Summarization of financial reports with AMUSE

This paper reports an approach for summarizing financial texts that combine genetic algorithms and neural document modeling. We treat summarization as the task of binary classification of sentences. Financial reports in the shared data of the FNS workshop are very long, have many sections, and are written in “financial” language using various special terms, numerical data, and tables. Our approach follows two main stages: (1) filtering the most irrelevant information with help of a supervised state-of-the-art summarizer and (2) extracting the most relevant sentences from the selected sentences in stage (1), using a novel deep neural model. As all participants of the Financial Narrative Summarization (FNS 2021) shared task, we used FNS 2021 dataset for training and evaluation.