Chinese stock market hot spot mining model based on financial big data

Abstract

Financial investment is running through the worst times because the general macro analysis of the economic environment is no longer an efficient way to construct an investment portfolio. In the past few years, emerging financial technology has brought a novel solution to financial challenges by combining machine learning and big data. This has brought prospects in investment portfolio construction. The methodology of analyzing the stock data related to big financial data and comparing the multiple types with single types in quantitative trading strategies. The financial news-driven stock market analysis has attracted more and more attention.

Semantic sentiment analysis is the favored stock market hotspot mining approach in the historical process. Its accuracy depends heavily on the quality of the cumbersome sentiment dictionary construction. And the sentimental tendencies unearthed from financial texts cannot quantitatively reveal the stock market hotspot. So, we proposed a stock market hotspot mining model that integrates the Latent Dirichlet Allocation topic model with the volume-price relationship based on the information theory to solve this problem.

To validate the efficiency and practicability of the model, we use big data, including time series stock pricing, listed company annual financial reports, company-related financial news, and user-generated content, which are available. Furthermore, it uses non-structured data from the company's financial statements and related financial information. The experimental index results show that the average maximum increase of the hotspots mined by the model is almost the same as the average of the benchmark index, which is essential for a better understanding of the stock market microstructure and trading behavior for future direction.