Post

History Embedding based Latent Factor Models

Based on the following lectures
(1) “Recommendation System Design (2024-1)” by Prof. Ha Myung Park, Dept. of Artificial Intelligence. College of SW, Kookmin Univ.
(2) "Recommender System (2024-2)" by Prof. Hyun Sil Moon, Dept. of Data Science, The Grad. School, Kookmin Univ.

Learning Objectives


  • 표현 학습(Representation Learning)
    • 사용자와 아이템을 공동의 잠재요인 공간에 표현하는 방법
    • 매칭 강도 추정 시 내적(Inner Product) 등 선형 유사도 함수를 적용함
    • 저차원(Low-rank) 유사도 구조를 효율적으로 포착할 수 있음
  • 매칭 함수 학습(Matching Function Learning)
    • 사용자-아이템 쌍을 입력으로 하여 매칭 함수를 직접 학습하는 방법
    • 복잡하고 비선형적인 매칭 함수를 근사할 수 있음

DMF


  • 문제 의식: 아이디 임베딩(ID Embedding) 입력 표현의 한계점
    • 아이디 임베딩 방식은 초기 표현(식별자)의 정보량이 부족하여 학습이 느리거나 성능이 제한됨
  • DMF(Deep Matrix Factorization): 사용자-아이템 상호작용 행렬과 그 전치 행렬을 초기 표현으로 사용하여 저차원 표현 학습을 수행하는 모형
    • Xue, H. J., Dai, X., Zhang, J., Huang, S., & Chen, J.
      (2017, August).
      Deep matrix factorization models for recommender systems.
      In IJCAI (Vol. 17, pp. 3203-3209).

Notation

  • $u=1,2,\cdots,M$: user idx
  • $i=1,2,\cdots,N$: item idx
  • $\mathbf{Y} \in \mathbb{R}^{M \times N}$: user-item interaction matrix
  • $\overrightarrow{\mathbf{u}}_{u} \in \mathbb{R}^{K}$: user latent factor vector
  • $\overrightarrow{\mathbf{v}}_{i} \in \mathbb{R}^{K}$: item latent factor vector
  • $\hat{y}_{u,i}$: interaction probability of user $u$ and item $i$

How to Modeling

01

  • user latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{u}}_{u} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{u*}) \end{aligned}\]
  • item latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{v}}_{i} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{*i}) \end{aligned}\]
  • Predict interaction probability of user $u$ and item $i$:

    \[\begin{aligned} \hat{y}_{u,i} &= \cos(\overrightarrow{\mathbf{u}}_{u}, \overrightarrow{\mathbf{v}}_{i})\\ &= \frac{\overrightarrow{\mathbf{u}}_{u} \cdot \overrightarrow{\mathbf{v}}_{i}}{\Vert \overrightarrow{\mathbf{u}}_{u} \Vert \cdot \Vert \overrightarrow{\mathbf{v}}_{i} \Vert} \end{aligned}\]

DeepCF


  • 문제 의식: 표현 학습 방식과 매칭 함수 학습 방식의 상호 보완적 관계
    • 표현 학습은 저차원 유사도 구조를 포착하여 사용자, 아이템의 일반화된 표현을 도출하는 데 강점
    • 매칭 함수 학습은 사용자와 아이템 간 복잡하고 비선형적인 상호작용 과정을 근사하는 데 강점
  • DeepCF: 표현 학습 모듈과 매칭 함수 학습 모듈을 병렬 학습하는 앙상블 모형
    • Deng, Z. H., Huang, L., Wang, C. D., Lai, J. H., & Yu, P. S.
      (2019, July).
      Deepcf: A unified framework of representation learning and matching function learning in recommender system.
      In Proceedings of the AAAI conference on artificial intelligence (Vol. 33, No. 01, pp. 61-68).
  • Components
    • CFNet-rl: Representation Learning
    • CFNet-ml: Matching Function Learning
    • CFNet: CFNet-rl & CFNet-ml Ensemble

Notation

  • $u=1,2,\cdots,M$: user idx
  • $i=1,2,\cdots,N$: item idx
  • $\mathbf{Y} \in \mathbb{R}^{M \times N}$: user-item interaction matrix
  • $\overrightarrow{\mathbf{u}}_{u} \in \mathbb{R}^{K}$: user latent factor vector
  • $\overrightarrow{\mathbf{v}}_{i} \in \mathbb{R}^{K}$: item latent factor vector
  • $\overrightarrow{\mathbf{z}}_{u,i}$: predictive vector of user $u$ and item $i$
  • $\hat{y}_{u,i}$: interaction probability of user $u$ and item $i$

How to Modeling

03

  • CFNet is CFNet-rl & CFNet-ml Ensemble

    \[\begin{aligned} \hat{y}_{u,i} &= \sigma(\overrightarrow{\mathbf{w}} \cdot [\overrightarrow{\mathbf{z}}_{u,i}^{\text{(RL)}} \oplus \overrightarrow{\mathbf{z}}_{u,i}^{\text{(ML)}}]) \end{aligned}\]

CFNet-rl

  • user latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{u}}_{u} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{u*}) \end{aligned}\]
  • item latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{v}}_{i} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{*i}) \end{aligned}\]
  • predictive vector of user $u$ and item $i$:

    \[\begin{aligned} \overrightarrow{\mathbf{z}}_{u,i} &= \overrightarrow{\mathbf{u}}_{u} \odot \overrightarrow{\mathbf{v}}_{i} \end{aligned}\]
  • if use CFNet-rl as a single prediction module:

    \[\begin{aligned} \hat{y}_{u,i} &= \sigma(\overrightarrow{\mathbf{w}} \cdot \overrightarrow{\mathbf{z}}_{u,i}) \end{aligned}\]

CFNet-ml

  • generate user latent factor vector through linear transformation:

    \[\begin{aligned} \overrightarrow{\mathbf{u}}_{u} &= \mathbf{W} \cdot \mathbf{Y}_{u*} \end{aligned}\]
  • generate user latent factor vector through linear transformation:

    \[\begin{aligned} \overrightarrow{\mathbf{v}}_{i} &= \mathbf{W} \cdot \mathbf{Y}_{*i} \end{aligned}\]
  • predictive vector of user $u$ and item $i$:

    \[\begin{aligned} \overrightarrow{\mathbf{z}}_{u,i} &= \text{MLP}_{\text{ReLU}}(\overrightarrow{\mathbf{u}}_{u} \oplus \overrightarrow{\mathbf{v}}_{i}) \end{aligned}\]
  • if use CFNet-ml as a single prediction module:

    \[\begin{aligned} \hat{y}_{u,i} &= \sigma(\overrightarrow{\mathbf{w}} \cdot \overrightarrow{\mathbf{z}}_{u,i}) \end{aligned}\]

J-NCF


  • 문제 의식
    • 아이디 임베딩(ID Embedding): 사용자와 아이템 표현을 무작위로 초기화한 후 매칭 함수 학습을 수행하므로 표현 학습이 미흡함 (ex. NCF)
    • 히스토리 임베딩(History Embedding): 사용자-아이템 상호작용 행렬과 그 전치 행렬을 활용하여 표현 학습을 수행하나 매칭 함수는 선형 유사도 함수에 의존함 (ex. DMF)
    • 앙상블(Ensemble): 표현 학습과 매칭 함수 학습을 분리하여 수행하므로 각 모듈이 서로의 학습을 보완하거나 강화하지 못함 (ex. CFNet)
  • J-NCF(Joint Neural Collaborative Filtering): 표현 학습과 매칭 함수 학습을 통합 훈련(Joint Training)하는 모형
    • Chen, W., Cai, F., Chen, H., & Rijke, M. D.
      (2019).
      Joint neural collaborative filtering for recommender systems.
      ACM Transactions on Information Systems (TOIS), 37(4), 1-30.

Notation

  • $u=1,2,\cdots,M$: user idx
  • $i=1,2,\cdots,N$: item idx
  • $\mathbf{Y} \in \mathbb{R}^{M \times N}$: user-item interaction matrix
  • $\overrightarrow{\mathbf{u}}_{u} \in \mathbb{R}^{K}$: user latent factor vector
  • $\overrightarrow{\mathbf{v}}_{i} \in \mathbb{R}^{K}$: item latent factor vector
  • $\overrightarrow{\mathbf{z}}_{u,i}$: predictive vector of user $u$ and item $i$
  • $\hat{y}_{u,i}$: interaction probability of user $u$ and item $i$

How to Modeling

02

  • user latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{u}}_{u} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{u*}) \end{aligned}\]
  • item latent factor vector representation learning:

    \[\begin{aligned} \overrightarrow{\mathbf{v}}_{i} &= \text{MLP}_{\text{ReLU}}(\mathbf{Y}_{*i}) \end{aligned}\]
  • matching function learning:

    \[\begin{aligned} \overrightarrow{\mathbf{z}}_{u,i} &= \text{MLP}_{\text{ReLU}}(\overrightarrow{\mathbf{u}}_{u} \oplus \overrightarrow{\mathbf{v}}_{i}) \end{aligned}\]
  • Predict interaction probability of user $u$ and item $i$:

    \[\begin{aligned} \hat{y}_{u,i} &= \sigma(\overrightarrow{\mathbf{w}} \cdot \overrightarrow{\mathbf{z}}_{u,i}) \end{aligned}\]
This post is licensed under CC BY 4.0 by the author.