[ Recommender System ]
16. AutoEncoder meets Collaborative Filtering
( 참고 : Fastcampus 추천시스템 강의 )
paper : AutoRec : Autoencoders Meet Collaborative Filtering ( Sedhain et al., 2015 )
( https://users.cecs.anu.edu.au/~akmenon/papers/autorec/autorec-paper.pdf )
1. Introduction
AutoRec = Autoencoder + Collaborative Filtering
- vision & speech task에서 성공을 거둔 NN을 적용
- representation & computation 모두 장점
2. The AutoRec Model
\(\begin{array}{c} u \in U=\{1, \ldots, m\} \rightarrow r^{(u)}=\left(R_{u 1}, \ldots, R_{u n}\right) \in \mathbb{R}^{n} \\ i \in I=\{1, \ldots, n\} \rightarrow r^{(i)}=\left(R_{1 i}, \ldots, R_{m i}\right) \in \mathbb{R}^{m} \end{array}\).
loss function
- \(r\)와 \(r'\)사이의 reconstruction loss+ weight penalty
- reconstruction loss : \(h(\mathbf{r} ; \theta)=f(\mathbf{W} \cdot g(\mathbf{V} \mathbf{r}+\boldsymbol{\mu})+\mathbf{b})\)
- \(\left.\min _{\theta} \sum_{i=1}^{n} \| \mathbf{r}^{(i)}-h\left(\mathbf{r}^{(i)} ; \theta\right)\right \|_{\mathcal{O}}^{2}+\frac{\lambda}{2} \cdot\left(\|\mathbf{W}\|_{F}^{2}+\|\mathbf{V}\|_{F}^{2}\right)\).
RBM-CF ( Restricted Boltzmann Machine + CF ) vs AutoRec
- Objective Function
- RBM-CF : maximize log likelihoood
- AutoRec : minimize RMSE
- optimization
- RBM-CF : Boltzmann machine
- AutoRec : gradient-descent
- characteristic
- RBM-CF : discrete rating에 적합 & 각 rating 값에 대한 parameter 추정
- AutoRec : 더 적은 수의 parameter 사용 & overfitting 위험 낮음
MF vs AutoRec
- (Non) Linearity
- MF : linear
- AutoRec : non-linear
- embedding
- MF : user & item
- (item-based) AutoRec : item