In the financial realm, profit generation greatly relies on the complicated task of stock
prediction. Lately, neural methods have shown success in exploiting stock affecting signals
from textual data across news and tweets to forecast stock performance. However, the
dynamic, stochastic, and variably influential nature of text and prices makes it difficult to train
neural stock trading models, limiting predictive performance and profits. To transcend this
limitation, we propose a novel multimodal curriculum learning approach: FinCLASS, which …

On computing the Stock-complexity Sd: First, we compute the bearish, neutral, and bullish
intent of each news item or tweet in the lookback period using class probabilities obtained
via fine-tuned Financial BERT (FinBERT) for English tweets [Araci, 2019] and Chinese
financial text [Rao et al., 2021]. For the US S&P 500 dataset, we use the pre-trained
implementation of FinBERT provided here. 1 We use the logits from the last layer to get the
sentiment scores for bearish, bullish and neutral intent for each English tweet posted in a …