2012/06/18 by Aaron Defazio, Defazio, Aaron, Tib rio S. Caetano +2
Computer Science · Social Sciences · #Advanced Graph Neural Networks #FOS: Computer and information sciences #Human Mobility and Location-Based Analysis #Information Retrieval (cs.IR) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Recommender Systems and Techniques
paper · pdf · doi:10.48550/arxiv.1206.4622
openalex publication_date 2012/06/18 · openalex created_date 2025/10/24 · openalex updated_date 2026/07/28
Item neighbourhood methods for collaborative filtering learn a weighted graph\nover the set of items, where each item is connected to those it is most similar\nto. The prediction of a user's rating on an item is then given by that rating\nof neighbouring items, weighted by their similarity. This paper presents a new\nneighbourhood approach which we call item fields, whereby an undirected\ngraphical model is formed over the item graph. The resulting prediction rule is\na simple generalization of the classical approaches, which takes into account\nnon-local information in the graph, allowing its best results to be obtained\nwhen using drastically fewer edges than other neighbourhood approaches. A fast\napproximate maximum entropy training method based on the Bethe approximation is\npresented, which uses a simple gradient ascent procedure. When using\nprecomputed sufficient statistics on the Movielens datasets, our method is\nfaster than maximum likelihood approaches by two orders of magnitude.\n