动机
在推荐系统中,交叉特征(Cross Features)可以深入挖掘特征之间的潜在关系,提升模型效果。例如,把职业特征 o c c u p a t i o n = { b a n k e r , d o c t o r } occupation=\left\{banker,doctor\right\} occupation={ banker,doctor}和性别特征 g e n d e r = { M , F } gender=\left\{M,F\right\} gender={ M,F}进行交叉组合,可以得到新特征 o c c u p a t i o n _ g e n d e r = { b a n k e r _ M , b a n k e r _ F , d o c t o r _ M , d o c t o r _ F } occupation\_gender=\left\{banker\_M,banker\_F,doctor\_M,doctor\_F\right\} occupation_gender={ banker_M,banker_F,doctor_M,doctor_F}
Factorization Machine(FM)是挖掘交叉特征有代表性的模型,它的表达式如下:
y F M ( x ) = w 0 + ∑ i = 1 n w i x i + ∑ i = 1 n ∑ j = i + 1 n v i T v j ⋅ x i x j y_{FM}(x) = w_0+\sum^n_{i=1}w_ix_i+\sum^n_{i=1}\sum^n_{j=i+1}v_i^Tv_j·x_ix_j yFM(x)=w0+i