Çok Modlu Veri ile Hisse Senedi Fiyati Tahmini

Abstract

In today's financial markets, the influence of dynamic factors is becoming increasingly evident, and markets are affected by real-time data from various sources. In this study, a multimodal machine learning approach is adopted by integrating traditional technical analysis metrics, tweets, and news articles with historical price data. Market sentiment and investor psychology are measured through sentiment analysis of textual data using both the FinBERT and ChatGPT-4o models, and the obtained outputs are combined with financial metrics to construct an LSTM-based stock price prediction model. To enhance the model's stability, LSTM models derived from different training sessions are merged using an ensemble learning method, and the two approaches are compared. The results demonstrate that the ensemble model outperforms the standard LSTM model, and integrating financial indicators with tweet and news data, as opposed to relying solely on price data, leads to increased overall profit. © 2025 Elsevier B.V., All rights reserved.