Sentiment Analysis for Stock Market Prediction Using Recursive Deep Neural Networks.

Sentiment analysis (SA) is used to identify the person's opinion from unstructured information. It is broadly applied to predict stock market movement direction to recognize the public opinion towards a company or products. The conventional techniques designed for SA do not provide higher accuracy, which impacts the reliability of stock market prediction. In order to improve the prediction performance, a Gensim Lovins Truncative Morisita–Horn's Broken-stick Regression-based Recursive deep neural network (GLTMBR-RDNN) is introduced for predicting the future outcomes in the stock market with a lesser error rate and minimal time. The customer reviews are collected from a large database. The GLTMBR-RDNN includes different layers for learning the input reviews. In the GLTMBR-RDNN technique, the first preprocessing of the text is carried out in the first hidden layer by removing stop words, stem words, truncation and so on. First, the Gensim tokenizer is applied in the preprocessing step to partition the text into a number of words. The proposed GLTMBR-RDNN technique uses a Sklearned model for stop word removal. Followed by Lovins truncative stemming process is carried out in the preprocessing step. Finally, the Normalization process is done for transforming the words into a standard form. After preprocessing, Morisita–Horn's Broken-stick Regression process is performed in the second hidden layer for predicting the future stock market value based on the classification of customer reviews by setting the breakpoint to the similarity score between the reviews. In this way, future stock market values are efficiently identified with enhanced classification accuracy in the output layer. The result of the suggested GLTMBR-RDNN technique is analyzed using metrics such as accuracy, F-measure, precision, recall and prediction time based on various input reviews. The results prove that the introduced GLTMBR-RDNN technique improves the performance of accuracy with less prediction time when compared to existing methods. [ABSTRACT FROM AUTHOR]