Extractive Text Summarization with Latent Topics using Heterogeneous Graph Neural Network

Anonymous

Abstract: This paper presents a heterogeneous graph neural network (HeterGNN) model for extractive text summarization (ETS) by using latent topics to capture the important content of input documents. Specifically, topical information has been widely used as global information for sentence selection. However, most of the recent approaches use neural models, which lead the training models more complex and difficult for extensibility. In this regard, this study presents a novel graph-based ETS by adding a new node of latent topics into HeterGN for the summarization (TopicHeterGraphSum). Specifically, TopicHeterGraphSum includes three types of semantic nodes (i.e., topic-word-sentence) in order to enrich the cross-sentence relations. Furthermore, an extended version of TopicHeterGraphSum for multi documents extraction is also taken into account to emphasize the advantage of the proposed method. Experiments on benchmark datasets such as CNN/DailyMail and Multi-News show the promising results of our method compared with state-of-the-art models.

Paper Type: short