Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science


Christie Ezeife




Recommendation systems not only aim to recommend products that suit the taste of consumers but also generate higher revenue and increase customer loyalty for e-commerce companies (such as Amazon, Netflix). Recommendation systems can be improved if user purchase behaviour are used to improve the user-item matrix input to Collaborative Filtering (CF). This matrix is mostly sparse as in real-life, a customer would have bought only very few products from the hundreds of thousands of products in the e-commerce shelf. Thus, existing systems like Kim11Rec, HPCRec18 and HSPRec19 systems use the customer behavior information to improve the accuracy of recommendations. Kim11Rec system used behavior and navigations patterns which were not used earlier. HPCRec18 system used purchase frequency and consequential bond between click and purchased data to improve the user-item frequency matrix. The HSPRec19 system converts historic click and purchase data to sequential data and enhances the user-item frequency matrix with the sequential pattern rules mined from the sequential data for input to the CF. HSPRec19 system generates recommendations based on frequent sequential purchase patterns and does not capture whether the recommended items are also of high utility to the seller (e.g., are more profitable?).The thesis proposes a system called High Utility Sequential Pattern Recommendation System (HUSRec System), which is an extension to the HSPRec19 system that replaces frequent sequential patterns with use of high utility sequential patterns. The proposed HUSRec generates a high utility sequential database from ACM RecSys Challenge dataset using the HUSDBG (High Utility Sequential Database Generator) and HUSPM (High Utility Sequential Pattern Miner) mines the high utility sequential pattern rules which can yield high sales profits for the seller based on quantity and price of items on daily basis, as they have at least the minimum sequence utility. This improves the accuracy of the recommendations. The proposed HUSRec mines clicks sequential data using PrefixSpan algorithm to give frequent sequential rules to suggest items where no purchase has happened, decreasing the sparsity of user-item matrix, improving the user-item matrix for input to the collaborative filtering. Experimental results with mean absolute error, precision and graphs show that the proposed HUSRec system provides more accurate recommendations and higher revenue than the tested existing systems. Keywords: Data mining, Sequential pattern mining, Collaborative filtering, High utility pattern mining, E-commerce recommendation systems.