Dynamic Label Adjusted and Key Term Based Product Review Analysis Framework for Weakly Labelled Data

Abstract

Sentiment analysis is the rising and the important field carried out on social network. Product reviews are very much important and one of the major factor which is considered by the new buyers. For upcoming buyers, these reviews help in making decisions. The key challenge lies in considering the weakly labelled data. The main objective of this work is to make an effective prediction of the product review sentiment in weakly labelled data. This system consists of two kinds of model to strengthen the review sentiments process with the help of RNN and CNN.  First model (KT_CNN) is constructed with the help Yake based Key Terms learning through Convolutional Network added with Blended Sentiment Polarity. The second model (DLA_RNN) is designed with polarity distance based dynamic label adjusted data with Long Short Term Memory network. The deep features extracted from the both models are concatenated to represent the sentiment of the product review. The proposed (KT_CNN_DLA_RNN) method archives the F1-Score up to 88.71% for the Amazon Data Products dataset