Abstract
Generell innføring av Elektroniske anbefalingsystemer. Presentasjon av de vanligste metodene, inkludert K-Nærmeste Naboer, Matrise Faktorisering, Begrensede Boltzmann maskiner.
Presentasjon av Netflix konkurransen og noen av deltakerne med deres bidrag. Også egne bidrag til Netflix konkurransen, inkludert Matrise Faktorisering og en egenutviklet "Metric Neighbourhood Predictor", samt en skisse til en logaritmisk modell presenteres. Problemer med sampling fra denne type data belyses.
This thesis will introduce the reader to Recommender Systems, including some examples from different methods.
I introduce some standard implementations of Recommender Systems including Matrix Factorization by Singular Value Decomposition and the K-nearest neighbours method.
The Netflix Prize is introduced along with a short discussion of its strengths and shortcommings.
Some of the entries to the Netflix Prize are reviewed, and my own three implementations are covered. Including an outline of a novell logarithmic model. I also take a look at the difficulties in sampling from such a interconnected set as the netflix movie dataset. Finally some concluding remarks and a preview of the road ahead is presented.