Deep-learning impact predictions based on sectoral news

Deep-learning impact predictions based on sectoral news

The Problem

Identification of drivers of ca. 3000 junior mining companies stock prices within up to 100 new daily specialized articles / news related to production and drilling results, permissions.

Our Solution

Cloud-based architecture that consists of asynchronous data structuring and analytics pipelines that constantly track latest news which are the basis of stock prices predictions. Due to complexity and length of drilling results articles the forecasting is based on the Longformer, a modified Transformer architecture with a self-attention operation that makes it easy to process long documents.

Outcome

Real-time automated data processing pipelines on production and drilling results, permissions providing email and dashboard alerts including stock prices predictions with up to 70% accuracy in direction forecasts. The results allow to quickly build trading strategies.

Key Takeaway

Analyzed 100% of the news data, evaluated its impact on price with up to 70% accuracy.

Technology Keywords

Python, DL/Transformer, ETL, Cloud-based, Forecasting