Towards a generic approach to providing proactive task support

Abstract

An increasing amount of task support resources has been placed online in a variety of forms such as help, references, wizards, cue-cards, examples, and interactive tutorials, to reduce users' need for training and task support from human experts. However, typically, users have to leave their task context and search for task support with a query in a separate context on a trial-and-error basis. There are several problems with this approach: Users may not be able to formulate proper queries; the process is often fruitless and frustrating because users are left alone to navigate through the query result. Most importantly, when given a new system to use, users have little desire to learn if they can use methods that they already know, regardless of their efficacy.

The basic idea in this research is to provide relevant task support for a computer-based application in a proactive but non-obtrusive manner. The task support system operates as an intelligent agent, which monitors the task progress and suggests relevant online resources continuously based on the user's task context. Advice and relevant domain knowledge are then displayed continuously in separate and persistently present advice windows side-by-side to the task window, and the display is updated at short intervals, without interfering with the user's task. An artificial neural network is used to identify the current task with the user's task progress as input. The artificial neural network recognizes one or more plausible tasks to approximate a user's task so that a range of relevant advice can be offered for the user's selective use.

A prototype, called Telephone Triage Assistant (TTA), has been built to support novice nurses in identifying diseases based on a phone interview with a patient. The usability of TTA has been assessed through a field study. Results show that, on average, 41% of the subjects' task time was spent on TTA and up to 70% of their questions appeared to be influenced by TTA. Although the post-task questionnaire data shows that TTA was perceived easy to use and useful for the task, it also reveals that subjects' perception of the continuous update was barely positive.

The proposed approach is expected to significantly alleviate the problems associated with conventional task support in the following ways: Users do not need to initiate search by themselves because relevant task support is presented to them at the moment of need. In addition, because of the continuous information update, advice to the user can be refined incrementally according to the task progress. Finally, users have full control over the task support with the options of either following up on the ask support or ignoring it completely. The main thrust of the proposed approach is its potential for enhancing access to online resources and literally bringing them to knowledge workers' fingertips.