A sentiment analysis model has developed by utilizing the state-of-art natural language processing models BERT, XLNet, FLAIR, ULMFit and word embeddings. This well-established NLP model on a large dataset has applied onto a smaller dataset, Turkish financial markets news. A prediction model has created by enhancing sentiment with financial metrics about the firm to forecast next day performance of the stocks with an ensemble learning technique (the ensemble model is a proprietary so haven't been commit to here). It has been presented as an industrial engineering graduation thesis and accepted to an international conference with 2 papers.
- M. S. Sivri, A.Ustundag, B. S. Korkmaz, “Combining News Sentiment Labels with Financial Variables to Improve Market Prediction”, INFUS, 2021.
- M. S. Sivri, B. S. Korkmaz, A. Ustundag,“From Statistical to Deep Learning Models: A Comparative Sentiment Analysis Over Commodity News”, INFUS, 2021.