Sentiment Analysis Model Based on Multi-Layer Perceptron for Social Networks

Abstract

Social media and other online platforms contain a large amount of data in the form of text, audio, video and image. Sentiment analysis (SA) has become a field of computational studies. In general, SA deals with the mining of information related to sentiments. Therefore, SA is necessary in texts in the form of messages or posts to determine whether a sentiment is negative or positive. SA is a focused on the extraction of emotions and opinions of people about products, events, movies, videos and music from structured, semi-structured or unstructured textual data. SA is an interesting topic in the field of research and technology. It combines natural language processing techniques with data mining approaches for building systems. The main problems that exist in the current techniques are: inability to perform well in different domains, inadequate accuracy and performance in sentiment analysis based on insufficient labelled data, incapability to deal with complex sentences that require more than sentiment words and simple analysing. Therefore, an approach that can classify sentiments into two classes, namely, positive sentiment and negative sentiment is proposed. A multilayer perceptron (MLP) classifier has been used in this document classification system. The aim of the present research is to provide an effective approach to improving the accuracy and speed of SA systems. This objective is achieved via four main steps. The first step is pre-processing aimed at noise removal or data filtering. It also involves prep-processing linguistic data using natural language processing (NLP) techniques. During this process, the input dataset is filtered and processed to provide highly accurate data, reduce the dataset size and shorten the processing time. The second step is applying feature extraction using the term frequency–inverse document frequency (TF-IDF) technique for dimensionality reduction by which an initial set of raw data is reduced to easily manageable groups for processing. The third step is applying feature selection using the chi-square method to reduce the features of documents and thereby shorten the processing time and improve system accuracy. In the final step, the special structures of the multilayer perceptron (MLP) classifier are designed to determine whether a sentiment is positive or negative. The proposed approach is applied to and tested on two datasets, namely, a Twitter dataset and a movie review dataset; the accuracies achieved reach 0.85% and 0.99% respectively