Efficient Sentiment classification with Minimal parameters using Average Embedding Approach

Abstract

Sentiment analysis is the area of research for analyzing customer opinions on services or products delivered by an entity. With the evaluation of deep learning, the recurrent neural network is picked as the preferred method for most of the sentiment analysis research. The goal of this paper is to build a model that uses minimum parameters without compromising too much on the performance. Three models are built on the publicly available dataset. The performance of these models is then evaluated. It is observed that the model using long-short term memory gives very good performance among all the models but uses too many parameters. The last model uses average of word embeddings which uses half of the parameters used in the previous model and its performance is very much near to the previous one.