ConvoSumm: Conversation Summarization Benchmark and Improved Abstractive Summarization with Argument Mining

While online conversations can cover a vast amount of information in many\ndifferent formats, abstractive text summarization has primarily focused on\nmodeling solely news articles. This research gap is due, in part, to the lack\nof standardized datasets for summarizing online discussions. To address this\ngap, we design annotation protocols motivated by an\nissues--viewpoints--assertions framework to crowdsource four new datasets on\ndiverse online conversation forms of news comments, discussion forums,\ncommunity question answering forums, and email threads. We benchmark\nstate-of-the-art models on our datasets and analyze characteristics associated\nwith the data. To create a comprehensive benchmark, we also evaluate these\nmodels on widely-used conversation summarization datasets to establish strong\nbaselines in this domain. Furthermore, we incorporate argument mining through\ngraph construction to directly model the issues, viewpoints, and assertions\npresent in a conversation and filter noisy input, showing comparable or\nimproved results according to automatic and human evaluations.\n