Inspiration
A lot of research work has gone into the development of financial stock trading agents, through Deep Learning and Reinforcement Learning. But none has actually been effective in terms of ROI, based on the dynamic and high volatile nature of the financial stock market.
What it does
This project uses Sentiment Analysis to critique online (twitter, blog, headlines etc.) posts by financial analysts, in order to understand the current trend of the market and thereafter proffer financial advice based on the understanding of the market, as captured by the financial analysts.
How we built it
The headlines and stock_name were retrieved from a list of online financial headlines. These were passed to the FinBERT model for inferences (+ve, -ve, neutral). Prompts may or may not include a date. Where a date is included, the model checks through the inference table and advises. Where date is not included, the current date is passed to the model.
Challenges we ran into
- identifying a suitable dataset
- classification of the headlines due to the large dataset
- compute costs
- prompts with current date would require retraining the model with current data, in order to ascertain current state of the market
Accomplishments that we're proud of
Though uncompleted, we are proud of being able to put forward an idea that we believe has huge potentials in providing financial advisory services.
What we learned
Try! It is never too late to start!
What's next for Fina-CHAT
We will be exploring ways to minimize compute costs, and options for addressing the challenge of ascertaining current market trend(s).
Log in or sign up for Devpost to join the conversation.