A Semi-supervised Topic-Driven Approach for Clustering Textual Answers to Survey Questions