What ’ s for Dinner ? Recommendations in Online Grocery Shopping

We present an approach for online grocery shopping recommendation, framed as a binary classification problem. We describe the methods we used to extract features from a recent data set by Instacart, and we compare the results of our recommendation experiments with those of traditional and enhanced market basket analysis approaches. We use approximate recommendation precision to measure the success of our recommender, and ROC AUC values to evaluate some of our underlying classifiers. Although our recommendations were better than the market basket analysis baseline, our best models learned primarily the purchase frequency of products.