Neural Ensemble for Organizing Business News-Streams for Corporate Libraries

Corporate libraries gathering business intelligence have to process an overwhelming stream of business news. Automatic categorization of these news articles based on their primary event ('NewsType') can significantly aid in better organization and disbursement. We propose an ensemble of a BiLSTM neural network and a Pre-trained Language Model (PLM) based zero-shot classification approach for solving the categorization task. We also show that fine-tuning the NLI model helps in better accuracy on the task. We evaluate our approaches on two datasets while comparing with multiple baselines.