Loading...

Media is loading
 

Paper Type

Complete

Description

In the field of Computational Advertising, the click-through rate (CTR) prediction is a primary research issue. The accuracy of CTR prediction affects the revenue of ad platforms and enhances the users’ experience. To predict CTR, users’ behavior sequence can be used to extract user's interest. However, due to the complexity of user behavior, many researchers only considered a single type of behavior and neglected other types. Multiple behavior sequences reflect users’ interest in various dimensions, and the latent interaction between these behavior sequences may better reveal users’ interests. To tackle this problem, we propose the Deep Multi-Behavior Interest Network (DMBIN) model, which employs multi-behaviors with hierarchical attention mechanism. Our experimental analysis on a real dataset shows that the DMBIN model is able to effectively extract users’ interests and capture the latent interaction between different behaviors. Compared with the benchmark models, the AUC and the Logloss of DMBIN are significantly better.

Paper Number

1459

Comments

SIG DSA

Share

COinS
 
Aug 10th, 12:00 AM

Advertising Click-Through Rate Prediction Based on Multi-Behavior Sequences

In the field of Computational Advertising, the click-through rate (CTR) prediction is a primary research issue. The accuracy of CTR prediction affects the revenue of ad platforms and enhances the users’ experience. To predict CTR, users’ behavior sequence can be used to extract user's interest. However, due to the complexity of user behavior, many researchers only considered a single type of behavior and neglected other types. Multiple behavior sequences reflect users’ interest in various dimensions, and the latent interaction between these behavior sequences may better reveal users’ interests. To tackle this problem, we propose the Deep Multi-Behavior Interest Network (DMBIN) model, which employs multi-behaviors with hierarchical attention mechanism. Our experimental analysis on a real dataset shows that the DMBIN model is able to effectively extract users’ interests and capture the latent interaction between different behaviors. Compared with the benchmark models, the AUC and the Logloss of DMBIN are significantly better.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.