Incorporating News feed to predict stock market using Deep Generative Recurrent and Anomaly ...
Predicting stock market prices has been one of the most challenging tasks in the modern business world and predictive models. In addition, using a news feed as a proxy indicator for the stock market prices change is adding one more complexity layer on any predictive model. The most common way to look at the stock market prediction problem is to consider it as a time-series problem, which is true to a certain level. Nonetheless, the confounding factors for stock changing are more variants to be accounted for. In machine learning realm, the go-to model is usually Recurrent Neural Network (RNN) considering that the stock market operates with trends and seasonality.
In this thesis work, a generative recurrent and anomaly sensitive model is introduced to incorporate not only time-series input from the stock market but also a news feed. This model is capable of memorizing past events, in addition, to extrapolate and reason for novel events. The traditional subcomponents of the proposed model are RNN, more specific, Long-Shot Time Model LSTM, Differentiable Neural Computer (DNC), and Isolation Forest.