Item

Abstract

Ambiguous information needs expressed in a limited number of keywords
often result in long-winded query sessions and many query reformulations.
In this work, we tackle ambiguous queries by providing automatically gen-
erated semantic aspects that can guide users to satisfying results regarding
their information needs. To generate semantic aspects, we use semantic an-
notations available in the documents and leverage models representing the
semantic relationships between annotations of the same type. The aspects in
turn provide us a foundation for representing text in a completely structured
manner, thereby allowing for a semantically-motivated organization of search
results. We evaluate our approach on a testbed of over 5,000 aspects on Web
scale document collections amounting to more than 450 million documents,
with temporal, geographic, and named entity annotations as example dimen-
sions. Our experimental results show that our general approach is Web-scale
ready and finds relevant aspects for highly ambiguous queries.