Similitude Attentive Relation Network for Click-Through Rate Prediction

Deng, Hangyu; Wang, Yulong; Luo, Jia; Hu, Jinglu

doi:10.1109/ijcnn48605.2020.9207521

Cited by 3 publications

(2 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where W s and W t are learnable parameters, ReLU(•) is Rectified Linear Unit which guarantees the output is nonnegative vectors [20] and • 2 is two-norm. It is worth mentioning that mapping embedding vectors to nonnegative vectors is usually beneficial [21,22]. On the one hand, it can be considered as an efficient dimension reduction via Nonnegative Matrix Factorization [23], on the other hand, it restricts e i ,j within [0, 1] and avoids getting negative values, which may neutralize the final relational features in (10).…”

Section: Similarity Based Relational Modulementioning

confidence: 99%

See 1 more Smart Citation

Relation‐Level User Behavior Modeling for Click‐Through Rate Prediction

Deng

Tian

Luo

et al. 2021

IEEJ Transactions Elec Engng

Self Cite

View full text Add to dashboard Cite

Many recent user behavior based click-through rate models adopt a similar item-level paradigm: learn the user representation from a list of item representations via a sequence model and/or a pooling mechanism. However, sequence models are usually sensitive to the exact order of the behavior sequence, while item-level pooling mechanisms simply neglect the chronological information. In this paper, we balance the two approaches by decomposing the long item sequence into a group of extremely short sequences (item pairs) and conducting relational reasoning on them. More specifically, the relational reasoning mechanism consists of two parts, which are designed for capturing various transitional patterns in the behavior sequences. An attentive pooling layer is employed to emphasize those relation-level signals that are highly related to the target item. Therefore, our approach is essentially a middle ground between the previous two approaches. To verify the effectiveness of our method, we conduct extensive experiments on three public datasets. Experimental results show that our methods achieve better performance than others. Besides, we explore the properties of our model and verify the effectiveness of each component by controlled experiments.

show abstract

Section: Similarity Based Relational Modulementioning

confidence: 99%

“…It is possible to take h i ,j as the output of the module directly. However, MLP itself is believed to be not so good at handling entangled tasks in CTR prediction [14,22,24]. Instead, we separate relational reasoning and representation learning by wrapping h i ,j with items via inner product operation.…”

Section: Mlp Based Relational Modulementioning

confidence: 99%