Comparative Opinion Summarization via Collaborative Decoding

Findings of the Association for Computational Linguistics: ACL 2022, pages 3307 - 3324 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Comparative Opinion Summarization via Collaborative Decoding Hayate Iso� Xiaolan Wang� Stefanos Angelidis� Yoshihiko Suhara� �Megagon Labs �University of Edinburgh {hayate,xiaolan,yoshi}@megagon.ai s.angelidis@ed.ac.uk Abstract Opinion summarization focuses on generating summaries that reflect popular subjective infor- mation expressed in multiple online reviews. While generated summaries offer general and concise information about a particular hotel or product, the information may be insuffi- cient to help the user compare multiple differ- ent choices. Thus, the user may still strug- gle with the question “Which one should I pick?” In this paper, we propose the compar- ative opinion summarization task, which aims at generating two contrastive summaries and one common summary from two different can- didate sets of reviews. We develop a compar- ative summarization framework COCOSUM, which consists of two base summarization models that jointly generate contrastive and common summaries. Experimental results on a newly created benchmark COCOTRIP show that COCOSUM can produce higher-quality contrastive and common summaries than state- of-the-art opinion summarization models. The dataset and code are available at https:// github.com/megagonlabs/cocosum. 1 Introduction Widely available online customer reviews help users with decision-making in a variety of domains (e.g., hotel, restaurant, or company). After creating a list of candidate choices based on initial condi- tions (e.g., area, price range, restaurant type), the user often has to compare a few choices in depth by carefully reading the reviews to make a final de- cision. However, it is time-consuming and difficult for the user to detect differences and similarities between the candidates, as those pieces of informa- tion are often scattered in different reviews. The recent success of neural summarization tech- niques and the growth of online review platforms led to establishing the field of multi-document opin- ion summarization (Chu and Liu, 2019; Bražinskas et al., 2020b; Amplayo and Lapata, 2020; Suhara Let’s go to the ! ∩ \ Reviews of Reviews of Reviews of � \ Reviews of � � � � � � Single'Entity'Opinion'Summarization ———— ———— ——— ———— ———— ——— ———— ———— ——— The hotel is great! The location is good. The staff is friendly. ———— ———— ——— ———— ———— ——— ———— ———— ——— Summarize Great location and awesome staff! Summarize Which hotel should I pick..? Comparative'Opinion'Summarization ———— ———— ——— ———— ———— ——— ———— ———— ——— ———— ———— ——— ———— ———— ——— ———— ———— ——— Great location and awesome staff! If you wish to go shopping, this is a perfect hotel! A minute walk from beach! Contrastive Contrastive Common � � � � Figure 1: Overview of the comparative opinion sum- marization task. The model takes two set of reviews about different entities to generate two contrastive opin- ion summaries, which contain distinctive opinions, and one common opinion summary, which describes com- mon opinions between the two entities. et al., 2020; Iso et al., 2021). The goal of multi- document opinion summarization is to generate a summary that represents salient opinions in input …

Content truncated to 500 words