Financial Market Events Prediction Based On Deep Learning

Financial market event prediction refers to predicting what the next event might be given a series of historical events.The task can provide advice for investors and reduce their investment risks.However,there is a little research on financial event prediction tasks at present.And they have not fully explored the more complex relationships between events and lacked the introduction of external knowledge.We provide an in-depth analysis of these issues from two perspectives.The specific research content is as follows:Firstly,we construct a financial market event prediction method based on the fusion of event chain and event graph which aim at the problem of insufficiently mining the complex interaction between historical events and candidate events.A single relationship between events,such as sequential relationships in event chains or spatial relationships on event graphs,that previous studies only considered may miss the interdependent relationships.In order to obtain the rich relationships between events as much as possible,we propose a model that integrates sequential and spatial relationships.Above all,we integrate the historical events in the historical event chain to form a corpus,and use the Skip-gram algorithm to obtain the event embedding representation to adapt to the particularity of the financial market event data set.If there is a new event,this event will be represented by a zero vector.Next,the GRU model and the Gated Graph Neural Network(GGNN)model are used to get the hidden state of historical and candidate events,then these two states are spliced so that the event contains both temporal and spatial features.Thirdly,the attention mechanism is introduced to give different weights to historical events.Finally,our experimental results show the effectiveness of the fusion model.Secondly,we propose a financial market event prediction method based on stock price trend to aim at the lack of external knowledge in the past.Previous studies have shown that the introduction of external knowledge can improve the effect of financial market event prediction.To this end,we propose a novel external knowledge: stock price trends.At first,we analyze the stock price and construct eight refinement rules,each of which corresponds to a section.And use eight rules to project the stock price.Then select several mapped stock price fluctuation data to add to the historical events,and use the GRU model to obtain its stock price trend vector representation,as well as concatenate it with the historical event representation.Finally,the GRU model is used to obtain the hidden state of the spliced historical events and the candidate events,and the attention mechanism is used to obtain the different confidence levels of the historical events for the candidate events.Comprehensive experiments on the financial market event prediction data show that the model achieves good evaluation results.