Predicting sentiment analysis for Web users with a deep learning approach

In the last decade, Sentiment analysis has been an interesting research topic in natural language processing (NLP) and data mining fields. We have noticed that deep neural network (DNN) models are always applied in sentiment analysis researches to obtain good results. Many neural architectures are applied for sentiment analysis. Among these architectures, Long Short- Term Memory (LSTM) models are the most used ones. In this contribution, we collected our dataset from our developed web application. The obtained dataset contains different types of features; Quality of Service (QoS) metrics, Web Quality of Experience (QoE) metrics, user engagement metrics, comments to videos that are included in the application, and the Mean Opinion Score (MOS) expressed by users. After that, we trained our dataset on our proposed model LSTM-CNN-RNN to predict the MOS. As a conclusion, we obtained a good accuracy and a low loss rate.

I. Introduction

The main aim of sentiment analysis is to analyze and extract information from the subjective users expressions written on the Internet. Recently, sentiment analysis is getting great attention from both researchers and industrials in data mining and Natural Language Processing (NLP) fields. This interest results from the propagation of different types of applications together with the fast growing of web services [1]. Thus, recently, we notice the appearance of a variety of developed methods and tools able to specify the polarity of a document. The phenomenon of polarity detection is considered as a binary classification task. It is preliminary for sentiment analysis applications [2]. Most of previous sentiment analysis methods resulted in shallow models on efficient characteristics carefully designed to achieve satisfactory polarity classification results [3].