Improving Update Summarization by Revisiting the MMR Criterion

This paper describes a method for multi-document update summariza-tion that relies on a double maximization criterion. A Maximal Marginal Relevance like criterion, modified and so called Smmr, is used to select sentences that are close to the topic and at the same time, distant from sentences used in already read documents. Summaries are then generated by assembling the high ranked material and applying some ruled-based linguistic post-processing in order to obtain length reduction and main-tain coherency. Through a participation to the Text Analysis Confer-ence (TAC) 2008 evaluation campaign, we have shown that our method achieves promising results.