CLIN 2005 Abstracts
  • Question Answering Against Wikipedia
    Sisay Fissaha Adafre (ISLA, University of Amsterdam)
    Maarten de Rijke (ISLA, University of Amsterdam)
    Erik Tjong Kim Sang (ISLA, University of Amsterdam)
    Wikipedia is a free community based on-line encyclopedia. It has become one of the primary reference sites. However, its growth in size and popularity call for novel and effective access methods.

    We present work done for on so-called "undirected informational" queries of the form "Tell me about X" using Wikipedia as our source of answers. We show how the methods developed in the area of question answering (QA) can be used to fulfill the requirements of both the readers and authors. On the one hand, our method provides readers with an alternative view of Wikipedia pages by highlighting the important facts on pages. The same method can also be used to assign relative importance to snippets generated automatically from other sources such as the Web, thus serving as a seed generating mechanism for the development of new Wikipedia pages; authors can benefit from such services while creating or updating a new pages.

    In our experimental work, the above two tasks are reduced to reranking sentences in Wikipedia pages or snippets from other sources. For this purpose, we developed a system that exploits key features of Wikipedia, such as categories, link and layout structures, together with a graph-based reranking method. Manual evaluation of the resulting system shows very acceptable results.