Open AccessApplying uncertainty reduction strategy for improving performance of questionnaire technique of solving cold user problem
Author(s) -
Muhammad Abubakar,
K. Umar
Publication year - 2021
Publication title -
bayero journal of pure and applied sciences
Language(s) - English
Resource type - Journals
ISSN - 2006-6996
DOI - 10.4314/bajopas.v13i2.6
Subject(s) - movielens , collaborative filtering , computer science , recommender system , product (mathematics) , cold start (automotive) , reduction (mathematics) , uncertainty reduction theory , data mining , machine learning , information retrieval , mathematics , engineering , geometry , communication , sociology , aerospace engineering
Product recommendation systems are information filtering systems that uses ratings and predictions to make new product suggestions. There are many product recommendation system techniques in existence, these include collaborative filtering, content based filtering, knowledge based filtering, utility based filtering and demographic based filtering. Collaborative filtering techniques is known to be the most popular product recommendation system technique. It utilizes user’s previous product ratings to make new product suggestions. However collaborative filtering have some weaknesses, which include cold start, grey sheep issue, synonyms issue. However the major weakness of collaborative filtering approaches is cold user problem. Cold user problem is the failure of product recommendation systems to make product suggestions for new users. Literature investigation had shown that cold user problem could be effectively addressed using active learning technique of administering personalized questionnaire. Unfortunately, the result of personalized questionnaire technique could contain some user preference uncertainties where the product database is too large (as in Amazon). This research work addresses the weakness of personalized questionnaire technique by applying uncertainty reduction strategy to improve the result obtained from administering personalized questionnaire. In our experimental design we perform four different experiments; Personalized questionnaire approach of solving user based coldstart was implemented using Movielens dataset of 1M size, Personalized questionnaire approach of solving user based cold start was implemented using Movielens dataset of 10M size, Personalized questionnaire with uncertainty reduction was implemented using Movielens dataset of 1M size, and also Personalized questionnaire with uncertainty reduction was implemented using Movielens dataset of 10M size. The experimental result shows RMSE, Precision and Recall improvement of 0.21, 0.17 and 0.18 respectively in 1M dataset and 0.17, 0.14 and 0.20 in 10M dataset respectively over personalized questionnaire.