A Task-oriented Approach to Support Spontaneous Interactions among Users in. Urban Computing Environments. Angel Jimenez-Molina, Byung-Seok Kang, ...
A Task-oriented Approach to Support Spontaneous Interactions among Users in Urban Computing Environments
Angel Jimenez-Molina, Byung-Seok Kang, Jun-Sung Kim, In-Young Ko
Department of Computer Science Korea Advanced Institute of Science and Technology Daejeon, Republic of Korea {anjimenez, byungseok, junkim, iko}@kaist.ac.kr
to support social groups made from spontaneous interactions of a huge number of diverse users. Ubicomp and UrbComp share a common set of basic requirements: context awareness, user centricity, composability of services and dynamicity of the environment and of users' goals [2]. However, the challenge of UrbComp imposes a newer, distinguishable requirement: spontaneity. It is about supporting spontaneous interactions among users by coordinating available services during runtime, without having a previous defmition of applications into templates or other predefined descriptions. The related work about supporting interactions of users with available services in UrbComp environments has tended to focus on recognizing activities based on the identification of meanings attached to a place. However, these studies do not realize the new requirement of spontaneity. A way to realize spontaneity is the approach of task-oriented computing [3]. This approach is about representing social groups' goals in unit-tasks (basic configuration of abstract services), which may then be composed together to defme a larger task in an emergent way. In order to select appropriate unit-tasks for spontaneous social groups, a kind of place-awareness is required. This concept is concerned with reflecting placeness in the selection of unit-tasks. The concept of placeness integrates the aspects that create or inhibit opportunities for spontaneous social interactions in urban environments. Those aspects consist of spatial and temporal characteristics of urban environments, and personal and social features of people's behavior in these spaces [4]. Additionally, the task-oriented computing approach bridges the gap between unit-tasks and services. This paper extends the mediation process between unit tasks and services described in the framework for service provision for Ubiquitous and UrbComp environments [2], [5], and [6]. This framework consists of three composition layers - the task layer, the service composition-pattern (SCP) layer, and the service layer. The task layer is concerned with selecting appropriate unit-tasks based on placeness information. Moreover, it is concerned with composing those unit-tasks in a larger task that fully realizes the goal of a social group. The SCP layer provides a set of reusable composition-patterns that supports unit-tasks.
Abstract-Urban Computing is an extension of Ubiquitous Computing. predecessor,
It
shares
like
common
context
requirements
awareness,
user
with
its
centricity,
composabiJity and dynamicity. However, a challenging issue of urban computing is to provide available services to diverse, spontaneous social groups. A major requirement for this challenge is spontaneity, which is about supporting social groups by coordinating available services during runtime, without having a previous definition of applications into templates or other predefined descriptions. The approach of task oriented computing realizes these requirements. It is about representing users' goals in tasks. A task is composed of unit-tasks (basic configurations of abstract services), which are selected based on 'placeness'. Placeness consists of the aspects that
characterize
urban
environments
and
social
groups:
spatial, temporal, social and personal aspects. The focus of this paper is twofold: first, to provide a semantically-based unit task description model; second, to propose a unit-task selection algorithm based on placeness on top of the model.
The
appropriateness of the task selection algorithm is illustrated by a demostration scenario in our campus.
Keyword-Urban Computing; Ubiquitous Computing; Task oriented Computing; Spontaneous Users Interaction.
I.
INTRODUCTION
Urban Computing (UrbComp) is an extension of ubiquitous computing (Ubicomp). Similar to Ubicomp, the main goal of UrbComp is to enable users to access services available in an environment anytime, anywhere [1]. However, the difference is that the physical scale of an UrbComp environment is much larger than a Ubicomp space. Moreover, an UrbComp environment is denser in terms of its participating users, and more dynamic in terms of their behavior. The co-presence of users in an UrbComp environment provides opportunities for serendipitous interactions among people. Thus, by engaging in spontaneous social activities in a public place users conform to particular social groups. Those social groups are multiple and diverse in terms of the type of users and their service requirements. Supporting them with appropriate services available in an UrbComp environment is a challenging issue. Thus, a distinguishable goal of UrbComp is to select, compose and then deliver available services and information
978-1-4244-5328-3/10/$26.00 ©201O IEEE
176
the day of the week and the age range. However, fIltering by the joint pair (urban space, phase of the day) in the first term, and (day of the week, age range) in the second term, it is possible to recognize activities with up to 80% accuracy [8]. We name these properties as primary variables. Other properties like the range of age, gender, and preferences of the user are less effective for recognizing unit-tasks. However, they are effective for personalizing the selection of unit-tasks. The factors that govern a spontaneous interaction of users in an UrbComp environment have been extensively studied in recent years [9]. Many of those studies have focused on determining what factors make it more likely for users to adhere in physical or virtual social encounters in an UrbComp environment. Of particular interest are some empirical studies carried out either by analyzing users' feedback data or logs about co-presence history of users, among other techniques used to identify spontaneous social group creation in an UrbComp environment. It has been determined that social ties among users in an UrbComp environment are in a great majority of cases preferably governed by three types of social contexts: familiarity, similarity and favorability. Familiarity denotes the strength of a social tie among two users. Two users may be familiars if they have previous social ties and know each other very well. In contrast, two users may be perfect strangers, or people who have never had any contact with each other in the past. Moreover, a certain degree of knowledge among users in a specific UrbComp environment, like a bus stop or the corridors of a school, converts them into familiar strangers [10]. However, familiar strangers only interact in their common environment by visual contact, maintaining a barrier to personal relationships. As for similarity, this trait concerns the degree of similitude between every property of users' profIles preferences, age range, belonging to a place, gender, and occupation, among other properties. Favorability between two users may be obtained from users' feedback information about the performed activities. Based on the primary variables extracted from the Time-use Studies, and the social context for social encounters, we have designed a semantic representation model that arranges the generic properties of unit-tasks.
Additionally, an SCP consists of a coordination of abstract services defmed in advance by application and service developers. In the service layer, a unit-task is realized by mapping the abstract services of its SCP to service instances available in the UrbComp environment. The focus of this paper is on the provision of an adequate mechanism to select appropriate unit-tasks for spontaneous social groups based on placeness information. In order to meet this aim, we have developed a semantically-based unit-task description model. The essential semantic elements that compose this generic model are taken from two sources: first, from a real dataset of users' activities in urban environments known as Time-use Studies [7]; and second, from the essential variables that affect a spontaneous social interaction among users. Those variables have been taken from existing empirical studies about users' behavior in urban environments. We have also developed a unit-task selection algorithm based on this model. It systematically arranges the semantic properties of the model to select unit-tasks. The rationale of the arrangement is based on the results obtained by Partridge et al. [8], who analyzed variables that are the best in recognizing activities. Moreover, the algorithm also considers factors that lead to social encounters among users in an urban environment. The appropriateness of the unit task selection algorithm is illustrated by a demo scenario on our campus. The roadmap of this paper is as follows. Section II describes the semantically-based unit-task representation model. Section III proposes the unit-task selection algorithm. Section IV describes the implementation of the algorithm in the framework for service provision for Ubiquitous and UrbComp environments. Additionally, this section describes the demo scenario. Section V concludes the paper. II. A.
-
UNIT-TASK SEMANTIC REPRESENTATION MODEL
Semantic Elements Rationale
There exist many publically available datasets with records of humans' daily activities in urban and private environments; such datasets are called Time-use Studies [7], and have been populated with daily routines reported by hundreds of thousands of participants from different countries over decades. Additionally, those daily routine records have been enriched with information about user demographic data, day of the week, starting time, ending time, and space, among other contextual variables. Based on two years' of fmdings by Partridge et al. [8], derived from a statistical analysis of Time-use studies of different countries, it can be asserted that the Ubicomp community has realized the benefits of utilizing the results of variables that are best at predicting human activities to improve the accuracy of task recognition [2]. According to [8], the properties that are the best in predicting user's activities are location interpreted in our ontology as an urban environment , the phase of the day,
B.
Unit-Task Properties
A unit-task is described by spatial and temporal properties (see Figure 1). This aspect consists of urban environment and time (day of the week and phase of the day). An urban environment is represented by a set of properties: a place ID that identifies a specific space; a set of environmental constraints attached to the space level of noise, or degree of brightness and temperature allowed; the level of public-ness, from absolutely public, or quasi-public to private places; the potential of a place, which deals with the most likely type of activity or behavior occurring according to the environment's settings as manipulated by urban designers [4]. A unit-task is also described by the effects that its execution may produce. On the one hand, a unit-task produces environmental effects that may affect the -
-
-
177
context of the urban space, like the level of noise, or degree of brightness and temperature allowed. On the other hand, execution of a unit-task may expose the information delivered to users to a certain level of public ness. Additionally, a unit-task is described by personal and social properties. The social encounter property describes the type of spontaneous interaction that users may perform according to their personal and social characteristics. This interaction conforms to a social group. It may lead to a physical or a virtual encounter. The former, in a great majority of conditions, happens for familiar and familiar stranger users, while the latter is more likely to happen for strangers [10]. As for the social group type property, it is described by information about the users participating in a group. A social group is comprised of users with similar profiles and high favorability. Users may be insiders or outsiders to an urban environment, since in practice not all public spaces are open to everyone. This property is named belonging to the place. A social group is also described by the average and homogeneity of the age range. Finally, a social group is comprised of users that share common interests and preferences. It is represented by the property named has preferences.
specialized level - in accordance with the granularity of the unit-tasks. The generalized level is composed of coarse grained unit-tasks extracted from the ATUS, while the next level consists of a specialization of those unit-tasks. The unit-tasks allocated in the specialized level are fmer-grained. Nevertheless, those specialized unit-tasks are created from a user-centric perspective by application and service developers. Property values are defmed based on the unit tasks execution history. III.
The task selection algorithm is divided into two phases the generalized unit-tasks selection and the specialized unit tasks selection. The former makes use of the primary variables to select initial subsets of generalized unit-tasks from the unit-task ontology [2]. The latter is concerned with specializing the unit-task selection from those subsets, based on the requirements and features of the current potential social groups informed by a place-aware context manager. Since the properties involved in the first selection phase urban environment, phase of the day, day of the week and age range - change with low frequency over time, it is possible to run the generalized unit-task selection periodically. By doing so, feasible subsets of general unit tasks are extracted in advance. In contrast, the specialized unit-task selection phase needs to be executed during runtime based on the social groups' information (see Figure 2.)
C. Unit-Tasks Hierarchy
We have used the activities recorded in the American Time-Use Study (ATUS) [7] as the unit-tasks that populate our ontology. These unit-tasks have been arranged in a hierarchy of subsumption relationships. The hierarchy is divided into two levels - the generalized level and the {public,
A.
A..
private} placePublic-nessLevel
placePotential
Let Q
Literal
=
(qv q2)
be
the
�
I
primary
variables,
Spatial and Temporal Aspect {morning, afternoon, evening, night, etc.}
.. i4---placel
)---time--�
.
1 .n
Time
phaseOfDay----.!
{public, quasi-public, private}
I
Literal
1.-..0..1
timeSpaceContext
Public-nessEffect
and temperature degree. etc.
� � p Od r
with
dayofWeek
O n
r-;,.-,;-:--:-::-;---,1 1
Literal
Generalized Unit-Task Candidates
{leisure, entertainment, business, etc.}
L-'--'-'---"'-----7-'--'-'-"--'-'-'--'-"-'.J quasi-public,
I
UNIT -TASK SELECTION ALGORITHM
1 .. 1
executionType
I
Unit-Task
,--;,...,.-::-:-::-;---,
{independent, conditional, ground, interdependent}
uceEffects
Literal
Ontology
-----G
environmentalEffects
supportsUnitTask subClassOf supportsSocialEncounter
i-L�;�-�d:-----------------: Ontology
ii
I
familiarityType
r-;2::::,;---, {familiars, familiar strangers,
: Schema
l�____����_�_�:�:____
Literal
{physical encounter, virtual encounter}
------. : Property
::' :,0 ,
encounterTyp�
socialEncounter
0
friends, strangers}
L-_____________________________________�
Figure l.
Representation of Unit-tasks Semantics
178
q]=(urban space, phase of the day) and q]=(day of the week, age range). Let Cu be a set of constraints for the urban space u. The nodes in the generalized level of the unit-task ontology are denoted as the set N. The primary variables q] and q] are matched against N, which is denoted by the expression M atch( Q, N). We describe this mechanism in detail in [2]. This property matchmaking process is periodically applied in an urban environment, according to the current day of the week and phase of the day. For each age range aj a subset of totally matched unit-tasks is selected. Those subsets are then constrained by Cu, which produces a family T = {tal' ta2, ... , tan} of new subsets. The set T defmes the generalized unit-task candidates that are available to be specialized according to the dynamic social groups' information produced by the place-aware context manager. B.
subgroup is computed. If
I
Social Groups Input -
-
that case, it is not possible to determine an age range. Thus, the whole set T of generalized unit-tasks to perform the specialization in an analogous way is selected. 2) Support to Social Groups of Familiar Strangers: The specialized unit-task selection for the social subgroup Gfsj is performed almost in the same way. The difference is in the social encounter type applied to the specialization of the social groups composed of familiar strangers, which depends of other social behaviors. Of particular interest are the studies about how familiar strangers behave in urban environments. This issue has been extensively studied during the last four decades. According to a recent empirical study reported in [10], if a spontaneous social encounter of two familiar strangers occurs in an urban environment that is different from the space where they would treat each other as familiar strangers, most likely those users will agree to carry out a physical encounter. For instance, two students that usually have co-presence in the same corridor of the campus, but that have never interacted, most likely will treat each other as friends if they suddenly meet in a park or plaza far from the campus. Therefore, in such a case, the algorithm takes into account a specialization based on a physical encounter as in a social group of familiar users (see Figure 2.) On the contrary, if the co-presence is verifIed in the same environment where users keep their social distance as familiar strangers, the algorithm specializes based on a
C. Specialized Unit-Task Candidates
1) Support to Social Groups of Familiar Users: For each Gh the variance VGt/a) of the age of the users in that Input: Subgroups GIj; a set T
=
{tal' taz' ... ,tan} of generalized (Pl'PZ' ""Pf).
unit-task candidates; user preferencesP •
=
Output: A set S of specialized unit-task candidates. 0: for each subgroup GIj do I: VGf,(a); /I a: age 2: 3: 4: 5:
6: 7: 8: 9: 10:
II: 12: 13: 14:
ifVGf,(a)