AN IMPROVED LEXICON USING LOGISTIC REGRESSION FOR SENTIMENT ANALYSIS

Abstract

Recently, stock market activities are becoming dependent intensely on social media interactions to deliver significant information for an extensive number of users. This obliges frameworks to scale expeditiously to suit the surge of new as well as existing users going to the recommendations based on information extracted from social media. In this work, we propose an approach for assigning scores to lexicon items using logistic regression based relative scoring to address both the proposal quality and the framework versatility. We propose to assign a rich range of scores to items in the lexicon, as indicated by their web usage history and corresponding effects. We utilise a Deep Learning way to deal with scores of the lexicon to space where the comparability amongst items and their favoured effects is maximised. In this dissertation work, we will talk about a stock market lexicon that endeavours the sentiment regularities caught by a Logistic Regression model in microblog data. Most of the lexicon acquisition frameworks regard words as paired vectors under the exemplary sack of-words model; however, there is not an idea of relative comparability between words while depicting the same sentiment effect. This relativity is considered using the logistic regression model and the accuracy of the results is found to be improved significantly.