Analytics Challenge Webinar Series - FinBERT for Financial Sentiment Classification

This is part 2 of a 5-part webinar series featuring leading academic experts in big data analytics, information systems, machine learning, data mining, and large-scale data processing. Students interested in the CBSA-Wisers Analytics Challenge @HKUST 2021/22 are highly encouraged to attend the webinars. 

ABSTRACT

Contextual pre-trained language models, such as BERT (Devlin et al., 2019), have made a significant breakthrough in various natural language processing (NLP) tasks by training on a large scale of unlabeled text resources. The financial sector also accumulates a large amount of financial communication text. In this work, we pre-train a financial domain-specific BERT model, FinBERT, using a large corpus of financial communication. We empirically show that FinBERT outperforms simpler NLP tasks and the generic domain BERT model. 

KEYWORDS

Natural Language Processing, Machine Learning, Deep Learning, Textual Analysis, Sentiment Classification, Informativeness, Earnings Conference Call