An improved personalized item-based collaborative recommendation algorithm

Loading...
Thumbnail Image
Date
2022
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Teknologi Malaysia
Abstract
In the increasingly competitive online market, users are presented with a massive number of items to choose. Recommendation algorithms are used to assist users in making shopping decision conveniently. However, the existing recommendation algorithms are still limited in covering the diversity in the recommendation list. This research aims to investigate the recommendation algorithms and design a personalized recommendation algorithm to cover unpopular items and diversity, which leads to more opportunities to create new offerings. This research is divided into three phases: first, to identify the existing recommendation algorithms; second, to design and implement the proposed algorithm; and third, to generate a candidate list, item-based collaborative filtering is used. Then, k-means cluster based on the genre was proposed to rank the items based on the candidate list and user’s interests to increase diversity and coverage. The effects of different methods impacted the diversity level of the developed algorithm, and the effectiveness and shortcomings of the algorithm are identified. Next, the recommendation list was analyzed and compared with other recommendation algorithms. The results of the proposed algorithm showed an improved diversity by 12% in comparison to personal collaborative filtering (PCF), 14% compared to Div-Clust, and 20% compared to the baseline algorithm. The improvement is due to reranking items based on the user’s diversity level and genre clusters which were introduced in the proposed recommendation algorithm. In addition, the coverage percentage has improved by 35% in comparison to PCF and Div-Clust, and 23% compared to the baseline algorithm, which increases the range of offerings to users. The recommendation algorithm demonstrated that it is able to generate a recommendation list that increases the diversity and coverage of items over other algorithms.
Description
Thesis (Doctor of Philosophy)
Keywords
Algorithms, Arithmetic—Foundations, Programming (Mathematics)
Citation