Short text classification, such as snippets, search queries, micro-blogs and product reviews, is a challenging task mainly because short texts have insufficient co-occurrence information between words and have a very spare document-term representation. To address this problem, we propose a novel multi-view classification method by combining both the original document-term representation and a new graph based feature representation. Our proposed method uses all documents to construct a neighbour graph by using the shared co-occurrence words. Multi-Dimensional Scaling (MDS) is further applied to extract a low-dimensional feature representation from the graph, which is augmented with the original text features for learning. Experiments on several benchmark datasets show that the proposed multi-view classifier, trained from augmented feature representation, obtains significant performance gain compared to the baseline methods. © Springer-Verlag 2013.
CITATION STYLE
Long, G., & Jiang, J. (2013). Graph based feature augmentation for short and sparse text classification. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8346 LNAI, pp. 456–467). https://doi.org/10.1007/978-3-642-53914-5_39
Mendeley helps you to discover research relevant for your work.