Skip to content

Recommender Systems

Jump to bottom
Chih-Ming Chen edited this page Nov 29, 2020 · 3 revisions

Recommender System (RecSys)

Recommender System (RecSys) generally refers to the solutions that seeks to predict users' preference on given items. For example, in e-commerce sites, RecSys shows the products which the users would like to buy. In streaming services, RecSys displays the videos which could attract the users' interest. In search engines, RecSys helps rank the pages according to users' need. They are all RecSys and literally everywhere in our daily life.

Collaborative Filtering (CF)

To dig out each user's preference, the mainstream approach is to adopt the so-called Collaborative Filtering (CF). CF can be deemed a high-level concept that inference an individual user preference by utilizing enormous users' behavior. Concretely, to the best of our knowledge, we consider that almost all the developed algorithms belong to CF method. The major differences among them are how the algorithm models the enormous users' behavior, and then to benefit the recommendations. In this work, we compile a basic taxonomy of the surveyed CF models. Precisely, this work focuses on investigating the embedding-based approaches which directly embed the users and items in vector representations while remains other work in discussion section.

RecSys Algorithms

  • Statistical CF
    • User-based CF
    • Item-based CF
      • SWING
  • Factorization Model
    • User-to-Item Modeling
      • Matrix Factorization (MF)
    • Item-to-Item Modeling
      • Sparse LInear Method (SLIM), ICFM'11
    • Feature-based Modeling
      • Factorization Machine (FM), ICDM'10
  • Learning-to-Rank
    • Bayesian Personalized Rank (BPR), UAI'09
    • Weighted Approximate Rank Pair-wise (WARP) Loss, IJCAI'11 (paper WSABIE)
  • Graph Embedding
    • Random Walk
      • Item2Vec, arxiv
      • Enhanced Graph Embedding with Side Information (EGES) from Alibaba Group, KDD'18
    • Translation-based Modeling
      • Translation-based Recommendation, RecSys'17
    • Graph Convolution
      • GraphSage from Pinterest, KDD'18
      • LightGCN, SIGIR'20
    • Knowledge Graph
      • Knowledge Graph Attention Network (KGAT), KDD'19
      • Product Knowledge Graph from Walmart Labs, WSDM'20
  • Deep-Neural-Networks
    • Neural CF, WWW'17
    • Wide&Deep Learning from Google, DLRS'16
    • Deep Neural Networks (DNNs) from Youtube, RecSys'16
    • Deep Interest Network (DIN) from Alibaba Group, KDD'18
  • Others
    • Rating Prediction
    • Content-based Modeling
    • Bandit
    • Recurrent Neural Network
    • Reinforcement Learning

SMORe as the Framework of CF

Graph, Sampler, Mapper and Optimizer

Overview HERE ...

CM's work

  • TPR: Text-aware Preference Ranking for Recommender Systems, CIKM'20
    • User-Item-Text Triplet Sampler
    • Translation-based Mapper
  • Skewness Ranking Optimization for Personalized Recommendation, UAI'20
    • Skewness-Guided Optimizer
  • Item Concept Network: Towards Concept-based Item Representation Learning, IEEE TKDE 2020
    • Item Concept Graph
  • HOP-Rec: High-Order Proximity for Implicit Recommendation, RecSys'19
    • High-Order User-Item-Pair Sampler
    • Marginal Rank Optimizer
  • Leveraging Affective Hashtags for Ranking Music Recommendations, IEEE TAC'19
    • High-Order Neighborhood Sampler
  • NavWalker: Information Augmented Network Embedding, WI'18
    • Centroid-Aware Neighborhood Sampler
  • Query-based Music Recommendations via Preference Embedding, RecSys'16
    • High-Order Neighborhood Sampler
  • Music playlist recommendation via preference embedding, RecSys16
    • Playlist-based Graph
Clone this wiki locally