DAILY CONSUMER PURCHASE SUGGESTED LIST BY RECOMMENDED SYSTEM /
Rawahi, Sanaa Saeed Musallam Al
DAILY CONSUMER PURCHASE SUGGESTED LIST BY RECOMMENDED SYSTEM / SANAA SAEED MUSALLAM AL RAWAHI; SUPERVISOR: PROF. DR. ERBUĞ ÇELEBİ - 78 sheets; 31 cm. Includes CD
Includes bibliography (sheets 75-78)
Thesis (MSc) Cyprus International University. Institute of Graduate Studies and Research Information Technologies Department
Includes bibliography (sheets 75-78)
ABSTRACT
The aim of this study is to use algorithms to build a recommender
system that can generate a suggestion for consumer shopping list.
We analyzed the effectiveness of a recommender system using three
different methods collaborative filtering (CF), Content-based
filtering (C-B) and Hybrid). We investigate and test each approach
of the recommended system. Next we have evaluated the accuracy
of the system using Recall@5 scored 79% for popularity and 67%
for CF. Recall@10 scored for popularity and CF 91% and 79%
respectively, Precision@5 scored 77% and 75% for Hybrid and CF
approaches Precision@10 scored 88% for Hybrid and 82% for CF.
The harmonic evaluation for both Recall and Precision F1@10
scored 88% and 80% for CF and Hybrid approach. Finally, we
observe that the content-based scored the lowest which is we
conclude that it is superior to CF for typical consumer purchases.
We observed that hybrid approach provides the best results for
making recommendations to brand-new users with no prior
information.
Keywords: Collaborative Filtering, Content-Based Filtering, Hybrid
Filtering, Precision, Recall, Recommendation System
Recommender systems (Information filtering) --Dissertations, Academic
DAILY CONSUMER PURCHASE SUGGESTED LIST BY RECOMMENDED SYSTEM / SANAA SAEED MUSALLAM AL RAWAHI; SUPERVISOR: PROF. DR. ERBUĞ ÇELEBİ - 78 sheets; 31 cm. Includes CD
Includes bibliography (sheets 75-78)
Thesis (MSc) Cyprus International University. Institute of Graduate Studies and Research Information Technologies Department
Includes bibliography (sheets 75-78)
ABSTRACT
The aim of this study is to use algorithms to build a recommender
system that can generate a suggestion for consumer shopping list.
We analyzed the effectiveness of a recommender system using three
different methods collaborative filtering (CF), Content-based
filtering (C-B) and Hybrid). We investigate and test each approach
of the recommended system. Next we have evaluated the accuracy
of the system using Recall@5 scored 79% for popularity and 67%
for CF. Recall@10 scored for popularity and CF 91% and 79%
respectively, Precision@5 scored 77% and 75% for Hybrid and CF
approaches Precision@10 scored 88% for Hybrid and 82% for CF.
The harmonic evaluation for both Recall and Precision F1@10
scored 88% and 80% for CF and Hybrid approach. Finally, we
observe that the content-based scored the lowest which is we
conclude that it is superior to CF for typical consumer purchases.
We observed that hybrid approach provides the best results for
making recommendations to brand-new users with no prior
information.
Keywords: Collaborative Filtering, Content-Based Filtering, Hybrid
Filtering, Precision, Recall, Recommendation System
Recommender systems (Information filtering) --Dissertations, Academic