Topic Models for Comparative Summarization

This paper aims to sum up our work in the area of comparative summarization and to present our results. The focus of comparative summarization is the analysis of input documents and the creation of summaries which depict the most significant differences in them. We experiment with two well known methods – Latent Semantic Analysis and Latent Dirichlet Allocation – to obtain the latent topics of documents. These topics can be compared and thus we can learn the main factual differences and select the most significant sentences into the output summaries. Our algorithms are briefly explained in section 2 and their evaluation on the TAC 2011 dataset with the ROUGE toolkit is then presented in section 3.