Item recommendation in aggregated marketplace

In this work we introduce the task of item recommendation in an aggregated marketplace. An aggregated market place is a single place where items from multiple e-commerce sites can be viewed and compared. Our main objective while recommending items is to provide users with alternatives across different e-commerce sites. To accomplish this, we embed items in a vector space and apply retrofitting, a method used in language processing to inject relationships into word embeddings. Due to recent attention towards item recommendation using only click data without any user feedback and our particular constraints, we test our model on real click data derived from Samsung Mall App.