recommendation-itemKNN

implementation of item based collaborative filtering

■ 데이터셋 Description

Dataset	#Users	#Items	#Interactions
ML-small
ML-1m	6,040	3,706	1,000,209
Last-FM	23,566	48,123	3,034,796
Amazon-Book	70,679	24,915	847,733

-> 데이터셋은 Knoledge Graph 관련 실험을 위해 metadata, triple이 존재하는 것 기준으로 사용

■ Evaluation Results

- item POP & item KNN

[Movie Lens - small]

Model	HR@5	HR@10	HR@20	NDCG@5	NDCG@10	NDCG@20
itemPOP	0.188	0.292	0.408	0.122	0.1575	0.186
itemKNN	0.304	0.405	0.518	0.230	0.263	0.290

[Movie Lens - 1M]

Model	HR@5	HR@10	HR@20	NDCG@5	NDCG@10	NDCG@20
itemPOP	0.311	0.455	0.638	0.208	0.255	0.301
itemKNN	0.368	0.521	0.707	0.255	0.306	0.351

[Last - FM]

Model	HR@5	HR@10	HR@20	NDCG@5	NDCG@10	NDCG@20
itemPOP	0.344	0.470	0.621	0.242	0.284	0.322
itemKNN

- BPR

emb size experiment, K=10

K size experiment, emb size=16

■ Issue

- Abook은 Time, Computing 문제로 추후 진행 예정

- Last-FM 데이터셋 활용시 memory error

item KNN에 Last-FM 데이터셋 적용시 너무 큰 Matrix가 생성되어 오류

-> Dataframe sim 계산까지는 -> 로컬 메모리 32G 기준으로 8000명씩 잘라서 matrix 생성 및 평가

■ 데이터셋 Reference

- Movie-Lens small

출처 : https://www.kaggle.com/rounakbanik/the-movies-dataset

- Movie-Lens 1M

출처 : https://grouplens.org/datasets/movielens/1m/

- Amazon-Book & Last-FM

출처 : https://github.com/xiangwang1223/knowledge_graph_attention_network