A collaborative, contextual and content-based recommender
In 2017, Ryanair was launching a new accommodation service – Ryanair Rooms. As this was a new service, Ryanair had limited data on guests using this new service. Ryanair wanted to explore how it could offer a more personalised experience to these customers. The two main approaches to recommender systems are Collaborative Filtering and Content-based approaches, which do not perform well where there is limited data. Following an innovation workshop with ADAPT, the company decided to engage in a collaborative project to explore a novel room recommender system.
Challenges: – Data sparsity problem and lack of personalisation in Collaborative Filtering approaches – Content-based approaches suffer from over-specialisation – Current approaches ignore user’s context – Addition of new user or items is time and resource-consuming
A real-time hybrid recommender that combines different techniques and exploits all the available information about users, such as: User’s preferences to personalise recommendations Group preferences to capture tastes of similar people Data associated with items to apply content-based techniques Contextual information
It overcomes the user and item cold start problems and overcomes the shortcomings of content-based and collaborative filtering approaches.
Inputs: User’s preferences, contextual information.
Technology: Hybrid approach that blends elements of naïve collaborative filtering, content-based recommendation and contextual suggestion.
Outputs: Custom recommendations.
Does not require significant rating data. Generates personalised recommendations. Provides real-time and robust recommendations. Recommendations for Users in cold start contexts: few (or no) data about users.
Check out this video by Prof. Conlan’s on his Personalisation Research
