An “Extreme Cold-Start” Resistant method for Online Recommendation

Abstract

Recommender systems are commonly encountered online e.g. Amazon, Netflix and YouTube all use recommender systems to suggest content or products to their users. In order to build and deploy a recommender system, a significant volume of data is usually required. In situations where this data does not exist, or is highly sparse, it can be extremely difficult for recommender systems to gain an understanding of an individual user’s preferences, or the preferences and propensity of groups of users. In addition, in such contexts the use of machine learning and deep learning approaches is not feasible.

This seminar will present ongoing research, in collaboration with Ryanair, which attempts to overcome these limitations by delivering an approach to product recommendation which is resistant to extreme cold-start contexts. The approach blends content-based recommendation, collaborative filtering and contextual recommendation in order to make personalised product recommendations.

Short Bio

Venue

Large Conference Room, O'Reilly Institute