Learning Representation for Earnings Call Transcript via Structure-Aware Key Insight Extraction

Anonymous

Abstract: Learning representations for earnings call transcripts encounter significant challenges, such as the unreliability of the knowledge encoding process and specific domain-specific requirements in the financial context. To address these challenges, this work proposes a self-supervised transcript representation learning approach that utilizes structural information within transcripts to provide supervision signals. Additionally, it offers concise explanations for each decision made by the neural networks through a redundancy-aware key sentence extractor. Extensive experiments across various downstream tasks, such as risk prediction, information retrieval, and firm similarity analysis, demonstrate the effectiveness of our approach.

Paper Type: long

Research Area: Information Retrieval and Text Mining

Contribution Types: Model analysis & interpretability, Data resources

Languages Studied: English