home  .  people  .  projects  .  publications  .  internal page

augur:
probabilistic calendars

Informal communications are brief, unplanned, face-to-face meetings that are often supported by electronic calendaring systems. In addition to informing colleagues where one is at the moment, electronic calendars also indicate where one is going to be during the rest of the day. This information allows colleagues to seek out someone for a face-to-face conversation without having to wait for her to return to the office. A practice called “ambushing” occurs when someone waits at the location where they know the colleague will be at that time.

Unfortunately, electronic calendars exhibit common inaccuracies, including but not limited to:

• Incorrect recurrence boundaries. Users may overestimate or underestimate the end dates of a recurring appointment, causing free times to be represented as busy or vice versa.

• All day events supercede others. Users may have an “all-day” appointment such as a conference or vacation that renders the routine appointments for that day obsolete.

• Conflicting events. There may be two events scheduled for the same time, of which the calendar’s owner can only select one.

• Infrequently attended events. A recurring event may be scheduled, but the event is attended only sporadically by the owner.

These inaccuracies make it difficult for coworkers to use calendars as a means of initiating informal chats.

We are investigating the use of techniques from the field of intelligent systems to enhance the accuracy of current electronic calendar systems and promote the practice of “ambushing” colleagues for informal collaboration. In particular, we have developed Bayesian belief networks to represent a user's decision to attend. The models learn about each user by looking at his/her history of attending past events, and can generate predictions of attendance at future events.

In addition, we are interested in exposing the underlying models of our system to users so that they may diagnose and adjust them to suit their own personal attendance preferences. Academic schedules are prone to rapid turnover during changes in terms, and learning algorithms do not receive enough examples to reliably generate accurate predictions. Interactivity allows users to align the models to their own mental model of their attendance habits, and learning algorithms are then able to refine the models to reflect actual behavior.

See also:
augur
augur: social schemes

home  .  people  .  projects  .  publications  .  internal page