JISE


  [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15]


Journal of Information Science and Engineering, Vol. 39 No. 3, pp. 609-621


Enriching Aspect-based Recommendation System using Social Relation-Item Interaction


SEEMA SAFAR1, BABITA ROSLIND JOSE2 AND T. SANTHANAKRISHNAN3
1,2School of Engineering
Cochin University of Science and Technology
Kerala, 682022 India

3Naval Physical Oceanographic Laboratory
Defense Research and Development Organisation
Kerala, 682021 India
E-mail: {seemasafar; babitajose}@cusat.ac.in; tsanthan.npol@gov.in


Amazon, Netflix, and other e-commerce sites allow users to read reviews of products before buying them. Some reviews can be trusted, while others can be quite misleading. The reviews of the people are generally quite diverse. The e-commerce boom has increased the need for Recommendation systems now more than ever. In light of this, the Recommendation system should be improved. It is important to consider the review’s aspects for a clear understanding of the reviews. Most of the baseline models do not give enough importance to the aspect of the review. In this paper, the idea of incorporating social relation-item interaction along with the user-item interaction by understanding the aspect of the user review is incorporated to improve rating prediction. Simple linear transformations Fast Fourier Transform (FFT) are used in the proposed Recommendation system to model heterogeneous semantic relationships in texts. The attention mechanism used in traditional baseline models is not used here for representing feature learning and understanding them. The proposed model is investigated to overcome some of the limitations addressed in attention mechanism-based models. An investigative analysis is done on standard datasets: SemEval 2014 laptop reviews, and restaurant reviews. Analysis of the proposed model is also done on a real-world dataset: Yelp showing an accuracy of 80.01%.


Keywords: recommendation systems, aspect-based sentiment analysis, information retrieval, fast fourier transform

  Retrieve PDF document (JISE_202303_11.pdf)