A unified approach to building hybrid recommender systems

Content-based recommendation systems can provide recommendations for "cold-start" items for which little or no training data is available, but typically have lower accuracy than collaborative filtering systems. Conversely, collaborative filtering techniques often provide accurate recommendations, but fail on cold start items. Hybrid schemes attempt to combine these different kinds of information to yield better recommendations across the board.