A preference scoring technique for personalized advertisements on Internet storefronts

This paper describes a new personalized advertisement selection technique based on a customer's preference scores for product categories. This method performs well, despite having low data and analysis requirements, relative to other methods in use. Customer preference scores are updated based on a customer's initial profile, purchase history, and behavior in an Internet storefront, and are then used to select and display appropriate advertisements on Internet web pages when the customer visits the Internet storefront. Compared with currently available recommendation techniques such as collaborative filtering or rule-based methods, preference scoring techniques use only a single customer's data to select appropriate advertisements and do not require a learning data set, and yet have competitive performance and can reflect changes in a customers' preference. An experiment is performed to compare two alternative data storage structures, the preference table and the preference tree, with random selection and collaborative filtering.