Advancing aspect-based sentiment analysis with a novel architecture combining deep learning models CNN and bi-RNN with the machine learning model SVM

Over the last decades, the aspect-based sentiment analysis (ABSA) task has been given great attention and has been deeply studied by the scientific community. It was first introduced in 2002 to extract the users’ fine-grained sentiments from textual data by focusing on aspect terms. In this paper, we propose a machine learning-based architecture called CBRS (CNN-Bi-RNN-SVM) to enhance the ABSA of smartphone reviews. This architecture combines two deep learning models [convolutional neural network (CNN) and bidirectional recurrent neural network (Bi-RNN)] with the classical machine learning model support vector machine (SVM). The CNN and the Bi-RNN models are used to capture both local features and contextual information. The SVM model is applied to classify the sentiments, expressed towards aspect terms, as positive or negative. To evaluate the performance of the developed architecture, 8,000 French smartphone reviews, extracted from the Amazon website, are annotated to create a dataset including 15,411 positive aspects and 14,627 negative aspects. The obtained findings corroborated the efficiency of the designed architecture by achieving an F-measure value of 94.05%, for the smartphone dataset, and 85.70% for the SemEval-2016 restaurant dataset. A comparative study demonstrates that the overall performance of our proposed architecture outperformed that of the existing ABSA models.