Template-based Abstractive Microblog Opinion Summarization

Template-based Abstractive Microblog Opinion Summarization Iman Munire Bilal1,4, Bo Wang2,4, Adam Tsakalidis3,4, Dong Nguyen5, Rob Procter1,4, Maria Liakata1,3,4 1Department of Computer Science, University of Warwick, UK 2Center for Precision Psychiatry, Massachusetts General Hospital, USA 3School of Electronic Engineering and Computer Science, Queen Mary University of London, UK 4The Alan Turing Institute, London, UK 5Department of Information and Computing Sciences, Utrecht University, The Netherlands {iman.bilal|rob.procter}@warwick.ac.uk bwang29@mgh.harvard.edu {atsakalidis|mliakata}@qmul.ac.uk d.p.nguyen@uu.nl Abstract We introduce the task of microblog opinion summarization (MOS) and share a dataset of 3100 gold-standard opinion summaries to fa- cilitate research in this domain. The dataset contains summaries of tweets spanning a 2-year period and covers more topics than any other public Twitter summarization dataset. Summaries are abstractive in nature and have been created by journalists skilled in sum- marizing news articles following a template separating factual information (main story) from author opinions. Our method differs from previous work on generating gold-standard summaries from social media, which usually involves selecting representative posts and thus favors extractive summarization models. To showcase the dataset’s utility and chal- lenges, we benchmark a range of abstractive and extractive state-of-the-art summarization models and achieve good performance, with the former outperforming the latter. We also show that fine-tuning is necessary to improve performance and investigate the benefits of using different sample sizes. 1 Introduction Social media has gained prominence as a means for the public to exchange opinions on a broad range of topics. Furthermore, its social and tem- poral properties make it a rich resource for policy makers and organizations to track public opin- ion on a diverse range of issues (Procter et al., 2013; Chou et al., 2018; Kalimeri et al., 2019). However, understanding opinions about different issues and entities discussed in large volumes of posts in platforms such as Twitter is a difficult task. Existing work on Twitter employs extractive summarization (Inouye and Kalita, 2011; Zubiaga et al., 2012; Wang et al., 2017a; Jang and Allan, 2018) to filter through information by ranking and selecting tweets according to various crite- ria. However, this approach unavoidably ends up including incomplete or redundant information (Wang and Ling, 2016). To tackle this challenge we introduce Micro- blog opinion summarization (MOS), which we define as a multi-document summarization task aimed at capturing diverse reactions and stances (opinions) of social media users on a topic. While here we apply our methods to Twitter data readily available to us, we note that this summarization strategy is also useful for other microblogging platforms. An example of a tweet cluster and its opinion summary is shown in Table 1. As shown, our proposed summary structure for MOS sepa- rates the factual information (story) from reactions to the story (opinions); the latter is further divided according to the prevalence of different opinions. We believe that making combined use of stance identification, sentiment analysis and abstrac- tive summarization is a challenging but valuable direction in aggregating opinions expressed in microblogs. The availability of high quality news article datasets has meant that recent advances in …

Content truncated to 500 words