The paper is devoted to the issue of clustering short texts, which are free answers gathered during brain storming seminars. Those answers are short, often incomplete, and highly biased toward the question, so establishing a notion of proximity between texts is a challenging task. In addition, the number of answers is counted up to hundred instances, which causes sparsity. We present three text clustering methods in order to choose the best one for this specific task, then we show how the method can be improved by a semantic enrichment, including neural-based distributional models and external knowledge resources. The algorithms have been evaluated on the unique seminar's data sets.
展开▼
