Research on Stock Market Prediction via Coupled Matrix and Tensor Factorization

• 1、Key Laboratory of Trustworthy Distributed Computing and Service(BUPT), Ministry of Education, Beijing, 100876

Abstract：With the improvement of people\'s living standards and perception, investment has been widely paid attention. Meanwhile, with the advent of the Internet ero, the way to obtain information gets diversified. Financial institutions have not only used traditional quantitative data to carry out investment analysis, but began to use big data analysis, artificial intelligence,etc. to make investment predictions.Therefore, research on the stock market has become a hot topic in recent years based on social network and multi-source data fusion. For the purpose of investors\' better decision on stock trading,the paper utilizes the tensor to integrate the heterogeneous data,including financial Web news, public affection from social network and some quantitative data on stocks and proposes a coupled matrix and tensor factorization model to predict the stock price volitation.The stock quantitative feature matrix and the stock correlation matrix as auxiliary matrices are constructed to solve the problem of sparsity. And the intrinsic relationship among different information dimensions can be captured via the tensor.Eventually, the results of experiments on the dataset CSI100 stocks in the year 2015 and 2016 are compared with other approaches to forecasting stocks in order to demonstrate the effectiveness of the proposed model.

Keywords： social network stock correlation tensor decomposition；stock forecasting