Simple tf-keras code of paper [Deep Neural Networks for YouTube Recommendations].
Jupiter notebook files recommendation.ipynb
with detailed description.
유튜브 추천 시스템을 논문과 여러 자료를 참고하여 텐서플로 케라스로 구현하였습니다.
주피터 노트북 파일 recommendation.ipynb
에 자세한 설명이 포함되어 있습니다.
This code contains 'candidate generation' only.
This code does not include 'ranking'.
이 코드는 'candidation generaion' 부분만 포함합니다.
'ranking' 은 포함하지 않습니다.
Check recommendation.ipynb
file
- Result
Movie Recommendation for user 1 (movie_ID): [53, 1064, 1111, 510, 1488, 239, 230, 1157, 281, 162]
- Matrix (User 1 MAP)
### Model Prediction VS Random Pick ###
top N = 1 -------------------
model = Rank 1 Recall : 0.0072992700729927005 (1/137)
model = Rank 1 Precision : 1.0 (1/1)
random = Rank 1 Recall : 0.0 (0/137)
random = Rank 1 Precision : 0.0 (0/1)
top N = 10 -------------------
model = Rank 10 Recall : 0.014598540145985401 (2/137)
model = Rank 10 Precision : 0.2 (2/10)
random = Rank 10 Recall : 0.0 (0/137)
random = Rank 10 Precision : 0.0 (0/10)
top N = 20 -------------------
model = Rank 20 Recall : 0.014598540145985401 (2/137)
model = Rank 20 Precision : 0.1 (2/20)
random = Rank 20 Recall : 0.0072992700729927005 (1/137)
random = Rank 20 Precision : 0.05 (1/20)
top N = 30 -------------------
model = Rank 30 Recall : 0.029197080291970802 (4/137)
model = Rank 30 Precision : 0.13333333333333333 (4/30)
random = Rank 30 Recall : 0.0364963503649635 (5/137)
random = Rank 30 Precision : 0.16666666666666666 (5/30)
top N = 40 -------------------
model = Rank 40 Recall : 0.058394160583941604 (8/137)
model = Rank 40 Precision : 0.2 (8/40)
random = Rank 40 Recall : 0.014598540145985401 (2/137)
random = Rank 40 Precision : 0.05 (2/40)
top N = 50 -------------------
model = Rank 50 Recall : 0.06569343065693431 (9/137)
model = Rank 50 Precision : 0.18 (9/50)
random = Rank 50 Recall : 0.014598540145985401 (2/137)
random = Rank 50 Precision : 0.04 (2/50)
mAP@50 = 0.19529945953383454
- python
- tensorflow 2