PROJECT TITLE :

Financial Latent Dirichlet Allocation (FinLDA) Feature Extraction in Text and Data Mining for Financial Time Series Prediction

ABSTRACT:

Many financial time series predictions based on fundamental analysis have relied heavily on news. However, sifting through a large volume of news and data released on the Internet in order to forecast a market can be taxing. This work introduces Financial LDA, a topic model based on latent Dirichlet allocation (LDA) for discovering features from a mixture of text, particularly news articles and financial time series (FinLDA). FinLDA characteristics are used as supplementary input features for any Machine Learning system to improve financial time series prediction. We give Gibbs sampling posterior distributions for two FinLDA versions and suggest a framework for using the FinLDA in text and data mining for financial time series prediction. The experimental results reveal that the FinLDA features empirically contribute value to the prediction and produce better outcomes than the comparison features from the standard LDA, such as topic distributions.