Graph based feature augmentation for short and sparse text classification
- Publication Type:
- Conference Proceeding
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2013, 8346 LNAI (PART 1), pp. 456 - 467
- Issue Date:
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.
Please use this identifier to cite or link to this item: