Research On Chinese Concept Stocks Prediction Driven By News Polarity

Stock market forecasting has always been a hot topic in data research,but it is affected by many factors and its prediction is more difficult.News is an important factor affecting stock prices.Investors often rely on news for stock trading and decision making,so the analysis of news can provide investors with effective information.The use of news as non-structural data in stock forecasting is difficult,and with the development of machine learning technology and natural language analysis technology,the solution of this problem becomes possible.At present,the significant differences in capital market policies at home and abroad have led to more and more domestic companies listing in foreign countries,while little research has been done on the impact of Chinese news on China stock market forecasts.Behavioral finance theory finds that investors' investment behavior and trading decisions are influenced by news and other texts.Therefore,the polarity analysis of news texts helps investors to judge whether the content of news expression is positive or negative in real time,and makes corresponding decisions in a very short period of time,quickly grasping the opportunities of income and avoiding the risk of loss.Stock price trend forecasting belongs to the interdisciplinary subject of computer science and finance.In the case of machine learning and deep learning prediction ability,this paper uses financial news to predict stock price trends by referring to different algorithm ideas in two fields.The main work is as follows:(1)On the basis of the algorithm theory of support vector machine,the self-defined automatic machine labeling model is realized,the key phrase weight is added,and a new cyclic evaluation support vector machine model is proposed.The random seed is used to randomly assign the corpus to generate different training.The collection and verification sets are added to the loop evaluation mechanism for training.After reaching the set prediction accuracy rate,the loop is released to obtain the final training result,and testing through the out of sample collection to prove the rationality and effectiveness,so as to improve the research on the influence and prediction of news polarity drive on stock price in Chinese stock concept.At the same time,compared with the convolutional neural network model and the Naive Bayesian model,the prediction performance,prediction trend and simulation transaction are compared.It is demonstrated that the proposed cyclic evaluation support vector machine model has excellent prediction effect and can help investors identify the polarity of the news and get a higher return on investment.(2)The convolutional neural network and long short-term memory networks can extract the network characteristics of text features,and take into account the superiority of convolutional neural networks in feature extraction and the natural advantages of long short-term memory networks in processing time information.The dilated convolution can expand the receptive field during the convolution operation by setting the dilation rate,thus avoiding the phenomenon of information loss caused by the pooling operation of normal convolution.The LSTM-Di CNN model is proposed,which is compared with CNN,LSTM,CNN-LSTM,LSTM-CNN,Di CNN and Di CNN-LSTM models.The input layers of the seven models are connected behind the word embedding layer,and the input data is the financial news text data of the unified corpus.The embedded layer uses Glo Ve processing to generate word vectors,which prevents over-fitting by indirectly introducing external training data,and reduces the number of training parameters to improve training efficiency.Using a custom automatic machine labeling operation and the idea of cross-validation to randomly assign the corpus collection,the seven models compare the predicted performance,the simulated transaction yield,and the simulated transactions cumulative multi-empty portfolio yield,which proves that the LSTM-Di CNN model has the highest accuracy and highest return.In summary,this article studies the impact of news polarity on stock price volatility trends,conducts fundamental and technical fusion analysis through machine learning and deep learning algorithms.The proposed algorithms are applied to the actual trading strategy,and all have achieved good experimental results.