explanation is omitted.

  The information notification system according to the embodiment of the present invention takes into account two user internal states of information recognition by the user in executing information notification and interruption of the current user task by providing information. Control the presentation method. In other words, for users living in the real world where various information display terminals are scattered, and for users who have information display terminals, the user's information recognition is high and the user's task interruption is low. , Adjust the information notification timing and information expression, and notify the information.

  This suppresses an increase in user discomfort and an increase in load due to interruption / interruption of tasks by real-world users in proactive information notification, which did not occur in conventional user-driven (request) type information notification It becomes possible.

  Specifically, the user's attention state on the information display terminal is detected, and the task on the information terminal, the user's task proficiency level, the user operation state, the information content (information urgency level, information amount), information recognition speed, etc. Consider information in consideration. Such an information notification is not a sudden one-way pressing information notification that does not consider the user state, but an information notification state that considers the user's information request state, in other words, the acceptance state. Then, by adjusting the information notification timing and the information expression / display method, it is possible to perform information notification while improving information recognition and keeping the task interruption level low.

  As a format for presenting information to the user, any format such as text, image, video, and the like can be considered.

  In addition, the information notification system according to the present embodiment includes an arbitrary information notification device capable of presenting information, for example, a display type device. For example, a desktop PC, a television, a digital signboard, an information window, a notebook personal computer, a PDA (Personal Digital Assistant), a handheld terminal, a mobile phone, a wristwatch, a future notebook display, and the like are included as the information notification device.

  Next, the information notification system according to the present embodiment will be described with reference to FIG.

  The information notification system according to the present embodiment includes an information notification device 100.

  The information notification device 100 includes a notification user state management unit 110 as a notification user state management device expressing a user state related to information notification, a user interface unit 130 connected to the notification user state management unit 110, and an environmental user state. A management unit 140, a notification control unit 120 as a notification control device, and a sense unit 150 connected to the environment user state management unit 140 are provided.

  The user state management unit 110 for notification includes a user internal state estimation unit 112 as a user internal state estimation unit, a user state model update unit 115 as a user state model update unit connected to the user internal state estimation unit 112, and a user notification. User notification utility deriving unit 114 as utility estimation means, information notification state history storage unit 116 as information storage means connected to user state model update unit 115, and observability connected to information notification state history storage unit 116 A model accuracy improvement notification control unit 118 as state element determination means.

  The notification control unit 120 includes a notification information receiving unit 122, a notification information storage unit 124 connected to the notification information reception unit 122, a notification information storage unit 124, a user notification utility derivation unit 114, and a model accuracy improvement notification control unit 118. And a notification execution unit 126 connected to.

  The user interface unit 130 includes a feedback information collection unit 132 connected to the information notification state history storage unit 116 and a notification information display control unit 134 connected to the notification execution unit 126.

  The environmental user state management unit 140 includes an environmental user state generation unit 142 connected to the user internal state estimation unit 112 and the model accuracy improvement notification control unit 118, and a sensor connected to the environmental user state generation unit 142 and the sense unit 150. An information receiving unit 144, an environmental user state generation unit 142, a user internal state estimation unit 112, a model accuracy improvement notification control unit 118, and an environmental user state storage unit 146 connected to the information notification state history storage unit 116 are provided.

  Each part which comprises the information notification apparatus 100 mentioned above is comprised by a module, for example.

  First, each unit constituting the notification user state management unit 110 will be described.

  The notification user state management unit 110 can acquire the two internal state elements “information recognition degree” and “task interruption degree” which are human internal state elements, and can be acquired by the sense unit 150. Conditional probability of the relationship between observable state elements such as “gaze state”, “user operation state”, “task state”, etc. that have a causal relationship with “information expression” elements that determine how information is displayed at the time of notification Manage dynamically with expressions.

  The user internal state estimation unit 112 is based on the user state information input from the environmental user state management unit 140 and the environmental state information such as the state of the information display terminal, and the user internal state to be directly considered at the time of information notification and the related The user status and the status of the information display terminal are managed in association with each other. In addition, the user internal state estimation unit 112 predicts (estimates) and manages the user internal state from the relevance between the user internal state elements that are directly considered when notifying information and the user state elements related thereto. That is, the internal state is estimated from the observable state element. In particular, regarding the internal state, the degree of task interruption and information recognition are predicted from changes in tasks that the user is performing in addition to information notification, such as tasks such as Web browsing and mail operations.

  The user notification utility deriving unit 114 estimates the degree of utility (effect) to the user by information notification based on the relationship between the states. For example, a utility value indicating the degree of utility (effect) to the user by information notification is derived. That is, the utility value obtained by providing information at each time is derived from two elements that have a trade-off relationship between task interruption and information recognition. In addition, the user notification utility deriving unit 114 determines the optimal notification timing and “information expression”.

  The user state model updating unit 115 realizes updating of model information itself that provides a basic structure for managing user states. That is, the user state model update unit 115 learns a random variable indicating a relationship between state elements based on accumulated information accumulated in an information notification state history accumulation unit 116 described later, and dynamically changes and sequentially updates the state variable. By doing so, the internal state estimation accuracy is improved.

  The information notification state history accumulation unit 116 accumulates and manages the history of the state in which the information notification is performed and the state fed back from the user. For example, the feedback information from the user input from the feedback information collection unit 132 and the state element value at the time when the information notification is performed from the environmental user state storage unit 146 are stored and managed. Further, the information notification state history accumulation unit 116 may accumulate and manage the history of the state and the state fed back from the user even when the information notification is not performed.

  The model accuracy improvement notification control unit 118 detects a combination of state elements having a low occurrence frequency based on the accumulated information accumulated in the information notification state history accumulation unit 116, and monitors whether the combination occurs. When this occurs, a request for information notification is made to the notification execution unit 126. By doing so, the random variable indicating the relationship between the state elements and the entire structure are actively updated. That is, the model accuracy improvement notification control unit 118 positively generates an event for deriving a relationship that cannot be compensated by the user state model update unit 115, thereby improving the accuracy of the inter-element relationship.

  Next, each part which comprises the notification control part 120 is demonstrated.

  The notification information receiving unit 122 receives notification information provided from an information provider, for example, a service provider. The notification information storage unit 124 stores the notification information received by the notification information reception unit 122.

  The notification execution unit 126 determines whether to notify based on an expected utility value (to be described later) obtained from the user notification utility deriving unit 114, and actually notifies the information based on the determination. For example, the notification execution unit 126 finally notifies the information notification timing and the information expression method in the information notification.

  Next, each part which comprises the environmental user state management part 140 is demonstrated.

  The sensor information receiving unit 144 receives sense information such as sensor information and interaction information from the sense unit 150. The environmental user status generation unit 142 converts, interprets, and creates sense information, and generates environmental user status information including at least one of user status information and environmental status information used by the notification user status management unit 110. For example, the environment user state generation unit 142 performs identification processing such as preprocessing, normalization processing, feature extraction processing, and clustering on the received sense information. For example, environmental state information is created by processing sensor information, user information is created by processing interaction information, and environmental user state information is created.

  The environmental user status storage unit 146 stores and manages the created environmental status information and user status information together with time information.

  Next, each part which comprises the user interface part 130 is demonstrated.

  The notification information display control unit 134 presents notification information in accordance with an instruction from the notification execution unit 126. For example, the notification information is notified to the user based on the timing of the information notification input from the notification execution unit 126 and the information expression method in the information notification.

  The feedback information collection unit 132 acquires the interaction information from the user and inputs it to the information notification state history storage unit 116. Also, the feedback information collection unit 132 acquires interface information and inputs it to the information notification state history storage unit 116.

  Next, information notification processing of the entire information notification system realized by the information notification apparatus 100 according to the present embodiment will be described with reference to FIG.

  Here, two types of information notification processing, that is, inquiry type processing and event type processing will be described.

  First, the inquiry type process will be described with reference to FIG.

  In the inquiry type process, the notification execution unit 126 inquires the user notification utility deriving unit 114 about the expected utility value at each time, and performs information notification based on the notified expected utility value.

  The sensor information receiving unit 144 periodically collects sense information from the sense unit 150 (step S202) and inputs it to the environmental user state generation unit 142. The environmental user status generation unit 142 creates environmental user status information indicating the status of the environment and the user (step S204). The sense information includes information output from all devices such as an information display terminal used by the user and a device embedded in the environment. For example, it includes interaction information on the information display terminal of the user, sense information of the acceleration sensor, video information from a surveillance camera installed in the room, and the like.

  The environment user state generation unit 142 inputs the created environment user state information to the user internal state estimation unit 112 (step S206).

  On the other hand, the service provider or the information receiving user himself / herself inputs notification information to be notified to the information receiving user, such as content, to the notification information receiving unit 122 as needed (step S208). The notification information receiving unit 122 inputs the input notification information content to the notification information storage unit 124.

  Here, the notification execution unit 126 refers to the notification information storage unit 124 and inquires the user notification utility deriving unit 114 about the current expected utility value when there is notification information content to be notified (step S210). ).

  When there is an inquiry, the user notification utility deriving unit 114 calculates an expected utility value and its information expression index based on the environmental user state information input from the environmental user state generation unit 142 to the user internal state estimation unit 112 (step S212). The expected utility value and its information expression index are returned to the notification execution unit 126 (step S214).

  The notification execution unit 126 determines whether to actually notify based on the received expected utility value (step S216). When notifying, the notification information content and the information expression index are transmitted to the notification information display control unit 134 (step S218), and the notification information display control unit 134 is based on the display method indicated by the information expression index. It is displayed (step S220).

  After the information is displayed (step S222), the feedback information collection unit 132 collects the interaction information fed back from the user who has received the information (step S224) and inputs it as feedback information to the information notification state history storage unit 116 (step S226). ). In this way, not only the information notification state history can be accumulated, but also the notification user state management model can be updated from the accumulated information.

  Next, event type processing will be described with reference to FIG.

  In the event type processing, the notification execution unit 126 first notifies the user notification utility deriving unit 114 of the reference value of the utility value in advance. Next, the user notification utility deriving unit 114 notifies the notification execution unit 126 only when the notified reference value is satisfied. Next, the notification execution unit 126 performs information notification based on the notification that the reference value is satisfied.

  The event type process is basically the same as the inquiry type process, but only the trigger part of the information notification is different. In the inquiry type process, the user notification utility deriving unit 114 returns an expected utility value in response to an inquiry from the notification execution unit 126, and the notification execution unit 126 performs information notification based on the returned expected utility value.

  In the event type processing, the notification execution unit 126 notifies the user notification utility deriving unit 114 in advance of the reference value of the utility value, and the user notification utility deriving unit 114 always calculates the expected utility value and monitors the case where the reference value is satisfied. . When the case where the reference value is satisfied occurs, the user notification utility deriving unit 114 transmits the information expression index to the notification execution unit 126.

  The event type processing will be specifically described.

  The sensor information reception unit 144 periodically collects sense information from the sense unit 150 (step S302) and inputs the sense information to the environment user state generation unit 142. The environment user state generation unit 142 creates environment user state information indicating the environment and the user state, and inputs the created environment user state information to the user internal state estimation unit 112 (step S304).

  On the other hand, the service provider or the information receiving user himself / herself inputs the notification information content notified to the information receiving user to the notification information receiving unit 122 as needed (step S306). The notification information receiving unit 122 inputs the input notification information content to the notification information storage unit 124.

  On the other hand, the notification execution unit 126 inputs the reference value of the utility value to the user notification utility deriving unit 114. The user notification utility deriving unit 114 sets a reference value (step S310), always calculates an expected utility value, and determines whether the reference value is satisfied.

  When the calculated expected utility value satisfies the reference value, the user notification utility deriving unit 114 inputs the information expression index to the notification execution unit 126 (step S312).

  The notification execution unit 126 transmits the notification information content and the information expression index to the notification information display control unit 134 (step S314), and the notification information display control unit 134 uses the display method indicated by the information expression index. Based on the display (step S316).

  After the information is displayed (step S318), the feedback information collection unit 132 collects the interaction information fed back from the user who has received the information (step S320), and inputs it to the information notification state history storage unit 116 as feedback information (step S322). ). In this way, not only the information notification state history can be accumulated, but also the notification user state management model can be updated from the accumulated information.

  Next, the operation of each unit constituting the information notification apparatus 100 will be described in detail.

  First, the operation of each unit of the notification control unit 120 will be described.

  The notification information receiving unit 122 receives information from various information sources.

  Information sources include e-mail, electronic news, e-mail magazine, pager distribution system, radio teletext, WWW content, and instant message (IM). Also included are a scheduler registered by the user, a To-Do list, and the like.

  Types of information include news (politics, economy, society, culture, sports, weather), movie information, gourmet information, traffic information, TV programs, stock information, weather, fortune, event information, learning information (English word memorization service) ), Emails from acquaintances, communications from companies, emergency communications such as disasters, advertisements, menu services, lost items services, lost items services, health promotion information, and last train information.

  The notification information accumulation unit 124 accumulates and manages information received by the notification information reception unit 122. Further, the notification information storage unit 124 responds to the inquiry from the notification execution unit 126 and returns the stored information, the number of information, and meta information of the stored information, for example, valid time information.

  The notification execution unit 126 determines whether there is information to be notified to the notification information storage unit 124. If information exists, the notification execution unit 126 determines notification based on the expected utility value. Here, the notification execution unit 126 determines the information to be notified based on the order in which information is registered in the notification information storage unit 124 or the priority according to the content. For example, the information to be notified is determined based on the contents of the information to be notified, for example, the importance level and the urgency level.

  In the present embodiment, a case will be described in which the order is determined according to the effective time of information without paying attention to the importance and priority according to the contents. Here, the effective time of the information is information that depends on the time determined objectively, such as the time of the schedule, the end time of the advertisement content, and the time indicating the freshness of the news. In this embodiment, a case will be described in which these pieces of time information are added as notification information addition (side) information as meta information by the notification content provider.

  The notification execution unit 126 inquires of the user notification utility deriving unit 114 about the expected utility value, and determines the notification by determining the obtained expected utility value based on the threshold value.

  Further, the notification execution unit 126 inputs the reference value of the utility value to the user notification utility derivation unit 114, and causes the user notification utility derivation unit 114 to determine whether or not the expected utility value satisfies the reference value. Good. In this case, the user notification utility deriving unit 114 inputs the information expression index to the notification execution unit 126 when determining that the expected utility value satisfies the reference value.

  Further, the notification execution unit 126 may control to transmit as much information as possible while reducing the burden on the user, based on the information amount of the notification information, for example, the notification information content.

  Further, the notification execution unit 126 may be controlled to notify the information to the user's gaze position based on the user task.

  When the notification execution unit 126 determines to notify based on the expected utility value, the notification execution unit 126 inputs the notification information content to the notification information display control unit 134. For example, information such as media type, display position / size, text presentation method, and information update frequency (speed) is input as information representation.

  Next, the operation of each unit constituting the notification user state management unit 110 will be described.

  The notification user state management unit 110 manages mainly two elements of “information recognition degree” and “task interruption degree” to be considered in information notification. “Information recognition” refers to the degree to which a user perceives and recognizes the content of information. The “task interruption level” indicates the degree to which the work (task) currently being performed by the user is interrupted by the information notification. These two elements are relatively low-level elements in the human information processing process, and are less dependent on the information content.

  However, since these are elements that cannot be observed directly, elements that can be directly observed that have an influence on both elements are introduced, and the causal relationship with these elements is estimated. Specifically, the user internal state estimation unit 112 introduces at least one element out of the user operation state element and the task state element, and includes two elements from the introduced element and the information expression element to be controlled: That is, the “information recognition degree” and the “task interruption degree” are estimated.

  Here, Bayesian networks are suitable for handling events that cannot be completely observed, such as the user's internal intentions, or events that contain a lot of uncertain information, such as events that contain a lot of weather and noise signals. (Belief / probability / causal network) can be used.

  The Bayesian network is a model in which qualitative dependency relationships between a plurality of random variables are expressed by a graph structure, and a quantitative relationship between individual variables is expressed by the above conditional probability. That is, the Bayesian network is a probability model that is presented by a random variable, a conditional dependency between random variables, and a conditional probability. A variable is a node, and a dependency relationship between variables is represented by an effective link extending from the cause to the resulting variable. A node that comes before the link is called a child node, and a node that is the link source is called a parent node.

  The user internal state estimation unit 112 uses a Bayesian network to manage the unobservable elements and the dynamic relationship between the observable elements by using a random variable between the elements. Probabilistic prediction of one element. By using a Bayesian network, it is possible to handle a phenomenon that cannot be completely observed inside the user and effects from factors that are not considered, such as noise, in a unified and robust manner under the statistical probability method. Become. In addition, by using a Bayesian network, it is possible to deal with latent elements that cannot be observed directly or new elements that appear in the future.

  A causal relationship between two non-observed user state elements, that is, “information recognition degree” and “task interruption degree”, and other elements is expressed using a Bayesian network. Specifically, for example, at least one of the task state and the user operation state on the information display terminal and the information expression are causal factors, and a causal network for the information recognition degree and the task interruption degree is generated. An example of a Bayesian network is shown in FIG. However, the links (arrows) between the elements are merely examples, and are dynamically changed according to the causal relationship strength.

  Next, elements constituting the Bayesian network will be described with reference to FIG.

  The “user operation state” element indicates the operation state of the user. For example, “Staying”, “Walking”, “Running”, “Climbing stairs”, “Stairs” Discrete variable values such as “getting off”. As a method for acquiring these gaze states, a method based on image recognition using an external camera, a method of analyzing information from a wearable sensor, and the like can be applied (for example, see Non-Patent Document 3).

  The “task state” element indicates the work state of the user on the information display terminal. For example, “I do nothing”, “I am watching the information display terminal but do nothing”, “Web browsing”, “Editing E-Mail”, “Dictionary search”, “Image reference”, “Terminal settings” Discrete variable values such as “streaming appreciation” and “document reference” are included. These can be acquired from an OS (Operating System) on the information display terminal or application status information.

  However, “you are watching the information display terminal but are not doing anything” cannot be acquired from the OS or application status information on the information display terminal. For this reason, whether or not the user's line of sight exists on the information display terminal, that is, the gaze state, is a method by image recognition using an external camera, a method in which the user mounts a head mounted display (HMD), etc. A method of analyzing information from the sensor that has been used can be applied.

  The “information expression” element indicates an information expression / display method on the information display terminal. For example, it indicates media types such as visual media such as text, animation, and video, audio media, and vibration media. As for text media, there is a distinction between display position / size, text presentation method, and information update frequency (speed). As shown in FIG. 5, for the display position and size, for example, there are a display on the entire screen and a display on the lower part of the screen and a horizontal part of the screen, and the presentation method is a scroll display in the horizontal direction and the vertical direction, There is a switching type. Also, in scrolling and switching, there are high-speed display and low-speed display as the frequency (speed). These are controlled by a notification information display control unit 134 described later.

  When the “user operation state” element and the “task state” element as the observable state elements as described above are obtained, the user internal state estimation unit 112, the observable state element, the “information recognition degree” element, Probability calculation is performed based on a Bayesian network composed of “task interruption level” elements. By doing so, it is possible to calculate the probabilities of the “information recognition degree” element and the “task interruption degree” element.

  Further, in the information notification in the real world, when the user internal state estimation unit 112 determines that the user is paying attention to the information display terminal as in the mobile environment, the task interruption property is relatively low, and the information recognition property May be predicted to be high.

  The user internal state estimation unit 112, in addition to the degree of attention to the information display terminal, the state of the user and the information display terminal, the work (task) state that the user is performing on the information display terminal, and the user for those tasks Based on the proficiency level, the task interruption element and the information recognition element may be obtained.

  Also, the user internal state estimation unit 112 sets arbitrary values for the “information recognition degree” element and the “task interruption degree” element, and reverses at least one of the observable state value and the combination that can satisfy these values. The probability calculation of the “information recognition degree” element and the “task interruption degree” element may be performed depending on whether the estimated and inversely estimated value matches the observable state element.

  Here, learning by model parameter estimation for creating these structural models requires offline learning in advance. It is also possible to learn online. The latter will be described later. Off-line learning is realized by creating a situation in which all combinations of discrete variables that can be taken by the elements that make up the model occur and observing objective information. The model structure is determined using, for example, AIC (Akaike's information criterion), MDL (Minimum description length), and BIC (Bayesian information criterion).

  Next, the user notification utility deriving unit 114 will be described.

  The user notification utility deriving unit 114 calculates an expected utility value expected when notifying the user of information at the timing from the probability of the information recognition factor element and the probability of the task interruption factor. The user notification utility deriving unit 114 may provide an expected utility value at each time in response to a request from the notification execution unit 126 (inquiry type method), or a conditional expression given in advance, for example, a reference value When it is satisfied, information indicating that the reference value is satisfied may be provided (event type method).

  In the present embodiment, a description will be given of a case in which the effective time of information is given in advance by the information provider as the delivery timing information of the notification information in deriving the expected utility value. For example, in the case of sale information, the end time thereof, in the case of waiting information, in the waiting time, in the case of notification information informing that a lost item has been dropped, the time of the moment. The amount (length) of information is also described in the delivery timing information or can be acquired by P-parsing.

  Next, an example of a derivation method for obtaining the expected utility value from the delivery timing information and the above two elements, that is, the “information recognition degree” element and the “task interruption degree” element will be described with reference to FIG.

  Specifically, information recognition probability P (InfoRec | E) (where InfoRec∈ {infoRecHigh, InfoRecLow}), task interruption probability P (Inter | E) (where Inter∈ {InterHigh, Interlow}) The utility value given by information recognition is defined as U (InfoRec), and the utility value given by task interruption is defined as U (Inter).

From the above, the expected utility value (EU) is given by the following equation (1). Here, E is a collection of evidence that can be observed when information is notified, and E = {information expression, task state, user operation state} = {InfoRep, Task, Motion}. Therefore, P (InfoRec | E) is exactly P (InfoRec | InfoRep, Task, Motion) and P (Inter | InfoRep, Task, Motion). Here, W _InfoRec and W _inter are weighting factors of information recognition degree and task interruption degree, respectively.

In the following, it is assumed that the time index is t and the index of the notification information presented at the time index t is m. In this case, when t = t + 1, notification information of m + 1 is notified.

  Next, a specific example of the expected utility value EU will be described.

Information recognition sensitivity InfoRecSensitivity (InfoRec) corresponding to information recognition degree InfoRec (however, InfoRec∈ {infoRecHigh, InfoRecLow}), task load TaskLoad (Inter) corresponding to task interruption degree Inter (where Inter∈ {InterHigh, Interlow}) The information amount InfoLength (m) [bytes] of the notification information m and the presentation time T _display (m) of the notification information m are defined. Here, InfoRecSensitivity (InfoRec) is taken as the cognitive information processing speed per unit time, for example, the unit is [bytes / sec]. W (m, t) is a weight determined by the current time Time. For example, the higher the degree of urgency, the higher the value. For example, a decreasing function at time t is defined.

From the above, the utility value U (InfoRec) given by the information recognition level and the utility value U (Inter) given by the task interruption are calculated by the following equations. Here, T _fullrec (m) is the time required for all the m-th notification information to be recognized. If the presentation time T (m) is T _fullrec (m) or more, all the information is recognized. U (InfoRec) is equal to the amount of information. That is, U (InfoRec) indicates the amount of information that can be notified within the monitored time.

Therefore, Formula (2) and Formula (3) hold.
U (InfoRec, m) = InfoRecSensitivity (InfoRec) × T (m) (if T _fullrec (m)> T (m)) (2)
U (InfoRec, m) = InfoLength (m) (if T _fullrec (m) ≦ T (m)) (3)
However, T _fullrec = InfoLength (m) / InfoRecSensitivity (InfoRec (m))
U (Inter) is expressed by equation (4). TaskLoad (Inter) is the task load increase rate that increases due to notification, ProcessSpeed (MainTask) is the information processing speed of the user when performing the main task (MainTask), TI (Task Importance) is the importance of the main task (MainTask) And how much the information processing speed of the main task is slowed down by the information notification is multiplied by the importance weight of the task. That is, U (Inter) indicates the information processing amount of the main task within the monitored time.
U (Inter, m) = (ProcessSpeed (MainTask) / TaskLoad (Inter)) × TI (Maintask) (4)
The user notification utility deriving unit 114 calculates the expected utility value so that the sum of the information amount that can be notified within the monitored time and the information processing amount of the main task during that period is maximized.

Furthermore, when the information importance level (including the urgency level) W (m, Time) of the notification information m is defined, U (InfoRec, m) is expressed by the following equation (5) using W (m, Time). ) And equation (6).
U (InfoRec, m) = InfoRecSensitivity (InfoRec) × T (m) × W (m, Time)
(if T _fullrec (m)> T (m)) (5)
U (InfoRec, m) = InfoLength (m) × W (m, Time) (if T _fullrec (m) ≦ T (m)) (6)
On the other hand, not only the notification information m to be provided now (t = t), but also at least one of the past (t = t-1) notification information (m-1) and the future (t = m; m> 0) It is also possible to determine the EU value in consideration of notification information (m) that is expected to be notified. For example, the utility value having the current “information recognition” element and “task interruption” element and the “information recognition” element and “task interruption” element traced back for a predetermined time as constituent elements is calculated. In this case, utility values obtained by providing information at each time are derived.

  When EU values are determined in consideration of notification information (m) in the past (t = t-1) and notification information (m) expected to be notified in the future (t = m; m> 0) A conceptual diagram of this is shown in FIG.

Here, if the index of the notification information that can be provided within the future time window T _w (for example, 1 minute or 30 seconds) being considered can be expressed by m, m is 0 ≦ m ≦ M _max . Here, M _max is the maximum integer that satisfies the following expression (7).

Using the above M _max , EU can be obtained as shown in Equation (8). In equation (8), the first term on the right side is the utility value related to the previous provided information, the second term is the utility value related to the notification information that is currently being provided, and the final term is provided within _Tw time from the current time. Indicates the utility value based on M _max pieces of notification information. The EU is the sum of these utility values.

Here, since EU _m-1 is already provided information, P (InfoRec | E) is 1 and InfoRec is determined. Therefore, from the time indicated by the time index t = t-1 at the time of notification, Using the elapsed time T (m−1) to the time indicated by the current time index t, it is calculated by the above equation U (InfoRec, m−1).

On the other hand, since EU _m is an expected value, the observation element used in the Bayesian network calculates P (InfoRec | E) on the assumption that the current time state is continued. P (Inter | E) is also predicted and used using the provision time T (m) that is a parameter. However, the method of obtaining the expected utility value is not limited to the above, and other methods may be used.

  On the other hand, as described above, it is possible not only to calculate EU from U and probability P, but also to determine utility values by comparison with prior probabilities P (Inter) and P (InfoRec). For example, there is a method of calculating the utility value from U (InfoRec) and P (InfoRec | E) by comparing P (Inter) and P (Inter | E). This realizes a control not to notify when the estimated task interruption degree is higher than the prior probability.

Alternatively, it can be calculated from the ratio of P (Inter) and P (Inter | E).

  The user notification utility deriving unit 114 notifies the notification execution unit 126 of a maximum expected utility value (MEU) that is the maximum value of the expected utility value derived by the above calculation formula. The maximum expected utility value is calculated by the following equation (11).

When considering the past and future EU, the maximum MEU within T _w time is determined by the Lagrange undetermined multiplier method, etc., using the presentation time T (m) of 0 ≦ m ≦ M _max as a parameter. m) can be determined.

In addition, the expected utility value at each time related to information notification may be derived from the utility function having the “information recognition” element and the “task interruption degree” element estimated from the Bayesian network as variables. When calculated, the user notification utility deriving unit 114 transmits the maximum value MEU and the index of the information expression method that provides the maximum value to the notification execution unit 126. An example of a table of information expression index information transmitted by the user notification utility deriving unit 114 is shown in FIG.

  The information expression index table describes the media type, position / size, text display method, update speed, and the like for each information expression index.

The notification execution unit 126 compares the received maximum notification utility value with the threshold MEU _th to determine whether or not to transmit. In the case of transmission, the notification information and the information expression index information are transmitted to the notification information display control unit 134.

The threshold value MEU _th can be determined statically in advance or based on the past history. For example, from the combination of {information representation, task state, user operation state} information when providing information in the past and feedback state information {display speed adjustment operation, display size adjustment operation, task speed} information {information recognition degree, task P (InfoRec | InfoRep, Task, Motion, SpeedAdapt, SizeAdapt, TaskSpeed) and P (Inter | InfoRep, Task, Motion, SpeedAdapt, SizeAdapt, TaskSpeed) can be obtained. Then, the EU is calculated from them, and the EU when information was provided in the past is obtained and accumulated. In this way, the optimum MEU _th can be obtained.

  Next, an example of specific calculation is shown.

  In the calculation, a conditional probability table (CPT) indicating the structure of the Bayesian network and the relationship between each element is required. Therefore, an example of calculation when they are given as shown in FIG. However, in order to show an example of the calculation result, FIG. 9 deletes some links in the structure of FIG. Also, the discrete variables of each element are limited to simplify the calculation example.

  Probability P (Iner | Rep, Task) of “task suspension level = high” before information notification for information representation when user state and task state are given refers to conditional probability table for task suspension element Can be obtained. For example, when “information expression method = small window” and “task state = email”, the probability of “task interruption degree” P (InerHigh | RepSWindow, TaskEmail) = 0.4, P (InterLow | RepSWindow, TaskEmail) ) = 0.6.

  Similarly, by referring to a conditional probability table (not shown) for information recognition elements, for example, the probability of “information recognition degree” P (InfoRecHigh | RepSWindow, MotionWalking, TaskEmail) = 0.2, P (InfoRecLow | RepSWindow, MotionWalking, TaskEmail) = 0.8.

Here, when defined as U (InfoRecHigh) = 0.8, U (InfoRecLow) = 0.2 and U (InterHigh) = − 0.8, U (InterLow) = − 0.4, it can be calculated as follows.
EU = (0.2 × 0.8 + 0.8 × 0.2) + {0.4 × (-0.8) + 0.6 × (-0.4)} =-0.24
Based on the above calculation, MEU can be calculated by calculating EU for all information representation methods.

  Furthermore, as described above, it is also possible to make a determination by comparison with the prior probability P (Inter). For example, P (Inter) can be calculated as follows.

= P (InterHigh | TaskNothing, RepNothing) P (TaskNothing) P (RepNothing)
+ P (InterHigh | TaskNothing, RepSWindow) P (TaskNothing) P (RepSWindow)
+ P (InterHigh | TaskNothing, RepLWindow) P (TaskNothing) P (RepLWindow)
+ P (InterHigh | TaskMail, RepNothing) P (TaskMail) P (RepNothing)
+ P (InterHigh | TaskMail, RepSWindow) P (TaskMail) P (RepSWindow)
+ P (InterHigh | TaskMail, RepLWindow) P (TaskMail) P (RepLWindow)
= 0.1 × 0.7 × 0.1 + 0.2 × 0.7 × 0.5 + 0.3 × 0.7 × 0.4
+ 0.2 × 0.3 × 0.1 + 0.4 × 0.3 × 0.5 + 0.9 × 0.3 × 0.4
= 0.334
From the above, since P (InterHigh | RepSWindow, TaskEmail)> P (InterHigh), EU = 0 can be calculated.

Next, a method for determining the threshold MEU _th based on the posterior probability estimation will be described.

  For example, when information is notified to a user in the state of “information expression = small window” and “task state = e-mail”, the speed of the primary task (e-mail here) is reduced as a user reaction (ie, “ Task speed = slow ”), but there is no adjustment to the button for changing the size (that is,“ size adjustment = none ”) as a threshold condition. Therefore, the posterior probability P (InterHigh | TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) of “task interruption level = high” is obtained as follows from the past example when the above combination occurs.

First, the joint probability of these elements can be expressed by the following equation (14).
P (InterHigh, TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing)
= P (InterHigh | TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) × P (TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) (14)
On the other hand, similarly, it can be expressed as the following equation (15).
P (InterHigh, TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing)
= P (TaskSpeedSlow | InterHigh) × P (SizeAdaptNothing | InterHigh) × P (InterHigh | TaskMail, RepSWindow) × P (TaskMail) × P (RepSWindow) (15)
Therefore, the probability to be obtained can be derived from the Bayes' theorem as shown in the following equation (16).
P (InterHigh | TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing)
= P (TaskSpeedSlow | InterHigh) × P (SizeAdaptNothing | InterHigh) × P (InterHigh | TaskMail, RepSWindow) × P (TaskMail) × P (RepSWindow) / P (TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) (16)
Here, it can be calculated as follows.
P (TaskSpeedSlow | InterHigh) × P (SizeAdaptNothing | InterHigh) × P (InterHigh | TaskMail, RepSWindow) × P (TaskMail) × P (RepSWindow)
= 0.8 × 0.2 × 0.4 × 0.5 × 0.3 = 0.0096
On the other hand, P (TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) can be calculated as follows.
P (TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing)

= P (TaskSpeedSlow | InterHigh) × P (SizeAdaptNothing | InterHigh) × P (InterHigh | TaskMail, RepSWindow) × P (TaskMail) × P (RepSWindow) + P (TaskSpeedSlow | InterLow) × P (SizeAdaptNothing | InterLow) × P ( InterLow | TaskMail, RepSWindow) × P (TaskMail) × P (RepSWindow)
= 0.8 × 0.2 × 0.4 × 0.5 × 0.3 + 0.2 × 0.5 × 0.6 × 0.5 × 0.3
= 0.0186
From the above, the calculated value can be calculated as follows.
P (InterHigh | TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing)
= 0.0096 / (0.0096 + 0.009) = 0.516129
The a posteriori probability P (InterHigh | TaskMail, RepSWindow, TaskSpeedSlow, SizeAdaptNothing) can be calculated by the above calculation method, and the threshold value MEU _th is determined from the maximum value or the average value in these histories.

  As described above, the expected utility value EU can be calculated when a Bayesian network as shown in FIG. 9 is given.

  Next, the notification information display control unit 134 will be described. The notification information display control unit 134 actually presents information on the information display terminal using the notification information received from the notification execution unit 126, for example, the notification information content and the information expression index. The notification information display control unit 134 has a user interface capability capable of displaying notification information, and controls the display position / size, text presentation method, and information update frequency (speed). The interface allows the user to adjust the timing and expression method in information notification.

  Furthermore, the notification information display control unit 134 can accept an interaction from the user. For example, as shown in FIG. 10, a button that can change the display position / size, a display update frequency (speed), and the like. A button that can be changed, a button for stopping the display itself, and the like can be provided, and the user's preference can be met. It is also possible to access link information by clicking on the information itself.

  Next, the feedback information collection unit 132 will be described.

  The feedback information collection unit 132 acquires user interaction information for a button or the like of the notification information display control unit 134 and notifies the information notification state history storage unit 116 of the information as feedback information.

  Further, the feedback information includes a task execution speed in addition to the display speed adjustment operation and the display size adjustment operation, and the feedback information collection unit 132 measures the task execution speed. The feedback information collection unit 132 measures the speed of button operations until the start of various applications, the character input speed when editing emails, the browsing speed when browsing the web, etc. The change of the task speed on the information display terminal that the user has performed is acquired and fed back. By feeding back these pieces of information, the user state model update unit 115 can update the relationship between the task state and the information recognition factor element or the task interruption factor element. Further, the feedback information collection unit 132 measures the operation speed for other tasks on the information display terminal even when information is displayed or when information is not displayed, and notifies the information notification state history storage unit 116 of the operation speed. To do.

  It should be noted that the feedback information collection unit 132 collects feedback information via a button or the like in an interface corresponding to the notification information display control unit 134 to the last. On the other hand, applications that are displayed simultaneously with the user interface unit 130, for example, user interactions such as an interaction with a mail client, are collected from the sense unit 150. However, the feedback information collection unit 132 collects specific application interaction information designated by the feedback information collection unit 132 for model update.

  The information notification state history storage unit 116 stores the notified interaction information. These pieces of information are used for user state model update and notification control for improving model accuracy.

  The information notification state history storage unit 116 stores and manages a state history (corresponding to a snapshot) of elements constituting a model when information notification is performed. The information notification state history storage unit 116, when information notification is performed or when a specific combination of states is observed even when information notification is not performed, the variable values of all the elements constituting the model and All combinations with feedback information may be stored and managed as a table.

  Further, the feedback information collection unit 132 notifies the information notification state history storage unit 116 of the notification information that has been notified one or more times ago and the information recognizing information when the notification information is notified as feedback information. You may make it do. By doing in this way, since the user state model update part 115 can update a state model based on the information regarding information cognition, it can notify information based on the current information cognition. For example, control can be performed so that the amount of information that can be notified to the user is increased.

Here, whether or not information notification has been performed, including whether or not information notification has been performed, whether or not the information notification has occurred, the display time in the notification information display control unit 134 when the information notification has occurred, and a diagram for them. 11 is input from the feedback information collection unit 132 to the information notification state history storage unit 116. Further, environmental user status information matching the display time information is also input from the environmental user status storage unit 146. The information notification state history storage unit 116 manages the input history information. However, since “information recognition” and “task interruption” cannot be directly observed, {information expression, task state, user operation state} information and feedback state information {display speed adjustment that are observed when information is provided "Information recognition degree" and "task interruption degree" are estimated from the operation, display size adjustment operation, task speed} information. For example, it can be calculated and estimated by the method described in the section of deriving the threshold value MEU _th . Here, a table example of history information is shown in FIG. In the history information, the operation state, task state, information expression, and feedback are described with respect to time.

  Further, all combinations of environments and user states may be stored in advance as a table, and the occurrence frequency may be counted. An example of a table indicating the occurrence frequency in this case is shown in FIG. In the occurrence frequency table, an operation state, a task state, an information expression index, and an occurrence frequency are described for the index.

  Next, model update processing will be described.

  Regardless of whether the information is notified or not, the model is updated in order to improve the accuracy of the model by using the obtained feedback information from the user. That is, the offline initial teacher data can be realized by learning the training data. However, since these are only reference data for training, the model generated therefrom must be called the initial model. Therefore, it is necessary to improve the accuracy of the model by using this information notification system. The user state model update unit 115 implements this process.

  The user state model update unit 115 obtains a random variable between elements based on the data stored in the information notification state history storage unit 116 and updates the model structure. For example, the user state model update unit 115 updates the model structure each time new information is added to the user notification history storage unit 116 or whenever a certain number of new information is stored. In realization, it is realized by performing the same processing as when a model is created from training data.

  Further, the user state model update unit 115 may dynamically change (learn) the entire structure of the relationship between the state elements based on the data stored in the information notification state history storage unit 116.

  As described above, the Bayesian network described above can be dynamically updated and the user state model can be updated based on information notification and explicit or implicit feedback information from the user. . For this reason, adaptation (personalization) to each user is attained.

  However, according to the determination of the notification execution unit 126, it is difficult to acquire history information about all possible combinations of elements constituting the user state model only by realizing adaptive information notification. Therefore, it is difficult to accurately derive the causal relationship between elements as described above and to obtain a more suitable model structure, and as a result, it is necessary to improve the representation accuracy of the model itself.

  Therefore, for the purpose of positively improving the accuracy of the expression ability of the user state model, information notification is actively performed when a combination other than a combination that has occurred in the past occurs. Specifically, the model accuracy improvement notification control unit 118 inquires of the information notification state history storage unit 116 that manages the past information notification history to detect a combination state with a low occurrence frequency. Then, by monitoring the environmental user status information input from the environmental user status generation unit 142, when the combination occurs, the information execution request is notified to the notification execution unit 126.

  That is, a combination of state elements observed in the past is managed, and control is performed so as to actively generate a combination event with a low frequency. By doing in this way, a more accurate inter-element relationship structure can be constructed.

  Hereinafter, this notification process is referred to as “state combination event type information notification process” in distinction from the above two processes, that is, the inquiry type process and the event type process.

  On the other hand, the model accuracy improvement notification control unit 118 periodically acquires combination index information with a low occurrence frequency from the information notification state history storage unit 116. Then, the collation process is performed with the information from the environmental user state generation unit 142, and if they match, an information notification request with the information expression index information attached is sent to the notification execution unit 126.

  The process in the model accuracy improvement notification control unit 118 will be described with reference to FIG.

  First, the model accuracy improvement notification control unit 118 acquires combination index information with a low occurrence frequency from the information notification state history storage unit 116 (step S1402).

  Next, the model accuracy improvement notification control unit 118 acquires environmental user state information from the environmental user state generation unit 142 (step S1404).

  Next, a collation process is performed between the acquired environmental user status information and combination index information with a low occurrence frequency (step S1406).

  When the environmental user state information does not match the combination index information with a low occurrence frequency (step S1406: YES), the model accuracy improvement notification control unit 118 determines whether or not there is an end command signal in the environmental user state information. (Step S1408) If there is an end command signal (Step S1408: YES), the process ends. On the other hand, when there is no end command signal (step S1408: NO), the process returns to step S1404.

  On the other hand, when the environmental user state information matches the combination index information with a low occurrence frequency (step S1406: NO), the model accuracy improvement notification control unit 118 issues an information notification request (step S1410). Next, the model accuracy improvement notification control unit 118 determines whether or not there is an end command signal in the environmental user state information (step S1412). If there is an end command signal (step S1412: YES), the process ends. On the other hand, when there is no end command signal (step S1412: NO), the process returns to step S1402.

  Next, state combination event type information notification processing will be described with reference to FIG.

  The sensor information receiving unit 144 periodically collects sensor information from the sensing unit 150 (step S1502) and inputs the sensor information to the environmental user state generating unit 142. The environment user state generation unit 142 creates environment user state information indicating the environment and the user state, and inputs the created environment user state information to the model accuracy improvement notification control unit 118 (step S1504).

  On the other hand, the service provider or the information receiving user himself / herself inputs the content of the information notified to the information receiving user to the notification information receiving unit 122 as needed (step S1506). The notification information receiving unit 122 inputs the content of the input information to the notification information storage unit 124.

  Here, the model accuracy improvement notification control unit 118 periodically acquires and sets combination index information with a low occurrence frequency from the information notification state history storage unit 116 (step S1508). Further, the model accuracy improvement notification control unit 118 collates the environmental user state information input from the environmental user state generation unit 142 with the combination index information with a low occurrence frequency, and determines whether or not they match.

  If the combination with low occurrence frequency matches the environmental user state information, the model accuracy improvement notification control unit 118 sends an information notification request with the information expression index information attached thereto to the notification execution unit 126 (step S1510).

  The notification execution unit 126 transmits the notification information content and the information expression index to the notification information display control unit 134 (step S1512), and the notification information display control unit 134 displays the information content based on the display method indicated by the information expression index. (Step S1514). After the information is displayed (step S1516), the feedback information collection unit 132 collects the interaction information fed back from the user who has received the information (step S1518) and inputs it as feedback information to the information notification state history storage unit 116 (step S1520). ). By doing in this way, the precision of the expression capability of a user state model can be improved positively.

  Next, the sense unit 150 and the environmental user state management unit 140 will be described.

  First, the sense unit 150 will be described.

  The sense unit 150 has all the functions for acquiring and outputting information indicating the state of the user and the environment, such as various sensors such as an acceleration sensor, a vibration sensor, a microphone, and a camera, and a terminal such as a computer and a mobile phone. Point to. Of course, the user interface unit 130 and the sense unit 150 may be arranged on the same information display terminal. Information acquired by the sensing unit 150 is notified to the sensor information receiving unit 144.

  On the other hand, the sensor information receiving unit 144 inputs the information received from the sensing unit 150 to the environmental user state generating unit 142.

  The environment user state generation unit 142 removes some variation and noise by performing preprocessing and normalization processing according to the type of information. Then, feature vectors are created from these pieces of information, and meaningful information is generated by performing identification processing such as clustering. These processes differ depending on the target data and the target to be recognized. For example, with respect to user operation states constituting the network, for example, information on wearable acceleration sensors worn by the user is divided into certain fixed time windows to extract features. By doing so, it is possible to create status information such as “stopped”, “walking”, “running”, etc. in time units.

  The environmental user status information is input to the environmental user status storage unit 146 and stored and managed together with the time information.

  According to the embodiment of the present invention, it is possible to realize the information notification that grasps the situation of the user living in the real world and can keep the information recognizability high while keeping the user's task discontinuity low. . Since the two elements of task interruption and information cognition are human internal states that cannot be directly observed, observable elements that affect both elements are introduced, and the relationship structure between these multiple elements By dynamically holding and managing the state, adaptive state management according to the situation becomes possible, and appropriate information notification can be made based on the state.

  In addition, the notification method can be controlled while dynamically estimating the “information recognition state” and the “task interruption degree” by notification as human internal state elements. Also, the state of the relationship between these elements can be learned and updated dynamically while repeating the notification.

  The notification user state management device, notification control device, information notification device, information notification system, and information notification method according to the present invention can be applied to a system that performs information notification.

It is a block diagram which shows the information notification system concerning the Example of this invention. It is a sequence diagram which shows an inquiry type | mold information notification process. It is a sequence diagram which shows an event type information notification process. It is explanatory drawing which shows an example of the Bayesian network for information notification. It is explanatory drawing which shows an example of the information expression method. It is explanatory drawing which shows an example of the decision making network containing the determination of a notification utility value. It is explanatory drawing which shows an example of the decision making network containing the determination of a notification utility value. It is explanatory drawing which shows an information expression index table. It is explanatory drawing which shows an example of a Bayesian network and a conditional probability table. It is a schematic diagram which shows the example of a button interface for display adjustment. It is explanatory drawing which shows a feedback information index. It is explanatory drawing which shows a log | history information table. It is explanatory drawing which shows the occurrence frequency table. It is a flowchart which shows the process of the notification control part for model precision improvement. It is a sequence diagram which shows a state combination event type information notification process.

Explanation of symbols

100 Information notification device

Claims (17)

User internal state estimating means for estimating the internal state of the user based on the observable state element;
User notification utility estimation means for estimating utility to a user who receives information when notifying information based on the internal state of the user;
A user state management device for notification, comprising: In the user state management device for notification according to claim 1:
The user internal state estimation means is interrupted by an information recognition degree indicating a degree of information perception and recognition by the user as an element indicating an internal state of the user and an information notification of work performed by the user. A notification user state management device characterized by estimating at least one of task interruption levels indicating degrees. In the user state management device for notification according to claim 2:
The user internal state estimation unit uses a Bayesian network to estimate a probability of at least one of the information recognition degree and the task interruption degree. In the user state management device for notification according to any one of claims 1 to 3:
The notification user state management device characterized in that the user notification utility estimation means calculates an expected utility value indicating a utility expected when information is notified based on an internal state of the user. In the user state management device for notification according to claim 4:
The user notification utility estimation means calculates an expected utility value from at least one of an information amount that can be notified within a predetermined time and an information processing amount of a main task within the predetermined time. User state management device. In the user state management device for notification according to any one of claims 1 to 5:
Information storage means for storing at least one of the observable state element and feedback information from the user;
User state model updating means for updating the user internal state based on the accumulated information;
A user state management device for notification, comprising: In the user state management device for notification according to claim 6:
An observable state element determination unit that determines whether or not an observable state element that is equal to or lower than a predetermined occurrence frequency in the stored information matches a newly acquired observable state element;
A user state management device for notification, comprising: Notification execution means for determining whether to notify information based on the utility to the user who receives the information when notifying the information, and notifying the information based on the determination;
A notification control apparatus comprising: In the notification control device according to claim 8:
The notification execution means determines whether to notify information based on a comparison result between an expected utility value indicating expected utility when notifying information and a preset value, Notification control device. An information notification device comprising a notification user state management device and a notification control device:
The notification user state management device includes:
User internal state estimating means for estimating the internal state of the user based on the observable state element;
User notification utility estimation means for estimating utility to a user who receives information when notifying information based on the internal state of the user;
The notification control device comprises:
Notification execution means for determining whether to notify information based on the utility and notifying the information based on the determination;
An information notification apparatus comprising: An information notification system comprising a notification user state management device and a notification control device:
The notification user state management device includes:
User internal state estimating means for estimating the internal state of the user based on the observable state element;
User notification utility estimation means for estimating utility to a user who receives information when notifying information based on the internal state of the user;
The notification control device comprises:
Notification execution means for determining whether to notify information based on the utility and notifying the information based on the determination;
An information notification system comprising: Estimating the internal state of the user based on the observable state element;
Estimating the utility to the user who receives the information when the information is notified based on the internal state of the user;
Determining whether to notify information based on said utility;
Notifying information based on said determination;
An information notification method characterized by comprising: The information notification method according to claim 12:
The step of estimating the internal state of the user includes, as an element indicating the internal state of the user, information recognition degree indicating a degree of information perception and recognition by the user and information notification of work performed by the user. An information notification method comprising a step of estimating at least one of task interruption levels indicating a degree of interruption. The information notification method according to claim 12 or 13:
The step of estimating the utility to the user includes the step of calculating an expected utility value indicating the expected utility when the information is notified based on the internal state of the user. The information notification method according to claim 14:
The step of calculating the expected utility value includes a step of calculating an expected utility value from at least one of an information amount that can be notified within a predetermined time and an information processing amount of the main task within the predetermined time. A characteristic information notification method. The information notification method according to claim 14 or 15, wherein:
The determining step includes a step of determining whether or not the expected utility value satisfies a preset value. The information notification method according to claim 12:
Accumulating said observable state element;
Determining whether or not the observable state element that is equal to or lower than a predetermined occurrence frequency in the accumulated information matches the newly acquired observable state element;
An information notification method characterized by comprising: