Non-Relevance Feedback for Document Retrieval

H. Murata, T. Onoda, and S. Yamada (Japan)


relevance feedback, document retrieval, nonrelevant doc uments, classification learning, oneclass SVM


This paper reports a new document retrieval method which utilizes non-relevant documents. From a large data set of documents, we need to be able to find documents that re late to the subject of interest in as few iterations of testing or checking by a user as possible. In each iteration, a com paratively small batch of documents is evaluated to estab lish there relevance to the subject of interest. This method is called relevance feedback, and it requires a set of rele vant and non-relevant documents. However, the documents initially presented for checking by a user do not always in clude relevant documents. Accordingly, we propose a feed back method using information on non-relevant documents only. We name this method non-relevance feedback. Non relevance feedback selects a set of documents which are discriminated not non-relevant area and near the discrim inant function based on learning result by one-class Sup port Vector Machine (one-class SVM). Results from exper iments show that this method is able to retrieve a relevant document from a set of non-relevant documents effectively.

Important Links:

Go Back