What's Missing in your Shopping Cart? A Set Based Recommendation Method for "Cold-Start" Prediction

This thesis studies the problem of predicting the missing items in the current user's session when there is no additional side information available. Many recommender systems fail in general to provide a precise set of recommendations to users with limited interaction history. This issue is regarded as the "Cold Start" problem and is typically resolved by switching to content-based approaches which require additional information. In this thesis, we use a dimensionality reduction algorithm, Word2Vec under the framework of Collaborative Filtering to tackle the "Cold Start" problem using only implicit data . We have named this combined method: Embedded Collaborative Filtering ECF. We are able to show that the ECF approach outperforms other popular state-of-the-art approaches in "Cold Start" scenarios by 2-10% regarding recommendation precision. In the experiment, we also show that the proposed method is 10 times faster in generating recommendations comparing to the Collaborative Filtering baseline method.