Graph Based Feature Augmentation for Short and Sparse Text Classification

Publication Type:
Conference Proceeding
Lecture Notes in Computer Science, 2013, 8346 pp. 456 - 467
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail2013003626OK.pdf447.9 kB
Adobe PDF
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.
Please use this identifier to cite or link to this item: