Hybrid Sentiment Polarity Prediction Scheme in Social Networks using Attention Mechanism and Improved CNN

Sentimental Analysis (SA) focuses on determining whether a specific textual term of communication represents a positive or negative opinion with respect to a government, people or business. SA plays a significant part in Natural Language Processing (NLP) which in turn relatively increases volume of unstructured data leading to a challenge in handling extraction of useful information. In case of fine-grained sentimental classification, determining polarity with precision is still a challenging issue due to complex logics, ordering of tests and change in the length sequence. In this paper, Hybrid Attention Mechanism-Improved CNN-based Sentimental Polarity Prediction Technique (HAMICSPPT) is proposed for determining precise sentimental polarity of positive and negative opinions in social networks. This method of SA offers improved tokenization and pre-processing of tokens such that better sentiments with lower and higher polarities can be determined with maximized precision. It also uses the merits of Bag of Words (BoWs) model for identifying and extracting information related to different semantics of each sentiment under use. This model considers a set of collected tweets from the user and topic as input for achieving better sentimental polarity prediction. The experiments conducted with social network datasets confirm better Classification Accuracy, Recall, F-Measure and Precision compared to the baseline approaches.