INVESTIGATION AND ANALYSIS OF STATISTICAL ATTENTION MECHANISMS IN CLICK-THROUGH-RATE PREDICTION: THE IMPACT OF LAYER NORMALIZATION AND INTERACTION COMPONENT INTEGRATION

Date

2024-9-5

Author

Büyükbaş, Ege Berk

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

54
views

16
downloads

Cite This

The accurate prediction of Click-Through Rate (CTR) is a key metric for enhancing user experience and optimizing revenue in online shopping and e-commerce businesses. This study explores the suitability of various statistical attention mechanisms—mean attention, max attention, mean-max, mean-std attention, and bitwise attention—under different hyper-parameter candidate sets within the most common and conventional CTR prediction algorithms. By conducting extensive experiments across the most commonly used open-source datasets for CTR prediction, this empirical study examines whether these attention mechanisms can effectively boost the informational utility of each field's low-dimensional feature embedding, potentially leading to improved prediction accuracy. Our findings show that each attention mechanism behaves uniquely across different algorithms and datasets. The application of these attention mechanisms to traditional CTR prediction models may demonstrate significant improvements in prediction performance by implicitly focusing on relevant features and their interactions. This research aims to contribute to the field of CTR prediction by providing a comprehensive analysis of how different attention mechanisms can enhance the predictive ability of well-known conventional CTR prediction algorithms and offer insights for the future development of more sophisticated and accurate CTR prediction systems.

Subject Keywords

Click-Through Rate (CTR), Statistical Attention Mechanisms, Interaction Component, Deep Learning, Feature Embedding

URI

https://hdl.handle.net/11511/111058

Collections

Graduate School of Natural and Applied Sciences, Thesis