Natural Language Retrieval of Grocery Products
Grocery shopping is a fundamental everyday activity for most people. For many customers, a grocery list is central to the grocery shopping experience. Studies have suggested that the shopping list serves, e.g., as a memory aid, as a tool for budgeting, and as a way to efficiently organize the routine shopping visits. The way people write shopping lists contrasts with the way stores maintain product information. Namely, users tend to use natural language and generic descriptions, whereas stores tend to use highly structured and specific information about products. To bridge the semantic gap between customers and stores, we have developed an information retrieval system for grocery products. Our goal is to use this system as part of a mobile grocery assistant that allows users to create and manage shopping lists using natural language. The information retrieval system is then used to find products from the store that match the items in the customerâ€™s shopping list. In this paper we describe and evaluate our information retrieval system. We have developed the system using fourteen months of shopping basket data from a large Finnish supermarket. To evaluate the system, we gathered real shopping lists from customers of the supermarket, and performed a user evaluation of the system using these lists. The evaluation indicates that our system achieves over 80% precision at rank one, and the mean average precision is around 70% for the top ten query results. We also compare our system against an off-the-shelf information retrieval system, and show that our system significantly improves retrieval accuracy.
Keywords: Grocery Retrieval, User Evaluation, Retrieval Models