This chapter introduces Collaborative filtering-based recommendation systems, which has become an integral part of E-commerce applications, as can be observed in sites like Amazon.com. It will present several techniques that are reported in the literature to make useful recommendations, and study their limitations. The chapter also lists the issues that are currently open and the future directions that may be explored to address those issues. Furthermore, the authors hope that understanding of these limitations and issues will help build recommendation systems that are of high accuracy and have few false positive errors (which are products that are recommended, though the user does not like them).
In our day-to-day life, in order to deal with life’s many choices, we depend on recommendations from those whom we trust. With the advent of information overload, the number of choices is overwhelming and the manual “word-of-mouth” recommendations do not scale well. Recommendation systems are those that help deal with these increasing choices, by automating “word-of-mouth” recommendations. They suggest items of interest to the user based on user’s stated preferences either implicit or explicit.
Recommendation systems are being used by an ever-increasing number of E-commerce sites to help consumers find products that match their interest. The recommendation is based on the user’s preferences and known preferences of similar users. Recommendation systems provide tools for people to collaborate and help each other for better recommendations. In the recent decade, Recommendation systems have become an important application area and the focus of considerable academic and commercial interest.
E-Commerce sites like Amazon.com, WalMart.com, Netflix.com, etc offer millions of products in their online catalog. It is difficult for consumers to choose from this large number of options available. Collaborative filtering-based Recommender systems automatically suggest recommendations to the user thereby reducing the choices the user has to go through before making a purchase.
From the business perspective, E-Commerce sites can be personalized to suit each user’s needs and do targeted marketing and it has been observed that the effectiveness of such Collaborative filtering based recommendations vastly exceed those of traditional methods and content-based filtering and result in increased sales (Linden, G., Smith, B., & York, J., 2003). Collaborative filtering based recommendation systems enhance e-commerce sales, by converting browsers into buyers, increase cross-selling by suggesting additional products and it builds loyalty, provided the recommendation system generates high-quality recommendations.
Key Terms in this Chapter
Model-Based Methods: The process of learning the relationship model between the users and items given the user-item rating matrix.
Collaborative Filtering: Recommendation system that enables people to help each other to perform filtering through collaboration. If two users share similar preferences for a subset of items, they are likely to have similar preferences for other items in the collection.
Co-Clustering: The process of clustering on both the user and item space in parallel. Also, called as simultaneous clustering.
Semantic Heterogeneity or Synonymy: A case where a word is used to mean different things.
Imbalance in Data: Dataset where multiple subsets are present, with the sizes of the subsets varying a lot, with some subset of huge size and some subset of negligible size.
Coverage: Coverage is measured as the percentage of user-item pairs in the test set for which recommendations can be made.
Cluster Prototypes: The set of samples that represent a given cluster.