The consortium of the DeepFinance project created during the period 28/7/2020-27/7/2023 an integrated platform for semantic analysis and sentiment extraction from social network flows with Deep Learning methods, but also integrated financial portfolio management tools, which are able to fuse multimodal information extracted from diverse (heterogeneous) data sources. The agencies contributed to the completion of the platform Aristotle University of Thessaloniki, DataScouting. and SpeedLab GmbH

More specifically, the DeepFinance project consortium developed deep learning methods and models for portfolio management using data from stock market trades. For this purpose, deep reinforcement learning methods were extensively used, while deep learning models were developed for the semantic analysis of the opinions relayed on social media about the various goods, organizations and companies, financial indices, shares and investment products.

Innovative deep learning methods for information fusion from heterogeneous information sources were developed, as well as deep learning methods for extracting semantic information from partially labeled data streams from social networks, while the use of multiple models was also examined to further increase the accuracy of performing semantic analysis tasks and forecasting various aspects of stock markets. Another innovation of the project is the accurate simulation of stock market activity in reinforcement learning agent training environments. The developed simulation environments support the execution of Limit Orders, which is not possible with the existing environments that support simpler Market Orders.