ivanwang0730/recommendation-paper

Recommendation-Paper

CTR论文-MEITUAN

A Deep Behavior Path Matching Network for Click-Through Rate Prediction, WWW2023(short)

1. Intuition:

• 层次化Attention: Current Click Behavior > Click Behavior > Other Behavior
• 对比学习解决稀疏性

2. 模型去选择Key Points...? Behavior Adaptive Hierarchical Interest

Decision-Making Context Interaction Network for Click-Through Rate Prediction, AAAI2023

1. Motivation:

• Explicit decision-making context: the relation of the intra-page items
• Implicit decision-making context: the information introduced by personalization

2. Model:

• Irrelevance Suppression Unit (ISU): 过滤Explicit和Implicit context中不相关的items，MLP
• Relevance Interaction Unit (RIU): Query为clicked item K,V为scope内的itmes cross-attention做选择
• Adaptive Interest Aggregation Unit (AIAU): 类似DIN利用cross-attention对历史行为兴趣选择，单层Transfomer Decoder Layer

3. Datasets:

• Avito & MeituanAds

Contrastive Learning for Sequential Recommendation, ICDE2022

1. Motivation: 解决数据稀疏，只用传统信号监督网络不能学到很好的用户兴趣表示
2. CL4SREC: Sequential Loss + Contrastive Learning Loss
3. Three views: crop, mask, reorder 类似simcse中的dropout技巧