cross-validation is markedly superior for small
data sets; this fact is demonstrated dramatically by Goutte (1997) in a
reply to Zhu and Rohwer (1996)
Jackknifing
the probability of selecting the "best" does
not converge to 1 (as the sample size n goes to infinity) for leave-v-out
cross-validation unless the proportion v/n approaches 1
以下の二つの論文で扱われたデータセット。やはり、因子分析に向いたものとしてしましまさんから久保山さんにお薦めがありました。
-- M. Zanker, M.Jessenitschnig, D. Jannach and S. Gordea, Comparing Recommendation Strategies in a Commercial Context, IEEE Intelligent Systems, 2007, vol. 22, May/June.
-- M. Zanker, M.Jessenitschnig, Collaborative feature-combination recommender exploiting explicit and implicit user feedback, 11th IEEE Conference on Commerce and Enterprise Computing (CEC), Vienna, Austria, 2009.