Sentiment classification with interpolated information diffusion kernels

Information diffusion kernels - similarity metrics in non-Euclidean information spaces - have been found to produce state of the art results for document classification. In this paper, we present a novel approach to global sentiment classification using these kernels. We carry out a large array of experiments addressing the well-known movie review data set of Pang and Lee, a de facto benchmark, comparing information diffusion kernels with a standard RBF kernel machine. Our results show that interpolation of unigram and bigram information is beneficiary for sentiment classification.