Improving Aggregate Diversity in Hybrid Recommendation Systems using Temporal Slot-Based Re-Ranking

Abstract

A Recommendation System is used to offer content to users that is both relevant and interesting to them. Businesses use recommendation systems to identify user’s preferences and interests, and promote their products and services to the end-users. However traditional recommendation systems aim to increase the accuracy of recommendations made, without considering the quality and diversity of the recommendations being made. This results in the creation of a set of items that face starvation and are rarely recommended, called the long-tail. This long-tail is an economic liability to the business, so it must be either removed altogether or promoted further. A hybrid list is used, generated by combining the Collaborative Filtering techniques of Item-Based Collaborative Filtering and Matrix Factorization using Singular Value Decomposition. The work done primarily focuses on improving the aggregate diversity of recommendations made, without majorly affecting the accuracy of the system, using a temporal based re-ranking technique.