Attentional Extractive Summarization

In this work, a general theoretical framework for extractive summarization is proposed—the Attentional Extractive Summarization framework. Although abstractive approaches are generally used in text today, methods can be especially suitable some applications, and they help with other tasks such as Text Classification, Question Answering, Information Extraction. The proposed approach based on the interpretation of attention mechanisms hierarchical neural networks, which compute document-level representations documents summaries from sentence-level representations, which, turn, computed word-level representations. models under able to automatically learn relationships among document summary sentences, without requiring Oracle systems reference labels each sentence before training phase. These obtained result binary classification process, goal distinguish correct documents. Two different systems, formalized framework, were evaluated CNN/DailyMail NewsRoom corpora, corpora most relevant works summarization. results during evaluation support adequacy our proposal suggest that there still room improvement attentional