Stock Comment Text Orientation Analysis Via Attention Based Deep Learning Network

We have already entered the era of big data,there are a lot of data that we can used for analysis on the work,the major is text data.People will express their comments on the blog,post bar or other commentary areas about the product,movies or about stock.These data include the author’s opinions or views,if we can extract the emotion target from them,we can deeply improve the degree of auto level of the network public opinion risk analysis,product improvement,users attraction or the information forecasting.Therefore,the text analysis has important values on both scientific research and practical application.Currently,attention model has been intensive studied in the field of NLP,it can make the neural network more focus on the text which have important influences on the whole articles by weighted the intermediate variable of the encoder network,it also can reduce the program running time and improve the accuracy of the network.Attention model is usually used in text translation,in this paper,we make some improvement on this algorithm,we combine it with deep network,and add CNN layer to extract local context information in order to increase the accuracy of the text classification tasks.We also introduce the TF-IDF weight to speed up the initialization of the attention model.In order to verify the validity of the model on the proprietary field,we design a spider to get the stock comments on the Internet as training sample.