tm4roon / survey

Survey on machine learning.
14 stars 1 forks source link

Neural News Recommendation with Long- and Short-term User Representations #67

Open tm4roon opened 4 years ago

tm4roon commented 4 years ago

Neural News Recommendation with Long- and Short-term User Representations

ニュース記事推薦の研究。ユーザの長期的な嗜好と短期的な興味の両方を考慮した、ユーザのベクトル化手法を提案。モデルは(1) News encoder (2) User encoderの2つのコンポーネントにより構成される。

User encoderでは、(1) long-term user representationをshort-term user representationに利用するGRUの初期値として導入する方法 (2) long-term user representationとshort-term user representationを結合する方法のの2つが提案されている。

long-term user representationは、User IDのembedding layerにより獲得。short term user representationは、ユーザのブラウジング履歴を利用し、news encoder + GRUにより獲得。

文献情報