Osamu Furuse, Nobuaki Hiroshima, Setsuo Yamada, Ryoji Kataoka
We have introduced a search engine that can extract opinion sentences relevant to an open-domain query from Japanese blog pages. The engine identifies opinions based not only on positive or negative measurements but also on neutral opinions, requests, advice, and thoughts. To retrieve a number of opinion sentences that a user could reasonably be expected to read, we attempted to extract only ex-plicitly stated writer's opinions at the sentence-level and to exclude quoted or implicational opinions. In our search engine, opinion sentences are identified based on features such as opinion clue expressions, and then, the relevance to the query of each identi-fied opinion sentence is checked. The experimental results for various topics, obtained by comparing the output of the proposed opinion search engine with that of human judgments as to whether the sentences were opinions, showed that the proposed engine has promise as a practical application.
Subjects: 13. Natural Language Processing; 1.10 Information Retrieval
Submitted: Oct 16, 2006