A RNN Based Parallel Deep Learning Framework for Detecting Sentiment Polarity from Twitter Derived Textual Data

Social media platforms have become one of the primary mediums of communication nowadays. Along with communication, they are currently being utilized in a wide range of activities like digital marketing, customer care, e-learning etc. The unceasing use of social media is generating gigantic amount of textual data everyday. It is essential to properly analyze these data with the consideration of underlying human traits sentiments for exploring the full potential of these platforms. However, sentiment analysis from text has been considered as a challenging task because of the rapid use of informal and noisy words. Updated and powerful word embeddings are being invented almost every two years so that the machines could understand the underlying features of linguistics. Each of these embedding techniques excel in different aspects. In this paper, we present a novel RNN based sentiment polarity detection framework which feeds the power of three different word embeddings: Word2Vec, GloVe and SSWE into a single powerful network with three parallel branches. The proposed network effectively utilizes the semantic, syntactic and sentiment polarity wise embeddings in word vectors encoding three major aspect of language from the viewpoint of extracting sentiment information from text. Posts collected from twitter was used to train and validate the proposed network. The results demonstrate that the proposed network containing parallelly configured multiple word embeddings outperforms the single word vectorization techniques. Additionally, it shows comparable or better evaluation scores when compared to several contemporary state-of-the-art models.