Sentiment Analysis of User-Generated Reviews of Women Safety Mobile Applications

Abstract

Google Play Store is basically an app store from where we get various kinds of applications for our android certified devices which makes our life a lot easier and faster through the diverse functionalities the apps contain. Numerous users are using applications as per their needs and putting their experience, thoughts of using that application via reviews in form of ratings and texts. As the safety of women is threatened, whether or not applications like women’s safety apps are appreciated, can be detected through text reviews and ratings by the users. In this study, we try to analyze the polarity (positive, negative, neutral) of the sentences or text reviews that are given by the users of the women’s safety app through the google play store. To detect the emotions of the users through the given text reviews and star ratings, different machine learning (ML) and deep learning methods using natural language processing (NLP) are conducted to analyze the sentiments of the review given by the users. For this study, we have collected data from the app reviews and star ratings provided by the users of the women’s safety related applications whose main purpose is to provide necessary functionality that can keep women safe in any dangerous and unwanted situation. The purpose of this paper is to mine the opinion of the users and get their viewpoint about those apps of specific purpose whether it is positive, negative, or neutral. As the current user’s ratings, reviews, or as a whole their viewpoint helps the new user understand the performance of the applications and insights in advance, so the mining of their opinion is helpful for both parties - developers and general users. To detect the level of the sentiment, we have applied several supervised machine learning algorithms namely Multinomial Naive Bayes (MNB), Logistic Regression (LR), Support Vector Machine (SVM), and k-nearest neighbor (K-NN) as well as unsupervised deep learning algorithm Bidirectional Encoder Representations from Transformers (BERT). Among these algorithms, the BERT has outperformed all other algorithms in terms of accuracy (86.06%).