Abstract
This paper describes an application of SVMs (Support Vector Machines) to interactive document retrieval using active learning. We show that SVM-based retrieval have an association with conventional relevance feedback (Rocchio-based method) by comparative analysis of relevance evaluation. We propose a cosine kernel which has the meaning equal with cosine similarity suitable for SVM-based interactive document retrieval from the analysis. We confirm the effectiveness of this method and experimentally compared it with conventional system in the Boolean, TF and TFIDF as representations of document vectors. The experimental results using TREC data set shows that cosine kernel is effective against all document representations, especially TF representation.
