The bootstrapping method is known as an application of the Page-rank technique for documents and words. The technique calculates the score of the words by mutually propagating the score of the words and the documents. However, sometimes the result is far away from the initial query word. The problem is known as gtopic drifth. This paper proposes to restrict the words to be to the top t words in the process of bootstrapping. The method is simpler than the technique known so far. The method is applied for the real bankruptcy information documents to extract the bankruptcy causes strongly related to the query. It is confirmed that the method prevents the topic drift.
展开▼