Abstract:Advertising click-through rate (CTR) prediction aims to forecast the probability that a user will click on an advertisement in a given context, thus providing enterprises with decision support for product ranking and ad placement. However, CTR prediction faces challenges such as data sparsity and class imbalance, which adversely affect model training effectiveness. Moreover, most current CTR prediction models fail to fully explore the associations among user history, interests, and target advertisements from multiple perspectives, neglecting important information at different levels. To address these issues, this paper proposes an improved CTR prediction model based on the xDeepFM architecture. By integrating a multi-head attention mechanism, the model can simultaneously focus on different aspects of feature interactions, enhancing its ability to learn intricate patterns without significantly increasing computational complexity. Furthermore, replacing the linear model with a Factorization Machine (FM) model improves the handling of high-dimensional sparse data by flexibly capturing both first-order and second-order feature interactions. Experimental results on the Criteo dataset demonstrate that the proposed model outperforms other state-of-the-art methods, showing significant improvements in both AUC and Logloss metrics. This enhancement facilitates better mining of implicit relationships between features and improves the accuracy of advertising CTR prediction.

What problem does this paper attempt to address?

This paper attempts to solve several key problems in advertising click - through rate (CTR) prediction. These problems include data sparsity, class imbalance, and the fact that existing CTR prediction models fail to fully explore the multi - angle associations between user history, interests and target advertisements, ignoring important information at different levels. Specifically: 1. **Data Sparity**: In CTR prediction, due to the large but sparse amount of interaction data between users and advertisements, the model training effect is not good. The method proposed in the paper deals with high - dimensional sparse data by introducing Factorization Machine (FM), effectively capturing first - order and second - order feature interactions and improving the model's ability to handle sparse data. 2. **Class Imbalance**: There are fewer advertising click - through events compared to display events, resulting in an imbalance in the ratio of positive and negative samples. Although the paper does not directly discuss how to deal with class imbalance, it indirectly alleviates this problem by improving the overall prediction performance of the model. 3. **Multi - angle Feature Interaction**: Existing CTR prediction models often cannot comprehensively consider the complex relationships between user history, interests and target advertisements. By integrating the multi - head attention mechanism, the paper enables the model to simultaneously focus on feature interactions in different aspects, enhancing the model's ability to learn complex patterns without significantly increasing the computational complexity. To address the above challenges, the paper proposes an improved CTR prediction model based on the enhanced xDeepFM architecture. The main improvements include: - **Introducing the Multi - head Attention Mechanism**: Through the multi - head attention mechanism, the model can more effectively capture the correlation between the user's historical behavior sequence and the advertisement to be predicted, extract the user's latent interests, and at the same time reduce the interference of irrelevant historical behaviors on the prediction results. - **Replacing the Linear Model Component**: Replace the original linear model component with Factorization Machine (FM) to better learn low - order feature interactions and enhance the model's expressive ability. The FM model can flexibly capture the interactions between first - order features and second - order features, improving the model's ability to handle high - dimensional discrete data. The experimental results show that the proposed model outperforms other state - of - the - art methods on the Criteo dataset, with significant improvements in both AUC and Logloss metrics. This indicates that introducing the multi - head attention mechanism and using the FM model to replace the linear model component are effective strategies that can significantly improve the accuracy of CTR prediction.

An accuracy improving method for advertising click through rate prediction based on enhanced xDeepFM model

Click-Through Rate Prediction Algorithm Based on Modeling of Implicit High-Order Feature Importance

An adaptive hybrid XdeepFM based deep Interest network model for click-through rate prediction system

DeepFFMS: A Parallel Model for Advertise Click-Through Rate Prediction

XGBDeepFM for CTR Predictions in Mobile Advertising Benefits from Ad Context

Field-Aware Neural Factorization Machine for Click-Through Rate Prediction

A New Approach for Advertising CTR Prediction Based on Deep Neural Network via Attention Mechanism

Research on CTR Prediction Based on Deep Learning

Deep interaction network based CTR prediction model

DeepFM: An End-to-End Wide & Deep Learning Framework for CTR Prediction

An Attention-based Deep Network for CTR Prediction

Representation Learning-Assisted Click-Through Rate Prediction

A joint learning model for click-through prediction in display advertising

DeepFM: A Factorization-Machine based Neural Network for CTR Prediction

Balancing Efficiency and Effectiveness: An LLM-Infused Approach for Optimized CTR Prediction

Sparse Factorization Machines for Click-through Rate Prediction

Retrieval-Based Factorization Machines for CTR Prediction.

Deep Context Interest Network for Click-Through Rate Prediction

A Novel Click-Through Rate Prediction Model Based on Deep Feature Fusion Network