[RecSys] 한 눈에 보기 정리 - 1

Recommender System

1. 개요
   1. 사용 이유
   2. 문제 정의
   3. 평가 지표
      1. Offline Test
      2. Online Test
2. Simple Aggregate
   1. Popularity
   2. Rating
3. 연관분석 : Association Analysis / Association Rule Mining
4. 컨텐츠 기반 필터링 : Content-based Filtering / Content-based Recommendation
   1. TF-IDF : Term Frequency - Inverse Document Frequency
5. 협업 필터링 : Collaborative Filtering
   1. NBCF : Neighborhood-Based Collaborative Filtering - Memory-Based CF
      1. UBCF : User-Based Collaborative Filtering
      2. IBCF : Item-Based Collaborative Filtering
      3. Rating Prediction
         1. UBCF - Absolute/Relative Rating
         2. IBCF - Absolute/Relative Rating
         3. Top-N Recommendation
   2. MBCF : Model-Based Collaborative Filtering 
      1. Non-Parametic : kNN, SVD
      2. MF : Matrix Factorization
      3. Implicit Feedback VS Explicit Feedback
         1. MF for Implicit Feedback - ALS : Alternating Least Square
         2. BPR : Baysian Personalized Ranking
      4. Embedding
         1. Word2Vec
            1. CBOW
            2. SG : Skip-Gram
            3. SGNS : Skip-Gram Negative Sampling
         2. Item2Vec
      5. ANN : Approximate Nearest Neighbor
         
         1. ANNOY : spotify에서 개발한 tree-based ANN
         2. HNSW : Hierarchical Navigable Small World Graphs
         3. IVF : Inverted File Index
         4. PQ : Product Quantization
      6. Context-Aware Recommendation
         1. CTR : Click-Through Rate Prediction
         2. FM : Factorization Machine
         3. FFM : Field-aware Factorization Machine
         4. GBM : Gradient Boosting Machine
            1. XGBoost
            2. LightGBM
            3. CatBoost
   3. Deep Learning based Recommendation
      1. Recommender with MLP
         1. NCF : Neural Collaborative Filtering
      2. Recommender with AutoEncoder
         1. AutoRec
         2. CDAE
      3. Recommender with GNN
         1. NGCF : Neural Graph Collaborative Filtering
         2. LightGCN
      4. Recommender with RNN
         1. GRU4Rec
      5. CTR Prediction with DL
         1. Wide & Deep
         2. DeepFM
         3. DIN : Deep interest Network
         4. BST : Behavior Sequence Transformer
      6. MAB : Multi-Armed Bandit
         1. Greedy Algorithm
         2. Epsilon-Greedy Algorithm
         3. UCB : Upper Confidence Bound
         4. Thompson Sampling
         5. LinUCB