Abstract
E-commerce activities among prominent retailing firms in modern times is inconceivable without the ubiquitous presence of recommender systems. This article brings forth the more advanced topics like content based and context-aware methods. Content-based methods use the actions and ratings of the users to match the user to new items based on past ratings. The objective here is to create user profiles and subsequently subject the profiles to classification algorithms. Knowledge-based systems are for more customized products with little history of usage and therefore little past data to help in recommendations. Such systems rely on either case-based recommendations or on a set of relevant constraints to identify appropriate recommendations. And finally, ensemble recommender systems help in combining the prediction power from multiple data sources. Finally, the author presents a discussion on the evaluation methods for recommender systems. The article is aimed towards both academic and managerial audiences.
TopIntroduction
Author has already emphasized upon the importance of recommender systems for today’s firms irrespective of domain such as grocery, apparel, electronic goods, entertainment services etc in the first part of this series of two articles. Today most retailing companies have proprietary recommender systems that are often highly customized and already into the next generation in terms of bringing together many different algorithms together. Today’s recommender systems are no longer standalone proof of concept systems trying to gain credibility but rather are considered to be a major source of revenue booster for the organizations. Popular literature estimate suggests that almost 70% of content consumed online by Netflix viewers is through personal recommendations and according to Gomez-Uribe & Hunt (2016) they could save almost USD 1.2 billion by avoiding cancellations. Similarly, popular literature reports almost 30-35% revenues of Amazon originating through its recommender systems. It is now well established that these systems help in better customer engagement, upselling, customer retention and in boosting revenues.
Author started his discussion with basic collaborative systems and its variants in the earlier chapter and also emphasized on the growing need and logic (Aggarwal, 2015) for combining various methods. This second chapter of the series takes the on from exactly the point where we left off. Author shall start his discussion from content based systems and its variants, then move on to knowledge based systems and ensembles, subsequently author discusses context aware systems and evaluation methods for recommender systems. Finally, the author concludes the chapter with a brief discussion and emphasizing on the need for more work on recent topics in the field.
Key Terms in this Chapter
Knowledge-Based Recommender Systems: For products not related to past purchase or having uncommon characteristics ordinary ratings may not be useful and such systems are subject to either constraint based systems or case based systems in order to make appropriate recommendations.
Ensemble and Hybrid-Based Methods: Both ensemble and hybrids involve bringing together more than one algorithm for recommendation, however, ensembles involve off-the shelf algorithms whereas hybrids involve custom algorithms for the context.
Content-Based Recommender Systems: Content based systems work on data from users’ own ratings as well as attributes or features of the product to make recommendations by using unstructured data elements and applying NLP and other methods on the same.