Need Help Finding Something? Why Not Ask Your (Virtual) Community! A Collaborative Filtering Based Recommendation Engine in SAS ®
Ben Elbert · 2015
People have likely been recommending everything from where to eat to whom to meet since the dawn of man, but in the last twenty years we have seen the rise of automated, data driven, recommendation engines (also known as recommender systems, expert systems, etc.). One popular approach to recommending products to potential shoppers is collaborative filtering (CF), which relies on weighting the quantity, propensity, or rating of other shoppers by the similarity of the potential shopper’s transactions to those of the community of shoppers (traditionally at an individual level). This traditional implementation of a CF-System has the drawback that new or early tenure shoppers may not have enough informative transactional data to make valuable recommendations. In this paper we present a model that, when used with information available to banks and creditors such as geographic, demographic, and credit bureau data, overcomes the cold-start problem and make valuable recommendations to new or low tenure cardholders in addition to multi-transaction persons. We also demonstrate how to implement a model that leverages large amounts of data in SAS ® and optimizes run time by providing tips for testing the model to determine optimal parameters. Finally, we discuss how to extend our CF-System to areas of interest beyond the recommendation challenge.