the tool used by these individuals and the tool used
in the earlier knowledge worker study was more support for tracking detailed parts of information artifacts, such as parts of web pages. We analyzed data
collected as these individuals worked over periods
ranging from just over 300 to almost 600 days. We
found that the maximum number of task activations
per day ranged between 16 and 24. The average number of task reactivations per day was between 3 and 4
and the maximum number of task reactivations
across the period was between 85 and 148.
From an analysis of the task contexts associated
with these individuals’ work, we found that the average number of files per context exceeded the average
interesting files per context, suggesting decay was
effective in trimming the size of an individual’s task
context. We saw a similar trend for the web pages referenced as part of a task context.
From this study, we can see that tasks are revisited
substantially over time and that the contexts formed
for a task are used by knowledge workers. We also
learned there is a need for models of task context to
automatically trim the contents of the task context
and reflect a knowledge worker’s current needs.
Beyond Task Contexts
Task contexts provide a full history of the interaction
with a task. When considered alongside the rich history of documents that are often maintained through
version repositories, shared file systems, and the like,
there arises the possibility of rewinding both task
context and artifact state concurrently, enabling a
knowledge worker to easily return to a previous state
in their work history. In essence, such a facility could
be used to rewind to a previous activity state of a
To date, we have provided and experimented with
situations in which one task can have multiple task
contexts at different points in time contributed by
different individuals, but tasks do not share task contexts. At times, users of our tools have requested the
ability to allow task contexts to refer to other tasks
and their associated contexts. For example, if task A
has subtasks B and C, the context of task A could be
the composite of task B and C. Support for cascading
task contexts along such relationships would substantially change our treatment in the tools of task
contexts. Instead of a task context being a snippet of
one user’s interaction history, a task context could be
identified by a set of spans of an interaction history
and relations to other such spans. Activating a task
could then involve swapping in the corresponding
segments of interaction. Such a facility would further
support rewinding interaction (and hence activity)
across one or more tasks.
More recently, in a commercial context, we have
been gaining experience with the representation of
tasks, communication activity, and artifacts across
many different repositories serving many different
kinds of knowledge workers. This exposure has
helped us refine our model for representing task and
task context information. We see artifacts as playing
more than one role in any larger schema describing
tasks and their context. In particular, artifacts and
their versions need to be more strongly modeled as
parts of tasks at different points in time with task
contexts referring to these different versions.
Our work to date has focused on the representation
of task context once a task is indicated by the user.
Future directions to investigate include the automatic determination of task boundaries and the integration of user goals within a task to task context information. Work in activity recognition that can
consider the structure of user goals may be helpful in
improving the representation and population of task
contexts (Natarajan et al. 2008) and in enhancing
task-focused user interfaces with additional inference
There is no lack of information available to knowl-
edge workers today. Although current tools make it
easy for knowledge workers to browse and query the
information space with which they must work, these
tools do not support a knowledge worker in manag-
ing the information needed for the tasks the worker
performs. Proper and easy management of informa-
tion relevant to tasks is needed given the high veloc-
ity with which knowledge workers switch tasks.
Without appropriate support, knowledge workers are
burdened with repeatedly creating and recreating the
context they need to get work done.
We have found that the task-focused interface that
leverages knowledge workers’ episodic memory to
form and recall tasks along with a model for task context based on the knowledge workers’ interaction and
activity can provide a means to reduce the friction
and improve the flow of knowledge work.
This work was supported in part by IBM CAS, NSERC
and Tasktop Technologies. We thank the participants
in the studies and the Eclipse Mylyn community for
their ongoing input.
1. Merriam-Webster’s Collegiate Dictionary, 2003.
2. eclipse.org, verified 11/22/13
3. eclipse.org, verified 11/22/13
4. This interface was put into open source as the first version
of the Eclipse Mylyn project.
Belotti, V.; Ducheneaut, N.; Howard, M.; and Smith, I. 2003.
Taking Email to Task: The Design and Evaluation of a Task