질문, 건의, 요청 사항은 방명록(guest)에 올려 주십시오.

2011-10-04_facto neighbor_in_recommend.pdf
 


발표자

임용섭


개요

In this paper, they propose a new recommend system combining
neighbor hood and factorization models. They obtain significant 
improvement by considering implicit feedback such as information
that one rated some item.  Their method outperforms the previous 
methods appeared in Netflix competition in RMSE measure. Also 
they propose a new metric to complement  RMSE in which the gap
of the best and a "average" method is just 20%. This metric measures 
how well the method recommends top-K items, the proposed method
is also the best in finding top-K items.

신고
Posted by AALab
이전버튼 1 2 3 4 5 6 7 ··· 47 이전버튼