Compare&Contrast is a prototype system that discovers on the web relevant but different stories for news events. In particular, given a news story, the system aggregates stories describing similar situations to the initial story but involving different entities. The system is domain independent and does not require any knowledge engineering efforts. It dynamically discovers entities comparable to the main entity in the original news event and uses these comparable entities as seeds to retrieve information about comparable stories.
Compare&Contrast helps users understand new situations or solve new problems by comparing and contrasting a given situation with another that is similar but distinct. The system may be used by human users, such as business researchers and journalists to analyze the events of their interests. It may also be used to supply comparable cases for automatically or semi-automatically building case-bases for machine reasoning.
Compare&Contrast is part of our larger goal of exploring intelligent information retrieval to support the user by finding useful, not just similar, information. Underlying this effort is the view that documents with useful information should be similar to the user’s current document in certain aspects, but systematically different from the original document in certain other meaningful aspects.