A Multimodal Transformer Approach for Market Movement Prediction Using News and Social Media Sentiment

Abstract

Forecasting the financial markets is challenging due to the dynamic and intricate nature of economic systems. News headlines, social media conversations, and price trends from the past all have an impact on public and investor sentiment, which in turn affects the market. Literary sources frequently contain contextual information, in contrast to traditional forecasting algorithms that rely on numerical time-series data. In this study, multimodal transformers are used to forecast market movements by combining financial time-series data with sentiment embeddings from social media and news. Market data is modelled using transformer topologies that account for long-term temporal links, and textual semantics are developed using sentiment-aware embedding methods. The complex dynamics of the market and the public's emotional exchanges can be captured via fusion techniques, which integrate modalities. The proposed approach outperforms more conventional machine learning models and unimodal deep learning in terms of performance prediction when tested on real-world financial datasets. This tactic is reliable for all asset classes and markets. The results demonstrate that platform for intelligent financial decision-support systems can be provided by multimodal sentiment-aware transformers, which in turn enhance market projections.