| Patralekha Bhattacharya |
Recommendation systems have become increasingly popular in many industries, such as retail, movies, books, etc. In this paper, we build and validate a recommendation system for physical fitness products (e.g., exercise equipment, training manuals, books, etc.). All the recommendation models were built using implicit data, as we do not have explicit consumer ratings for the products. Instead, we use past user behavior (or purchases) as a proxy for user preferences whenever such past data is available. Our dataset consists of customers who purchased a fitness product during the year 2022 or earlier and those who have never purchased in the past. Customers were divided into three groups: (i) non-purchasers, (ii) recent purchasers, and (iii) older purchasers, different modeling methods were used to recommend items to different sets of customers. Recommendations were created both for existing products (which had been purchased by some customers in the past) and for newly introduced products (for which no purchase history exists). Different methods (such as collaborative filtering, content-based, trigger-based, and popularity-based techniques) were used. For the final solution, we used a hybrid system that used the strengths of both Python and SAS to develop the best set of recommendations for each customer. Within the first three months of implementation, our solution generated significant improvements in customer purchase behavior, compared to before the system was put in place. |