RecSmart: Data augmentation to facilitate recommendation using skewed and sparse data of restaurant loyalty programs

Abstract

A good reward recommender system can effectively help retain customer by recommending personalized rewards. To provide accurate recommendation, the recommender needs to accurately predict a customer’s preference, an ability difficult to acquire. Conventional data mining techniques, such as association rule mining and collaborative filtering can generally be applied to this problem, but rarely produce satisfying results due to the skewness and sparsity of transaction data. In this paper, we describe a Recommender System built for a marketing company running Restaurant Loyalty Programs, the challenges we faced with the reward-response data and how we augmented the data to mitigate the challenges. We learnt that a collaborative filtering method based on ratings (e.g., GroupLens) to perform personalized reward recommendation is not sufficient. Instead, data augmentation can be more effective in handling skewness and sparsity of data.