Recommendation system using Deep Learning-based sentiment analysis
Ikram Karabila, Nossayba Darraz, Anas El-Ansari, Nabil Alami, Mohamed Lazaar, Mostafa El Mallahi
Abstract
With the exponential growth of online data, there is a growing demand for advanced recommendation systems (RS) due to their crucial role in providing personalized item recommendations to users. This work aims to overcome the limitations of traditional RSs by introducing an innovative approach that seamlessly integrates sentiment analysis (SA) into recommender system techniques. The primary goal is to enhance the performance of SA and improve the reliability and accuracy of user recommendations. The suggested framework comprises of three essential steps: SA, where the Bi-GRU technique is utilized to assess the emotional context within user reviews; the generation of personalized recommendations; and the seamless integration of sentiment analysis into the recommendation system. A comprehensive evaluation conducted on a Musical Instruments dataset demonstrates significant improvements in evaluation metrics. The Bi-GRU technique achieves an impressive accuracy rate of 89%. User-based recommendations using SA yield an RMSE of 1.83 and an MAE of 1.71. Additionally, item-based recommendations result in an RMSE of 2.18 and an MAE of 2.00. This pioneering approach represents a substantial contribution, promising more reliable and precise user recommendations, ultimately enhancing overall user satisfaction and the user experience.