A Fuzzy Logic Inspired Approach for Social Media Sentiment Analysis via Deep Neural Network

In this paper, we present an efficient method of classification of sentiment in social media texts, each consisting of single or multiple sentence(s) that most of the time includes pop culture texts. In our experiment, we present an architecture that derives vector representations (i.e., word2vec) of the phrase level sentences. We use some combination of quantitative and qualitative methods for training a recurrent neural network with empirically cross-validating gold-standard array of lexical features, which are precisely synced with sentiment in microblog-like pieces. We leverage a new technique that expands upon previous works on sentence-level lexical sentiment classification, using recurrent fuzzy neural network and use it jointly with a Recursive Neural Network to further improve the classification. We have tested our algorithm against the other state-of-the-art methods on various platforms for better demonstration of our experiment with satisfactory and competitive results.