Augmenting human intelligence via externalized knowledge representation and intelligent information retrieval

Abstract

With a representation of a user's mental model in hand, a computer system can continuously query against a knowledge base of past work sessions or information on the Web and generate a set of recommended resources for the user to consider. In this thesis, I have developed an interface and a representation that allows a computer system to build a model of a user's intent and generate recommendations. I have designed, prototyped, and deployed the Mental Model Browser, a web application that infers a user's intent during a Web browsing session and provides recommendations for related URLs. The application includes a web browser extension for recording the URLs a user visits and a feedback interface that hosts a dialogue between the computer system and the user. The Mental Model Browser identifies important concepts in the session by leveraging an API provided by the Delicious web bookmarking service, a rich data corpus of crowd-sourced web page tags. The identified concepts are presented to the user to confirm their validity and trigger a query for recommended web pages. I conducted a pilot study with 22 active participants who engaged in 56 web browsing sessions. The results of the study show that users were able to readily adapt to the workflow of the application. Users reported quickly discovering how to help shape the system's model of the session through the use of tags. Several users reported receiving valuable recommendations that they did not find through search alone. Finally, I lay out visions for near-future technologies such as multi-modal knowledge capture and activation, as well as knowledge-based social networks that are enabled by the concepts I have explored in this thesis.