Integrating Attention Mechanism and Low- Rank Approximation into xDeepFM for Advanced Recommendation System

Abstract

Integrating Attention Mechanism and Low-Rank Approximationintion xDeepFM for Advanced Recommendation System ZHANG YIWAN Department of Computer Science The Graduate School of Hanyang University Supervisor Professor: Inwhee Joe Due to the sheer volume and variety of data in industrial production, manually creating these features can cause significant costs. Therefore, appropriate feature processing methods become crucial. Most existing feature processing methods abstract feature engineering as a feature search problem, i.e. searching for feature transformations that optimize the performance of a model. However, for automated feature engineering, the number of searches and the number of transformation combinations are huge. Because of this, we adopt a model based on factorization, which measures interactions in the form of a vector product that automatically learns the patterns of combined features and generalizes them to unseen features. Prior to this paper, the algorithms DeepFM, and xDeepFM. The LRCIN proposed in this paper is based on the improvement of the CIN network in the xDeepFM method, and then introduces a low-rank approximation method in the CIN network to reduce the number of parameters, an attention mechanism is added after the CIN to ensure the accuracy of the model. Our experimental results conclude that our method can effectively reduce the time complexity of the model and improve the accuracy of the model.