A model and system for automating user tasks on the World-Wide Web.

摘要

In this dissertation, we investigate the applications and extensions of the Programming By Demonstration (PBD) paradigm to the distributed hypertext medium exemplified by the World-Wide Web. We analyze existing approaches to accommodating repetitive tasks on the WWW, and proceed to identify the primary challenges of PBD on the WWW: dealing with change, reasoning about state, overcoming information opaqueness, generalizing to automate choice and iteration, and parameterization. We present a general model of a user task on the Web that can be applied in other HCI domains and applications. We describe the design and implementation of our WebMacros system for automating repetitive Web tasks by demonstration, discussing design decisions, implementation and software artifacts. We discuss the results and amplications of our user-centered evaluation of the system, covering subjective evaluation by the participants and performance on tasks. The results of the user-centered evaluation support our primary hypothesis that automating common tasks brings statistically significant improvements in user performance and satisfaction. In practice, increase in performance varied from twofold to fourfold, depending on the task. We also describe the case studies of using the WebMacros system as a Web developer's assistant on two real-world scenarios.; A significant challenge over the course of this work proved to be determining the identity of a hypertext page, and classifying pages into groups based on their structure, while ignoring the dynamic content. We developed a novel representation of hypertext page structure based on fingerprints of path expressions in the parse tree, and evaluated three similarity measures to determine if any of them is appropriate for identifying and clustering pages based on structure. Based on 60 daily data sets from 2 popular commercial sites, we identified Jaccard's similarity as the best measure for page clustering. Implications of this work allow the construction of by-example Web crawlers and page-structure verification scripts.