Extractive Summarization of Financial Earnings Call Transcripts: Or: When GREP Beat BERT

Abstract

To date, automatic summarization methods have been mostly developed for (and applied to) general news articles, whereas other document types have been neglected. In this paper, we introduce the task of summarizing financial earnings call transcripts, and we present a method for summarizing this text type essential for the financial industry. Earnings calls are briefing events common for public companies in many countries, typically in the form of conference calls held between company executives and analysts that consist of a spoken monologue part followed by moderated questions and answers. We show that traditional methods work less well in this domain, we present a method suitable for summarizing earnings calls. Our large-scale evaluation on a new human-annotated corpus of summary-worthy sentences shows that this method outperforms a set of strong baselines, including a new one that we propose specifically for earnings calls. To the best of our knowledge, this is the first application of summarization to financial earnings calls transcripts, a primary source of information for financial professionals.