Analysis of Online Grocery Recommendation Systems

Recommender system has been recognized as the most effective method for information overload problem. But a generic approach with small changes is used to get results in most types of recommender systems. Grocery recommendations are unique to this generic approach because of the possibility of reordering items. This reordering criterion can very well be taken as a measure of preference compared to general rating system that is followed. This paper focuses on using the idea of reorders to make and compare different systems of online grocery recommendations. Two types of systems have been tested for grocery recommendation taking reorders into account and a comparison between them has been made at the end. The methods and metrics used for making these two recommender systems including truncated SVD, cosine similarity and k-nearest neighbours. The paper has also discussed the results of reordering criteria for grocery sales.