A Prediction Of Stock Trading Signal Based On Multi-model Fusion And Emotion Of Financial Text

The prediction of price trend in stock market is a challenging task due to its high nonlinearity and time-varying characteristics.It also provides a vast stage for machine learning,which tests and promotes the progress of machine learning technology.Meanwhile,with the further development of reform and opening-up,the security market plays an increasingly important role in the society and economic development of China,and the ratio of individual investors are chronically high,thus the prediction of security market becomes a major issue to maintain the steady and healthy development of the market.The prediction of stock turning points is one of most popular issues in stock prediction field.However,there are many insufficiencies that are needed to be improved.In view of the shortcomings of existing methods,we propose MMF-PLR-RF and CA-RF trading signal prediction models which are respectively based on technical analysis theories and financial text analysis,and verify the validity and superiority of the models by experiments.In the first part,in view of the deficiencies of IPLR-WSVM framework,we improve the turning point generation method,classification algorithm and trading signal decision method of IPLR-WSVM,and propose PLR-RF model,which can obtain more valuable information and filter false signals.Then PLR-RF models are trained by the core features extracted from three technical analysis theories respectively.Finally,we obtain the stock trading signal prediction model MMF-PLR-RF based on multi-model fusion.In the second part,we train word embeddings with a large amount of corpus related to stock investment field which are collected from the Internet,and obtain the domain-related sentiment lexicon using the dictionary auto-expansion technology.Based on aforementioned lexicon,we propose the investor sentiment index and the announcement comment emotion model CA-RF,which can quantify the comment text and provide investors with comprehensive guidance.The MMF-PLR-RF model achieves more stable investment returns than PLR-RF models on a set of randomly selected stocks for two years.Compared with IPLR-WSVM model,MMF-PLR-RF has significant improvement in the evaluation indeces.The case on the three stocks shows that there is a strong correlation between investor sentiment index and volume index,which has important guiding significance in actual investment.The experiment also shows that the CA-RF model can effectively identify announcement information and explain the turning point that can not be predicted by technical analysis,and improve the returns of model.