Browsing by Author "SIYAM, BILAL YACOUB"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Deep Learning for Aspect-Based Sentiment Analysis of Government Mobile Apps Reviews
    (The British University in Dubai (BUiD), 2018-09) SIYAM, BILAL YACOUB
    The number of smartphone users in the United Arab Emirates (UAE) has increased by 25% in the last four years (Statista.Com 2018c). Government entities in the UAE aim to provide their services through various digital channels, including smartphones among others. This offering will allow customers to use their preferred digital channels and cover the largest customers’ segment. As life is accelerated and people are seeking a quick and efficient way to consume services, government entities become concerned in providing easily accessible services through smart applications. On the other hand, smartphone users are interested in considering others’ opinions and reviews before downloading or using an application. Thus, it is essential for government entities to take into consideration these opinions and comments for the purpose of developing and improving their applications and providing the intended value to their customers. Sentiment Analysis (SA) is the process of analysing textual information and understanding the intent related to emotions, feelings and behaviours. SA is classified into three levels of analysis, namely document-level, sentence-level and aspect-level. This study focuses on aspect-based sentiment analysis (ABSA) of Dubai government mobile apps reviews. ABSA takes into account analysing various features or aspects stated in the review. This study aims at exploring the use of deep learning techniques in improving the performance of ABSA in government app reviews domain. Our approach uses Convolutional Neural Network (CNN) framework which is one of the deep learning approaches. Deep learning is chosen in this study over other approaches since it requires less human intervention and less effort in features engineering. In addition, it simulates how humans think in representing patterns and simplifying them. The approach utilized word embeddings to represent the reviews as vectors and inducted them into the input layer of the deep learning model. The proposed CNN framework consisted of three hidden layers: convolutional, pooling and classification. Further, several techniques and hyper-parameters have been explored as well as their effect on the accuracy of our framework. The performance measures showed that (GloVe) outperformed other word embeddings models. Additionally, utilizing variety of activation functions, namely Sigmoid, Tanh and Softmax enhanced the model performance by 6.5%. The proposed framework achieved high performance results with 89.42% accuracy. This study highlighted several advantages of using deep learning approach. For instance, it does not require much human experience in the domain of the problem compared to the rule-based approach. Further, it needs less effort in identifying features for training.