FINET: Fine-grained Feature Interaction Network for Click-through Rate Prediction

Abstract

Click-Through Rate prediction (CTR) is an critical task in recommender systems and computing advertisements because the prediction accuracy affects the user experience and the revenue of merchants. The instances in the CTR prediction task are usually sparse and high-dimensional. In real scenarios, an effective prediction relies on feature engineering which is difficult and requires domain knowledge. With the great success of deep neural networks (DNNs) in various fields, many models have been proposed to extract useful and important feature interactions from high-dimensional and sparse features. However, most models extract the feature interactions in a simple manner, and there is no comprehensive extraction of fine-grained and high-order feature interactions. In this paper, we propose a novel Fine-grained feature Interaction Network (FINET), which can automatically learn finegrained and high-order feature combinations of input features. We first use convolutional neural network (CNN) mechanism which leverages the fine-grained feature interactions to generate higher-order feature interactions. Then useful and important features can be selected through attention mechanism to further reduce the difficulty of learning a deep neural network. We conduct comprehensive experiments on two real-world datasets to demonstrate the effectiveness of FINET.