Improving recommendation diversity and serendipity with an Ontology-based algorithm for cold start environments

Abstract

Every real-life system has several development phases. In the beginning, it has no interactions; wecall this the cold-start problem. After a while, where user interactions occur in the system, the sys-tem gets to a state where one part of the catalogue has some interaction that could be enough touse CF algorithms. However, there’s still a larger amount of items with a small amount of inter-action, so-called cold-item. We call this state a transition state. If most of the item in the systemhas interactions, we call this a regular state. The system needs a slightly different behaviour fromthe recommendation system at each stage. At the cold-start stage, we need exploration of thecatalogue and the involvement of the maximum number of cold-items to use CF as soon as pos-sible. In the transition phase, we need diversification because there are still many items with asmall amount of interaction. In the regular phase, we want to offer new relevant items to usersto surprise them and keep them from getting into the information bubble and thus have a betterexperience using the system. Our approach uses Ontology, which can provide semantic item simi-larity if the text attributes contain the exact words but even if the text attributes look different.This paper presents a Universal Ontological-Based Algorithm (OBACS) that uses diversificationmethods such as serendipity and diversity to improve user experience in all system developmentphases. We verify our approach experimentally on the Movie Database and describe the results.