Generating Topic-Oriented Summaries Using Neural Attention

Proceedings of NAACL-HLT 2018, pages 1697–1705 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Generating topic-oriented summaries using neural attention Kundan Krishna Adobe Research kunkrish@adobe.com Balaji Vasan Srinivasan Adobe Research balsrini@adobe.com Abstract Summarizing a document requires identifying the important parts of the document with an objective of providing a quick overview to a reader. However, a long article can span sev- eral topics and a single summary cannot do justice to all the topics. Further, the interests of readers can vary and the notion of impor- tance can change across them. Existing sum- marization algorithms generate a single sum- mary and are not capable of generating mul- tiple summaries tuned to the interests of the readers. In this paper, we propose an attention based RNN framework to generate multiple summaries of a single document tuned to dif- ferent topics of interest. Our method outper- forms existing baselines and our results sug- gest that the attention of generative networks can be successfully biased to look at sentences relevant to a topic and effectively used to gen- erate topic-tuned summaries. 1 Introduction Automatic text summarization is the task of gen- erating/extracting short text snippet that embodies the content of a larger document or a collection of documents in a concise fashion. Traditionally, researchers have used extractive methods for sum- marization - where a set of sentences is selected from an article and concatenated as-is to form the summary. Extractive methods are limited by their inability to paraphrase the content of the article using new sentences and hence are very different from the summaries created by humans, who para- phrase a given article to generate summaries. Re- cent works on summarization have come up with sequence-to-sequence models for generating sum- maries in a word-by-word fashion, thus ‘generat- ing’ new sentences. As articles get longer, it might span several topics and therefore, a single summary is often insufficient to satisfy the interests of the reader. In these cases, it is desirable to produce sum- maries that are aligned to the topical interests of reader to enable better consumption. Extractive summarization uses sentence-level features (Yang et al., 2017) that have been leveraged for pro- ducing query-focused or topic-based summaries. However, for RNN-based frameworks, such tuned summary generation is non-obvious due to the ab- sence of explicit content based features. To extend the advantages of abstractive summa- rization and generate topic-tuned summaries, we propose a neural encoder-decoder based frame- work which takes an article along with a topic of interest as input and generates a summary tuned to the target topic of interest. Our method works by training a neural framework to pay higher at- tention to parts of the input articles relevant to given topic. To overcome the lack of dataset con- taining articles with multiple topic-oriented sum- maries, we use a novel approach to artificially create a topic-centric training corpus from the CNN/Dailymail dataset (Hermann et al., 2015; Nallapati et al., 2016) for training our model. Table 1 shows …

Content truncated to 500 words