Research On Modeling And Evolution Of Financial Topics Based On LDA Model

Compared with traditional financial statistics,the information in the financial text can effectively reflect the economic ecology at the micro level,covering all economic life and related fields,and at the same time has a stronger timeliness.Current financial text mining is widely used in various economic fields,such as stock price prediction,economic uncertainty measurement,and economic cycle prediction.Accurate modeling of financial topics and ensuring the timeliness of the model are effective supplements to financial statistic,which is essential and beneficial to promote forecasting market fluctuations and trends in the economic field,suppress inflation,and improve investment strategies.However,the traditional topic modeling technology has problems such as non-dynamic,sparseness,noise,high computational complexity,and single dimension of topic evolution analysis.Therefore,it is difficult to accurately identify the topic and effectively analyze the dynamic evolution of financial topics using traditional methods.The LDA model can discover topic categories in large-scale text collections,improve calculation accuracy and reduce complexity.In addition,time-related topic models can track topic changes in the text stream and effectively identify the dynamic laws of financial topic models.However,the current research on the topic model still has the following deficiencies: the segmentation of time leads to the discontinuity of topics between different time slices,the study of corpus only focuses on sparseness and ignores noise,and the failure to systematically study the evolution of topics,which results in poor text topics modeling and evolution analysis.In view of the above problems,based on the in-depth study of the LDA model,this thesis considers the time factor,and based on the sliding-window technique,introduces the genetic factor of financial topic and the common financial topic.Therefore,an SGC-LDA financial topic model and a multi-dimensional financial topic evolution analysis model are proposed,which are used to identify the topics and analyze the topics evolution law of financial texts,respectively.The main research work of this thesis includes:(1)This thesis first designed a financial topics modeling framework,then based on the LDA model,combined the two methods of isochronous segmentation and equal number of document segmentation to segment the user corpus text,and then introduced the genetic topic of financial topics to maintaining the continuity of topics between different time slices,and defining the common financial topics for capturing general semantics and noise interference words,so as to build the SGC-LDA model,and use this model to model financial topics.(2)Based on the analysis of the evolutionary relationship of financial topics,this thesis combines the three dimensions of topic intensity,topic state,and topic path to construct a multidimensional evolution model of financial topic,and designs a topic evolution analysis process,in order to carry out evolutionary analysis of financial topics.(3)This thesis applies the SGC-LDA model to financial topic modeling,and uses financial news text corpus for empirical analysis to verify the superiority of the SGC- LDA model.And then this thesis combines the qualitative and quantitative methods to confirm the effectiveness of the evolution model and analyze the characteristics of financial topics evolution.