Dissertation Defense

Effective Faceted Browsing

Manish Singh

ABSTRACT: Faceted browsing is a popular paradigm for end-user data access. It is, at present, the defacto
standard for almost all e-commerce. A typical faceted interface has two main component panels:
a query panel and a result panel. Faceted browsing is primarily designed to help users quickly get to
a specific item if they know the characteristics they are looking for. However, limitations in the query
and the result panel deter effective faceted browsing, especially for users unfamiliar with the data.
In this dissertation, we highlight two such limitations, one each in the query and the result panel.
We propose add-on extensions to address each of these limitations. In a faceted interface, users
progressively select a sequence of facet values to get to their desired result set, which is called an
exploration path. If the dataset is high-dimensional, the query panel can only show a few of those
dimensions as queriable facets. Users cannot see in the query panel the overall space of available
exploration paths, and thus end up choosing an inferior exploration path. Many users have difficulty
in selecting or understanding an exploration path when there are many non-queriable facets and
query panel has very limited information of interaction between facets. We address this limitation by
showing users an integrated summary of facet interaction that summarizes their chosen exploration
path, and by presenting a two-phased faceted interface that provides users a facetwise way to compare
the available exploration paths. The result panel that is normally used for presenting relational tuples,
including faceted interface, cannot support fast browsing. When a user scrolls fast through data
having alphanumeric values, then everything seems like a fast changing blur. To help the user get a
quick sense of data, we propose a novel variable-speed scrolling interface, which provides the user
a good impression of the data through selected representative tuples that are chosen based on the
user's scrolling speed and browsing history.

Sponsored by

H.V. Jagadish