A Hybrid B2B App Recommender System Alexandru Oprea 1 , Thomas Hornung 2 , Cai-Nicolas Ziegler 3 , Holger Eggs 1 , and Georg Lausen 2 1 SAP Commercial Platform, St. Leon-Rot & SAP Research, Darmstadt, Germany {alexandru.dorin.oprea,holger.eggs}@sap.com 2 Institute of Computer Science, Albert-Ludwigs-Universit¨ at Freiburg, Germany {hornungt,lausen}@informatik.uni-freiburg.de 3 American Express, PAYBACK GmbH, M¨ unchen, Germany cai-nicolas.ziegler@payback.net Abstract. Recommender systems are integral to B2C e-commerce, with little use so far in B2B. We present a live recommender system that operates in a domain where users are companies and the products being recommended B2B apps. Besides operating in an entire new domain, the SAP Store recommender is based on a weighted hybrid design, making use of a novel confidence-based weighting scheme for combining ratings. Evaluations have shown that our system performs significantly better than a top-seller recommender benchmark. 1 Introduction and Motivation The SAP Store caters to SME companies that aim to drive their business via B2B apps, e.g., for customer relation management or compliance. Many of these apps are geared towards specific industries and their needs. As the number of partners producing them is growing, so is the number of apps in the store itself and thus the complexity for the user (who represents a company) to actually find what he is looking for. To actively help the user, we propose a hybrid recommender system that addresses exactly the needs of this specific B2B scenario. The system puts to use both knowledge-based, collaborative, and content-based sub-recommenders. Moreover, we present a novel hybrid weighting scheme [1] that incorporates confi- dence scoring for the predictions produced, so that sub-recommenders contribute for recommendations according to their confidence weight. The system is live and can be used by logged-in users 1 . We have conducted empirical evaluations via hold-out testing that show that the recommender out- performs the non-personalized top-seller recommender. 2 Recommender System Architecture The architecture of the recommender is depicted in Figure 1. Overall, we have three different information sources for generating new recommendations: the 1 See http://store.sap.com F. Daniel, P. Dolog, and Q. Li (Eds.): ICWE 2013, LNCS 7977, pp. 490–493, 2013. c  Springer-Verlag Berlin Heidelberg 2013