A new graphic kernel method of stock price trend prediction based on financial news semantic and structural similarity

Lots of researches try to predict the stock price movement using financial news based on machine learning represented by SVM (Support Vector Machine). But almost all of them focus on the news contents while very few consider the information hiding in the relationship between different news. In this paper, we proposed a new kernel based on SVM concerning not only the contents themselves but also the information structures among them. As both the news contents and the information structures are imported into our kernel, this kernel is named as semantic and structural kernel, referred to S&S kernel. Medical industry financial news is used to illustrate the efficiency of our kernel. By comparing the predicting accuracy of S&S kernel with other kernels, such as linear kernel, we find our method outperforms the others by at least 5% on accuracy, which is a quite meaningful promotion. The result also confirms the information structure contained in daily financial news can offer extra information helping to predict the trend of stock price. (C) 2018 Elsevier Ltd. All rights reserved.