Python実装 of 「推薦システム実践入門」, OREILLY, 2022

NOTE: This is not official

Official implementation.

環境構築

• Create virtualenv.
  • python3 -m venv .venv
• Activate virtualenv.
  • source .venv/bin/activate.fish (If you use fish shell.) Or you can call source .venv/bin/activate.
• Install dependencies
  • I installed manually like pip install scikit-learn.

Prepare MovieLends data

• Create data/ directory at the project root.
  • mkdir data
• Download moveilens data into data/ directory. Run the following command in the project root.
  • wget -nc --no-check-certificate https://files.grouplens.org/datasets/movielens/ml-10m.zip -P data
• Unzip downloaded data. Run the following command in the project root.
  • unzip -n data/ml-10m.zip -d data/

Finaly, the directories are as follows.

.
├── README.md
├── algorithms/
├── data/
│   ├── ml-10m.zip
│   └── ml-10M100K/
│       ├── movies.dat
│       └── ...
└── utils/

Performance of Algorithms

ユーザー1000人分のデータを用いた場合の各推薦アルゴリズムのメトリクス参考値。

Move to algorithm/ directory and run each algorithms' file.

Algorithm	RMSE	Precision@K	Recall@K	Source
RandomRecommender	1.88	0.0	0.0	random_recommender.py
PopurarityRecommender	1.06	0.0	0.0	popularity_recommender.py
AssociationRecommender	NaN	0.014	0.043	association_recommender.py
UMCRecommender	0.952	0.002	0.005	umc_recommender.py
RandomForestRecommender	0.996	0.0002	0.004	randomforest_recommender.py
SVDRecommender	1.04	0.020	0.065	svd_recommender.py
NMFRecommender	1.048	0.019	0.060	nmf_recommender.py
MFRecommender	1.027	0.010	0.034	mf_recommender.py
IMFRecommender	NaN	0.023	0.073	imf_recommender.py
BPRRecommender	NaN	0.022	0.069	bpr_recommender.py
FMRecommender	1.055	0.013	0.041	fm_recommender.py
LDAContentRecommender	NaN	0.0	0.0	lda_content_recommender.py
LDACollaboprationRecommender	NaN	0.018	0.057	lda_collaboration_recommender.py
Word2VecRecommender	NaN	0.001	0.003	word2vec_recommender.py
Item2VecRecommender	NaN	0.027	0.085	item2vec_recommender.py

未実装のアルゴリズム

• RNN (Session-based recommendations with RNN, Balaz Hidasi et al, 2015)
• item2vec (Neural item embedding for collaborative filtering, Oren Barkan and Noam Koenigsten, 2016 & Mihajlo E-commerce ub your inbox: Product recommendations at scale, Grbovic et al, 2015)
• BERT (BERT4Rec: Sequential recommendation with bidirectional encoder``` representations from transformer, Fei Sun et al, 2019)
• Nerural Collaborative Filtering (Xiangnan He, et al, 2017)
• Wide and Deep (Heng-Tze Cheng et al, 2016, Google)

DeepLearning for recommendationsライブラリ

• Recommenders (microsoft)
• Spotlight (maciejkula)
• RecBole (recbole)

注目論文

• Are We Really Making Much Progress? A Worring Analysis of Recent Neural Recommendation Approaches (Maurizio Ferrari Dacrema et al, 2019)
• A Survey on Contextual Multi-armed Bandit (Arxiv)