The improvement of Item-based collaborative filtering algorithm in recommendation system using similarity index

• 第一著者: Yao, Ma
• フォーマット: 学位論文
• 言語: 英語; 英語
• 出版事項: 2022
• 主題: T58.6-58.62 Management information systems; HF5001-6182 Business; T Technology (General)
• オンライン･アクセス: https://etd.uum.edu.my/9764/1/permission%20to%20deposit-grant%20the%20permission-823872.pdf; https://etd.uum.edu.my/9764/2/s823872_01.pdf; https://etd.uum.edu.my/9764/

The extensive and increase use of high-tech product in business has generated a huge amount of business information to be processed in many fields. Thus, a recommendation system is introduced as an effective strategy to manage the business information overload problem. The system aims to filters enormous information and proposes appropriate suggestions to users. A collaborative filtering algorithm is one of the algorithms applied in the recommendation system. However, the collaborative filtering algorithm faces cold-start problem, where new items in the shopping list are not identified and recognized by the system. Hence, this study proposes an improved collaborative filtering algorithm which aims to alleviate the cold-start problem by combining the item rating and item attributes in similarity index. The performance of enhanced algorithm was compared to existing collaborative filtering algorithms in term of precision rate, recall rate and F1 score using Movielens dataset. The algorithm’s efficiency, objectiveness, and accurateness towards its performances were measured. Finally, the experimental results showed that the proposed algorithm get 15 percent precision rate, 6 percent recall rate and 9 percent F1 score. Thus, it proved to be more effective in deal with cold-start problems by using new similarity index, and also can make recommendations on new items in different fields with satisfactory accuracy for better recommendation result. Theoretically, this study contributes to improve the collaborative filtering algorithm in recommendation system for overcome the cold-start problem by analyzing more item attributes to extract more information to the algorithm. Besides, the proposed algorithms can be applied in many fields for cold-items recommendation and to enhance the quality of the recommendation system.