[2D1-GS-9] Natural language processing, information retrieval: Support technology

[2D1-GS-9-01] Proposal of Hierarchical Attention Model for Summarization and Extension Method for Pointer Generator Network

Hiroki Teranishi¹, 〇Makoto Okada¹, Naoki Mori¹ (1. Osaka Prefecture University)

Keywords:Text Summarization, AI, Neural Network

Automatic summarization is one of the key issues. Therefore, research on document summarization has been active. In the fields, methods using neural networks are attracting such as an Attention based Summarization model and a Pointer Generator Network (PGN) model. Such models treat a set of multiple sentences as one sentence. Therefore, it is not possible to process each sentence individually and to summarize considering the relationship between the sentences. Therefore, an architecture of processing sentences individually and capturing the dependencies of sentences may lead to the improvement of document summarization. In this research, we propose a hierarchical attention model, which makes each sentence vector from input document. In addition, we propose a model that incorporates a gated convolutional neural network or self-attention to obtain the relation between sentences. Moreover, we extend our architecture to use a PGN. Through the experiments, we show the effectiveness of our proposed model.