Analysis on Hidden Layer in Deep Artificial Neural Network for Classification of Sentiments

Data in online business contains opinion/comments from the user regarding the purchase that can be used for the benefit of businesses in various aspects. These comments are referred as sentiments on the particular purchase that is required to help taking decisions for further improvement in business. This kind of data produce massive stream, and it is almost impossible of manual tracking and extracting useful information. Hence, every online business has the process of extracting sentiments or opinions from reviews given by users on a particular subject, area or product with Natural Language Processing (NLP) that categorize the opinions of people about the products or the reviews. This research work focuses on hotel industry in which users post comments and views in online mode regarding the hotels they stayed, and NLP is used to label the sentiments on the posts already. Hence, to categorize the database on various sentiments this work implements deep learning architecture with multi feedforward artificial neural network. This neural network has three layers namely input layer, more than one hidden layer and output layer. Among them, fixing number of hidden layers, neurons in hidden layer and dropout ratio has a significant role in obtaining accuracy of the model. This research analyzes the fixation of hidden layers, neurons and dropout ratio by comparing variant sizes and ratios with ReLU activation function. The work is done and visualized in Rapidminer tool.