Learning Opinion Summarizers by Selecting Informative Reviews

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9424–9442 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9424 Learning Opinion Summarizers by Selecting Informative Reviews Arthur Bražinskas1, Mirella Lapata1 and Ivan Titov1,2 1 ILCC, University of Edinburgh 2 ILLC, University of Amsterdam abrazinskas@ed.ac.uk, {mlap,ititov}@inf.ed.ac.uk Abstract Opinion summarization has been tradition- ally approached with unsupervised, weakly- supervised and few-shot learning techniques. In this work, we collect a large dataset of summaries paired with user reviews for over 31,000 products, enabling supervised training. However, the number of reviews per product is large (320 on average), making summariza- tion – and especially training a summarizer – impractical. Moreover, the content of many reviews is not reflected in the human-written summaries, and, thus, the summarizer trained on random review subsets hallucinates. In or- der to deal with both of these challenges, we formulate the task as jointly learning to se- lect informative subsets of reviews and sum- marizing the opinions expressed in these sub- sets. The choice of the review subset is treated as a latent variable, predicted by a small and simple selector. The subset is then fed into a more powerful summarizer. For joint training, we use amortized variational inference and pol- icy gradient methods. Our experiments demon- strate the importance of selecting informative reviews resulting in improved quality of sum- maries and reduced hallucinations. 1 Introduction Summarization of user opinions expressed in on- line resources, such as blogs, reviews, social media, or internet forums, has drawn much attention due to its potential for various information access ap- plications, such as creating digests, search, and report generation (Hu and Liu, 2004; Medhat et al., 2014; Angelidis and Lapata, 2018; Amplayo and Lapata, 2021a). Although significant progress has been observed in supervised summarization in non- subjective context, such as news articles (Rush et al., 2015; Nallapati et al., 2016; See et al., 2017; Lebanoff et al., 2018; Gehrmann et al., 2018; Fab- bri et al., 2019; Laban et al., 2020), modern deep learning methods rely on large amounts of an- notated data that are not readily available in the Verdict If you like the idea of a glass feeder, this is the one to get. It has a lot to offer for the price. Pros • Has a large opening that makes it easy to get in and out of the feeder • Has a nice design that’s easy to clean Cons • The lid is a little flimsy, and it’s not as durable as some of the other models Reviews ... looks just as nice as the glass feeders ||... Very happy with the value, quality and function ... ||... the cheapest most flexible "jar" I’ve ever seen ...||... Nice large opening so it’s easy to pour the sugar water||... This feeder has a nice large opening ... ||... this is the perfect design and size ...||The hummingbirds liked it and had no trouble feeding or perching....||... The main compartment is easy to clean... ||... The top …

Content truncated to 500 words