Product Review Sentiment Analysis Using Weakly-supervised Deep Embedding Using CNN and MLSTM Techniques

Abstract

Decisions making of upcoming buyers are assisted by product reviews. For this, proposed various opinion mining techniques. In this major difficulty lies in judging orientation of review sentence. Problems of sentiment classification can be solved by using a deep learning method.  In Mining of online user generated content, sentiment analysis is a major difficulty. Reviews of customer are concentrated in this work. It is opinionated content’s important form. Every sentence’s semantic orientation is identified by this.

Substantial human efforts are involved in traditional sentiment classification methods. Feature engineering and lexicon construction are the examples of it. Problems of sentiment classification can be solved by using a deep learning method.  Without human efforts, useful representations can be learnt automatically by neural network intrinsically. Availability of large scale training data defines the deep learning method’s success. For review sentiment classification, novel deep learning framework is proposed in this work. Prevalently available ratings are employed as weak supervision signal.

There are two steps in this framework. They are, high level representation learning, adding of classification layer on top of embedding layer. For supervised fine tuning, labelled sentences are used. Through rating information, sentences general sentiment distribution is captured using this high level representation. Efficiency and superiority of proposed method is shown by the experimentation done using a Amazon’s review data.