NOVEL RECOMMENDATION SYSTEM BASED ON USER PREFERENCES USING SENTIMENT ANALYSIS

• Published in: JOURNAL OF ENGINEERING SCIENCE AND TECHNOLOGY
• Main Authors: Sopiah, Nyimas; Kurniawan, Tri Basuki; Dewi, Deshinta Arrova; Zakaria, Mohd Zaki; Basri, Muhammad Tashdiq
• Format: Article
• Language: English
• Published: TAYLORS UNIV SDN BHD 2023
• Subjects: Engineering
• Online Access: https://www-webofscience-com.uitm.idm.oclc.org/wos/woscc/full-record/WOS:001148675500001

When searching online for something to buy, many consumers will look at an online review or rating for a product. 'Unknown' brands of action cameras, watches, and headphones are receiving thousands of reviews, as they are unverified. There is no evidence that the reviewer bought or used the product. The problem is that some people deliberately want to take advantage of others by making a fake review on the product just to make the product appear to be in the market. This problem also happens to users who want to buy a novel from an online website. Reading each review could take a lot of time, and determining whether the reviews are positive or negative would be difficult because some reviews might be fake. In this research, a prototype of the novel recommendation system based on a review of websites is developed. This research's methodology includes an initial investigation, knowledge acquisition, data collecting, analysis, and categorization. The study focuses on how evaluations are differentiated between favourable and unfavourable texts. Based on user choices, the categorization outcome suggests a good novel. The Decision Tree, Backpropagation Neural Network, Naive Bayes, Support Vector Machine, and Recurrent Neural Network classifiers were all used in this research. For this research, the Recurrent Neural Network obtained the best result for classifying the sentiment, with an average accuracy of 87.85% and an average error rate of 9.1%. The classifier was trained with a learning rate of 0.001 and 100 epochs. The result classification can be increased by training more datasets or tuning the hyperparameter value. This statement shows that the future of this research does not end here but can improve from time to time.