OpinionDigest: A Simple Framework for Opinion Summarization

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5789–5798 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 5789 OPINIONDIGEST: A Simple Framework for Opinion Summarization Yoshihiko Suhara∗1 Xiaolan Wang∗1 Stefanos Angelidis2 Wang-Chiew Tan1 1Megagon Labs 2University of Edinburgh {yoshi,xiaolan,wangchiew}@megagon.ai s.angelidis@ed.ac.uk Abstract We present OPINIONDIGEST, an abstrac- tive opinion summarization framework, which does not rely on gold-standard summaries for training. The framework uses an Aspect-based Sentiment Analysis model to extract opinion phrases from reviews, and trains a Transformer model to reconstruct the original reviews from these extractions. At summarization time, we merge extractions from multiple reviews and select the most popular ones. The selected opinions are used as input to the trained Trans- former model, which verbalizes them into an opinion summary. OPINIONDIGEST can also generate customized summaries, tailored to specific user needs, by filtering the selected opinions according to their aspect and/or sen- timent. Automatic evaluation on YELP data shows that our framework outperforms com- petitive baselines. Human studies on two cor- pora verify that OPINIONDIGEST produces informative summaries and shows promising customization capabilities1. 1 Introduction The summarization of opinions in customer reviews has received significant attention in the Data Min- ing and Natural Language Processing communities. Early efforts (Hu and Liu, 2004a) focused on pro- ducing structured summaries which numerically aggregate the customers’ satisfaction about an item across multiple aspects, and often included repre- sentative review sentences as evidence. Consider- able research has recently shifted towards textual opinion summaries, fueled by the increasing suc- cess of neural summarization methods (Cheng and Lapata, 2016; Paulus et al., 2018; See et al., 2017; Liu and Lapata, 2019; Isonuma et al., 2019). ∗Equal contribution. 1Our code is available at https://github.com/ megagonlabs/opiniondigest. Opinion summaries can be extractive, i.e., cre- ated by selecting a subset of salient sentences from the input reviews, or abstractive, where summaries are generated from scratch. Extractive approaches produce well-formed text, but selecting the sen- tences which approximate the most popular opin- ions in the input is challenging. Angelidis and Lap- ata (2018) used sentiment and aspect predictions as a proxy for identifying opinion-rich segments. Ab- stractive methods (Chu and Liu, 2019; Braˇzinskas et al., 2019), like the one presented in this paper, attempt to model the prevalent opinions in the input and generate text that articulates them. Opinion summarization can rarely rely on gold- standard summaries for training (see Amplayo and Lapata (2019) for a supervised approach). Recent work has utilized end-to-end unsupervised architec- tures, based on auto-encoders (Chu and Liu, 2019; Braˇzinskas et al., 2019), where an aggregated rep- resentation of the input reviews is fed to a decoder, trained via reconstruction loss to produce review- like summaries. Similarly to their work, we assume that review-like generation is appropriate for opin- ion summarization. However, we explicitly deal with opinion popularity, which we believe is crucial for multi-review opinion summarization. Addition- ally, our work is novel in its ability to explicitly control the sentiment and aspects of selected opin- ions. The aggregation …

Content truncated to 500 words