Collaborative Filtering Using Associative Neural Memory

There are two types of collaborative filtering (CF) systems, user-based and item-based. This paper introduces an item-based CF system for ranking derived from Linear Associative Memory (LAM). LAM is an architecture that is founded on neuropsychological principles and is well studied in the neural network community. We show that our CF system has a user-based interpretation. Given a random subset of all users, our CF system is an unbiased estimator of predictions made from all users. We further apply standard neural network techniques, such as magnitude pruning and principle component analysis, to improve the system’s scalability. Results from experiments with the MovieLens dataset are shown.

• Type: Conference
• Year: 2003
• Where: IJCAI
• Authors: Chuck P. Lam
• Keywords: IJCAI 2003, IJCAI 2007, Item-based Cf, Linear Associative Memory, Neural Network