Research On User Investment Analysis Based On Heterogeneous Data

Despite the continuous development of China's financial market and increasing investor enthusiasm,many individual investors are unable to systematically evaluate and analyze their investments.The rate of return calculation is important for user investment analysis.Traditional rate of return calculation methods have problems in the complex investment scenarios.Traditional user investment analysis methods provide fewer dimensions and the results are not intuitive.In addition,with the rapid rise of Internet information platforms,financial news is rapidly increasing.Financial events in the news can make users understand the event factors in the fluctuations of their investment profits,thereby understanding the internal rules of the securities market and improving their investment capabilities.This paper mainly studies how to use heterogeneous data composed of structured data such as stock market data and stock industry information and unstructured data of financial news to analyze user investment.The research contents include the following aspects:Research on user investment analysis methods using structured data.In order to solve the problem of calculating the rate of return in complex investment scenarios,this paper proposes a time-weighted and cost-weighted rate of return based on the SEA grid,which is compared with the traditional methods to verify its effectiveness.An investment analysis method based on structured data is proposed,and calculation methods for six dimensional indicators such as profitability,capital liquidity,and time-choosing ability are designed and the user investment analysis platform is built.Research on the extraction method of financial events based on news data.In order to use the event information in the financial news to analyze the profit fluctuations of users' investment,this paper studies event extraction methods.Considering that most of the existing methods are based on corpus with trigger word annotation,and labeling trigger words takes more effort,this paper studies a joint extraction method of financial event types and subjects based on a multi-layer pointer decoding network without trigger word annotation.The label attention enhancement network is designed to integrate event type information into the model to achieve better results.Considering that the traditional pointer decoding network only uses the start and end pointer prediction results to decode multiple event subjects,it is prone to match misalignment due to a pointer prediction error.This paper adds matching prediction of start and end pointers to improve the accuracy of decoding event subjects.The experimental results show that the performance of this method exceeds other methods,and the weighted F1 value reaches 92.11%.This method is used to extract 21,615 financial events from the news text,laying a solid foundation for subsequent analysis.Research on the event analysis method of user investment combining structured data and financial event data.In order to quantify the impact of financial events on stock prices,this paper introduces the event study method to analyze the impact of events,and design weight factors to quantify the degree of impact.Design a method of event analysis of user investment,and propose a reliability index to evaluate the analysis result.Experiments show that by combining event data,it can effectively provide event factors in the fluctuation of user investment profit,and help users better analyze their investment.