Deep Learning Based Extractive Text Summarization: Approaches, Datasets and Evaluation Measures

• Published in: 2019 Sixth International Conference on Social Networks Analysis, Management and Security (SNAMS) pp. 204 - 210
• Main Authors: Suleiman, Dima, Awajan, Arafat A.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2019
• Subjects: Attention Mechanizm; Convolutional Neural Network; Decoding; Deep learning; Extractive text summarization; Feature extraction; GRU; Recurrent Neural Network; Recurrent neural networks; Restricted Boltzmann Machine; ROUGE; Semantics; Training; Variation Auto-Encoder; LSTM
• DOI: 10.1109/SNAMS.2019.8931813

Recently, the number of online documents witness huge increase in volume. Thus, these documents need to be summarized in order to be effective. This paper reviews the most recent extractive text summarization approaches that are based on deep learning techniques. These approaches are classified into three categories based on deep learning techniques which are Restricted Boltzmann Machine, Variation Auto-Encoder and Recurrent Neural Network. The mostly used datasets for extractive summarizations are Daily Mail and DUC2002. Moreover, ROUGE is the mainly used evaluation measure to assess the quality of the extractive summarization process. The results show that SummaRuNNer approach which is based on Gated Recurrent Unit Recurrent Neural Network achieved the highest values for ROUGE1, ROUGE2 and ROUGE-L over Daily Mail datasets. On the other hand, the approach that is based on Recurrent Neural Network achieved the best results over DUC2002 datasets in term of ROUGE1 and ROUGE2.