Inexpensive data collection and storage technologies and a global thirst for information have led to data repositories so large that users may become disoriented and unable to locate desired items, leading to dissatisfaction and session abandonment. Automatically personalizing interfaces can resolve navigational difficulties and improve satisfaction. Recommender systems, which suggest items to users, have become a popular automatic personalization tool, but quality and speed continue to haunt practical applications. In this paper, we identify two fundamental problems with recommender systems. We address these problems by presenting a detailed framework for classifying, understanding and generating personalization heuristics, including recommender systems. We present a high-level survey of current personalization systems showing which areas of the framework have received the most (and least) attention. This analysis provides guidance for future studies and a novel paradigm for coordinating development within the field.