Guided summarization for Indonesian news articles

The development of online news media grew in number in Indonesia. One technique of news articles summarization is guided summarization where the summary should contain important aspect information. Guided summarization techniques have been developed in the Text Analysis Conference (TAC) 2011 and one of the best methods is SWING by Jun-ping, et al. The purpose of this study is to adapt the methods of SWING system to Indonesian news articles as well as integrating with News Aggregator system. In this research, the experiments have purpose to determine the best features and system configuration when adapted to Indonesian news articles. ROUGE-2 and ROUGE-SU4 is used to evaluate the results of the summary where a summary of the system results compared to the human-made summaries. The best system configuration produces summary with evaluation of ROUGE-2 0,31 and ROUGE-SU4 0,22 which is very close to the human-made summaries with a value of ROUGE-2 0,32 and ROUGE-SU4 0.24. In addition, the update summarization component can be run by giving a summary of updates without repeating the information. Adaptation from SWING system to Indonesian news articles is employing features such as sentence length (SL), category relevance score (CRS), category KL-Divergence (CKLD), bigram DFS (BDFS), Top n NE corpus, Top n NE topic, quote sentence removal, and building SVR model for each news category.