Handling cold-start scenarios in point-of-interest recommendations through crowdsourcing

Point-of-Interest (POI) recommender systems are used to suggest places or venues for a target user based on her preferences. The traditional recommender systems utilise the historical data to learn user preferences and then select places that relate to them. Therefore, knowledge of the features of a POI is important along with the preferences of the target users for a traditional recommender system. This technique faces a serious problem when a new POI emerges in a city. A 'new' POI has no historical data and hence a recommender system fails to learn about its features. This results in absence of the 'new' POIs in the recommended list. Such a scenario is popularly known as the POI cold-start scenario. Online social networks such as Yelp, TripAdvisor, Foursquare, etc. that provide POI recommendations as a service mostly face this particular problem. To address this issue, the proposed work gathers content on the cold-start POIs by crowdsourcing other online social networks and subsequently the dominating features at POIs are identified from the collected review and rating data. These features are utilized to address the cold-start problem. Finally, we develop a POI recommender system that can handle the POI cold-start scenario. We experimented on the real-world data provided by Yelp and the results are found to be significantly better than the state-of-art techniques in handling cold-start scenarios.