An AI-powered Sentiment Indicator tailored for online crypto communities and meme trading.
Containing several models finetunned using data primarily extracted from reddit, aimed to better understand modern terminology used by contemporary traders and the communities surrounding them.
All the models make use of a DistilBert backbone that was pre-trained using a masked language modeling objective on a collection of crypto-oriented datasets extracted from the aforementioned social network, then a multi layer perceptron (MLP) head is used to further finetune the models on various of the most popular sentiment analysis datasets (such as "financial phrasebank", "Stanford Sentiment Treebank", etc.). Currently supporting sentiment classification (meassuring by classes e.g. either negative or positive) and sentiment regression (mesuring from 0: negative, all the way up to 1: positive).
Check out the Datasets on Kaggle!
Example: Inference on Bitcoin language data extracted from reddit through 2021
For reference: 0=Negative, 1=Neutral, 2=Positive
Average daily sentiment from 2021/01/01 to 2021/05/01
Average daily sentiment from 2021/05/02 to 2021/08/16
