Aspect-Controllable Opinion Summarization

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6578–6593 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 6578 Aspect-Controllable Opinion Summarization Reinald Kim Amplayo Stefanos Angelidis Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh reinald.kim@ed.ac.uk10 Crichton Street, EH8 9AB s.angelidis@ed.ac.uk mlap@inf.ed.ac.uk Abstract Recent work on opinion summarization pro- duces general summaries based on a set of in- put reviews and the popularity of opinions ex- pressed in them. In this paper, we propose an approach that allows the generation of cus- tomized summaries based on aspect queries (e.g., describing the location and room of a hotel). Using a review corpus, we create a synthetic training dataset of (review, summary) pairs enriched with aspect controllers which are induced by a multi-instance learning model that predicts the aspects of a document at dif- ferent levels of granularity. We fine-tune a pre- trained model using our synthetic dataset and generate aspect-specific summaries by modify- ing the aspect controllers. Experiments on two benchmarks show that our model outperforms the previous state of the art and generates per- sonalized summaries by controlling the num- ber of aspects discussed in them. 1 Introduction Consumers oftentimes resort to review websites to inform their decision making (e.g., whether to buy a product or use a service). The proliferation of online reviews has accelerated research on opinion mining (Pang and Lee, 2008), where the ultimate goal is to glean information from reviews so that users can make decisions more efficiently. Opinion mining has assumed several guises in the literature such as sentiment analysis (Pang et al., 2002), as- pect extraction (Hu and Liu, 2004; He et al., 2017), combinations thereof (Mukherjee and Liu, 2012; Pontiki et al., 2016), and notably opinion summa- rization (Hu and Liu, 2006; Wang and Ling, 2016), whose aim is to create a textual summary of opin- ions found in multiple reviews. Text summarization models, both extractive (Narayan et al., 2018; Zheng and Lapata, 2019; Ca- chola et al., 2020) and abstractive (See et al., 2017; Gehrmann et al., 2018; Liu and Lapata, 2019), op- erate under the assumption that salient content is General The room was clean and comfortable. The staff was very friendly and helpful. It was a great location, just a short walk to the beach. There wasn’t much to do in the area, but the food was good. Location The location was great, right on the Boardwalk, and close to the Venice beach. Rooms The room was very clean and the bathroom was very nice. The bathroom had a large separate shower. There was a TV in the room. Location and Rooms The location is great, right on Boardwalk, and the beach is very nice. The room was very clean and the bathroom was very nice and the shower was great. Cleanliness, Location, Room, and Service The staff was very friendly and helpful. The room was very clean, and the bathroom was very nice. It was a great location, right …

Content truncated to 500 words