-Providing proactive service requires careful treatment because it has many risks that may force the user to be insufferable with the inadequate services. The context-based interaction pattern is an important factor to personalize the services because it is the practical foundation to adapt to the user. It can be extracted based on the user's activity relative to time, space, and object. In this paper, we describe the methodology to extract user interaction pattern with an example, smart home.
Index Terms—Personalization, context-aware systems, proactive service, smart spaces, ubiquitous computing
Tviding the proactive services to the user. The system may
HE purpose of building context-aware system is for
prorecognize the situation of the user with some assumptions, and
then it provides the related service to user by itself. However,
providing proactive service needs to be careful because it has
many risks that may force the user to be insufferable with the
inadequate services [
In this reason, the proactive services may need two elements
to judge the accurate service; the user context and the user’s
preference. The user context has been issued by many
researchers of context-awareness [
In this paper, we propose the methodology for modeling context-based interaction pattern from the system’s experience with the user. In Section 2, we describe the entities of context-based personalization; the user context and system experience. The user context model is explained more detail in Section 3. In Section 4, we describe the user interaction model to extract the context-based user interaction pattern. Context Cube as a visualization tool is proposed in Section 5. Finally, the conclusion and future works are discussed in Section 6.
II. ENTITIES OF CONTEXT-BASED PERSONALIZATION
The user context is the user-centered situational information.
It is the basic information to trigger the related services
according to the current user’s situation. The user context is
generally categorized by who the user is, where the user is,
what the user is doing, when the user does it, and on which the
user focuses. They are the elements to represent the current
situation of the user in the interaction space. When the user
context indicates a certain situation, the related service may be
invoked. Therefore, it is the trigger to call the related services
[
The system experience is information that the system can gather from the interaction with the user (e.g., the repeated actions in a certain situation, or the interrupt against the proactive service that the system provides). It means that the system experience is not the directly defined options by the user, but the information which is gathered seamlessly through the interaction with the user. It may be reused as context to decide the service method. It is independent on the pre-designed scenarios, and it can be adapted to the user according to the activity patterns of the user dynamically.
We distinguish system context (lower-level) from user
context (upper-level) as the whole device states available for a
context-aware system [
Space class means the physical domain in the environment (e.g., home). It may have some districts (e.g., living room, kitchen, and bedroom). Each space consists of zones which are the square area in plane coordination. Zone has four properties defining the bounds of each corner (e.g., lower left, lower right, upper left, and upper right). Using the space ontology, the system can map the semantic domain from the physical space. Each object in this space also may have the semantic domain property. The system identifies the objects by their own UUID (Universal Unique Identifier). In the user context, the objects may be abstracted and classified according to their own properties and services. Therefore, defining the object ontology requires the common vocabulary which depends on the domain because each object may be manufactured by various vendors and their properties may be heterogeneous. When a new object involves in the environment, the user context should make an instance according to the common vocabulary from the system context. Table II shows the example of light object model based on Service Type SwitchPower Dimming
Variables PowerTarget PowerStatus LoadLevelTarget LoadLevelStatus OnEffectLevel OnEffect StepDelta RampRate IsRamping RampPaused RampTime
Type Boolean Boolean ui1 ui1 ui1 String ui1 ui1 Boolean Boolean ui4
Allowed Value true, false true, false 0-100 0-100 0-100 OnEffectLevel, LastSetting, Default 1-100 0-100 true, false true, false 0-4294967295
This upnp-enabled dimmable light has two available services; switch
power and dimming. It is enabled to control these services over the network.
UPnP device specification [
Time is the basis to model the user interaction pattern as it
increases infinitely, but the conceptual time loops recursively
(e.g., daytime, night, day, day of the week, week, month,
season, and year). The user’s repetitive actions may have the
relations with the conceptual time. The time ontology defines this
conceptual time, and when the user activity occurred, the user
context may be represented with it [
The activity is the model of characterizing the status of the user. It is used as a trigger to invoke the related services. That is, a basis to judge when and how the system reacts. The possible activities in an environment are dependent on the domain. Therefore, the activity ontology defines the whole possible activities. Each activity may be inferred from the system context by its own rules.
The user interaction model is a pattern of occurring the user’s activities. When the user is on a specific activity, the chained actions may be occurred according to the relativities for the task. It is possible modeling a pattern from the repetitive and chained operations by the user and it would be the useful context which can be reused at the next similar situations. To provide the proactive service which satisfies the user, it is important that the system uses the user’s personalized interaction model as a context. There are several elements for modeling user interaction pattern as following:
This is composed by the repeated activities which are dependent on conceptual time (e.g., waking up, eating, going to church, sleeping). If an activity is occurred recurrently at a certain conceptual time, the activity has a relation with time.
The whole activities are occurred in a space. However, the activities may have dependencies for the specific spaces (e.g., taking a shower, sleeping, cooking) or not (e.g., listening the music, talking). If an activity is occurred recursively at a certain space, the activity has a relation with space.
Although the whole activities are not dependent on the specific objects, many activities are dependent on the objects (e.g., sleeping, shaving, listening to music). If an activity is occurred recursively with a specific object, the activity has a relation with the object.
The core dependencies of activity are independent on each other. Therefore, the activity may depend on the whole dependencies or not. Fig. 3 shows that an activity has the relations among time, space, and object dependency. There are the additional dependencies for user activity.
The schedule is a reserved activity. If the user has a schedule in a little time, his current activities may relate to the schedule (e.g., morning call, meeting appointment).
If the proactive service is not triggered, the user may command the task manually. Then, it should be reflected in the next similar situations. And also when the user interrupts and rollback the provided service, the system should be reflected.
We think visualizing interaction patterns is useful because it
makes easier for grasping and analyzing by end-user. Context
Cube is our approach to the visualization tool for modeling the
personalized interaction patterns in terms of domain activity,
space, object, and time properties. Bauer et al. denoted their
sensor device as ‘ContextCube’, to serve as an information
source [
It may show just consecutive activity logs at early stage. However, as the specific activity is occurred repeatedly, it grows the possibility that can be occurred at next similar situation. Context Cube has five properties defining the possible elements depends on the domain as following:
Domain activity property defines the whole possible user activities in the interaction space. As we mentioned in section three, the activity is defined with its own rules. When the environment situation satisfies the rules, the activity is occurred. Each activity has the weight variable and it grows when the same activity that has the similar relations with space, object, or time properties.
Space property defines the boundary and districts of the physical domain. A space may contain objects or users; therefore it also groups the objects surrounding the user.
Object property contains the current object resources available in the interaction space. The whole objects have the location attribute which is related with space property.
Time property consists of the interpretation factors classified by the level of detail. The conceptual abstraction of time can be the period of user activity (e.g., Monday, July, and evening).
Personalized interaction pattern is a set of the activities. As the specific activity is occurred repeatedly, its weight value will be greater by counting.
There are several benefits when we model context-based interaction pattern using Context Cube. First, we can model multiple personalized interaction patterns for each user. As the each end-user may set their own interaction patterns, the user would fine it very useful, and as it is stated earlier, some of those patterns would be found automatically and it would be very user-friendly.
Second, the 3D visualized model of Context Cube is easy for understanding and customizing the services by end-user. Because each interaction patterns are related with something they are used to (i.e., time, space, and object) the user would find it easy to understand new patterns or setting new patterns for the future use.
Third, this generalized model can be applied to the various domains. The three factors of the Context Cube are universal so it should be able to apply to the other areas of the fields. Although it is not proof of the rules of natural world, we believe this generalized context-based personalization model assists to develop the personalized and proactive services. In future work, we will apply the probabilistic model to extracting the interaction pattern.
We thank to Hyun-Jhin Lee. The working of Context Cube is inspired from her research.