Predictive Accuracy of Recommendation Algorithms

Prachita Mane, Purdue University

Abstract

Recommendation systems (also known as recommender systems) have become ubiquitous due to the growing popularity of the Internet and the overload of available information. Recommendations help users make informed choices by listing similar and relevant items that might be interesting for the user. Research on recommendation systems as an independent area has been going on since mid-90s and there is no sign of them becoming out-of-date because of the increasing use of the internet for online shopping, music streaming and other applications. This has made recommendation systems an important area of research. With the rise of open source technologies, developers are constantly looking for available libraries that can be embedded into projects to build a suitable recommendation system. In this thesis, a literature review was conducted on the various recommendation approaches, the available open source libraries and accuracy metrics. An open source library “SurPRISE” was chosen to conduct accuracy tests on various algorithms available in this library. Hybrid algorithms are also constructed and tested for accuracy. Results of the measured accuracy of these algorithms on the chosen datasets were noted. Finally, a linear relationship between the dataset features and the prediction accuracies is suggested in this study.

Degree

M.S.

Advisors

Springer, Purdue University.

Subject Area

Information Technology

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS