Research Area:  Machine Learning
With the developments of e-commerce websites, user textual review has become an important source of information for improving the performance of recommendation systems, as they contain fine-grained user-s opinions that generally reflect their preference towards products. However, most of the classical recommender systems (RSs) often ignore such user opinions and therefore fail to precisely capture user-s specific sentiments on products. Although a few of the approaches have attempted to utilize fine-grained user-s opinions for enhancing the accuracy of recommendation systems to some extent, most of these methods basically rely on handcrafted and rule-based approaches that are generally known to be time-consuming and labour-intensive. As such, their application is limited in practice. Thus, to overcome the above problems, this paper proposes a recommendation system that utilizes aspect-based opinion mining (ABOM) based on the deep learning technique to improve the accuracy of the recommendation process. The proposed model consists of two parts: ABOM and rating prediction. In the first part, we use a multichannel deep convolutional neural network (MCNN) to better extract aspects and generate aspect-specific ratings by computing user-s sentiment polarities on various aspects. In the second part, we integrate the aspect-specific ratings into a tensor factorization (TF) machine for the overall rating prediction. Experimental results using various datasets show that our proposed model achieves significant improvements compared with the baseline methods.
Author(s) Name:  Aminu Da, Naomie Salim, Idris Rabiu, Akram Osman
Journal name:  Information Sciences
Publisher name:  ELSEVIER
Volume Information:  Volume 512, February 2020, Pages 1279-1292
Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S0020025519310060