Query-based summarization for Indonesian news articles

Query-based summarization not only finds important information, but also relevant information to query given by users. This research aims to determine the best technique and configurations of query-based summarizer for Indonesian news articles. We develop two techniques adapted from the first rank summarizers on Document Understanding Conferences which are using Wordnet-based and word cluster-based sentence scoring. We also compared three sentence selection methods i.e.: Maximal Marginal Relevance (MMR), Integer Linear Programming (ILP), and dynamic programming. Based on evaluation using ROUGE-2, the first technique which is adapted from NUS summarizer has the best recall of 0.1232, while the second one (which is adapted from IIIT Hyderabad summarizer) 0.1555. Both techniques give the best result when using MMR as sentence selection method.