Text

theses

Recommender systems are by far one of the most successful applications of big data and machine learning. The goal of recommender systems is to find what is likely to be of interest, thus enabling personalization and tailored services. Among all the recommendation algorithms, collaborative filtering is the most common technique.
In this thesis, we present our research results in the field of ranking-based collaborative filtering. We view the task of recommendation as providing a personalized ranked list of items for each user and thus formulate it as a ranking problem. In order to generate accurate recommended lists, we look into the technique of combining learning-to-rank with conventional collaborative filtering methods for solving the recommendation task and comprehensively discuss the challenges and advantages of this approach. Particularly, we propose an improved pairwise ranking model and a multi-objective ranking framework to address the issues which occur during the combination of learning-to-rank and collaborative filtering. In addition, we propose a new ordinal approach to modeling the ordinal nature of user preference scores, which demonstrates distinct superiority compared to the numerical and (nominal) categorical views of user ratings.

ETD_9228

Ph.D.

Includes bibliographical references

by Jun Hu

doi:10.7282/t3-27m3-e107

ETD doctoral

The author owns the copyright to this work.