Sentence structure-based summarization for Indonesian news articles

• Published in: 2017 International Conference on Advanced Informatics, Concepts, Theory, and Applications (ICAICTA) pp. 1 - 6
• Main Authors: Reztaputra, Raihannur, Khodra, Masayu Leylia
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.08.2017
• Subjects: clustering; Data mining; dependency tree; Earthquakes; extraction; Feature extraction; Fuses; Geology; Information retrieval; maximal marginal relevance (MMR); part-of-speech (POS) tag; Redundancy; selection; sentence structure; summarization
• DOI: 10.1109/ICAICTA.2017.8090983

Automatic multi-document summarization may help news readers retrieve information from digital news media efficiently. The summarizer create a concise summary containing important information from a collection of articles, enabling readers to read only one text to gain information from multiple text sources. Reflecting on previous researches, we propose an automatic summarization system using sentence structure information (subject, object, predicate, complement). The system consists of four main components, preprocessing and feature extraction, sentence structure information extraction, sentence clustering and fusion, and sentence selection. The system will extract the necessary information using dependency tree, cluster sentences using Density Based Spatial Clustering for Application with Noise (DBSCAN), fuse sentences with sentence structure information graph, and select sentences using Maximal Marginal Relevance (MMR). The evaluation shows that the proposed system performs with 0.276 average ROUGE-2, with many chances of improvements. Sentence structure extractor has 0.75 f1-measure score.