Scholarlink

Click-Through Rate (CTR) prediction, which estimates the probability of a user clicking on a particular item, plays an important role in online advertising and recommender systems. However, data sparsity remains a significant challenge in achieving accurate CTR prediction results. Fortunately, semi-supervised node classification techniques offer a potential solution to this problem. To this end, this paper proposes a semi-supervised node classification model named Soft Feature Selection Graph Neural Network (SFSGNN), which is designed for alleviating data sparsity in CTR prediction. Firstly, the integration of key modules, including node generation, graph construction, and soft feature selection networks, enables SFSGNN to effectively leverage the limited known user-item interaction data to predict the unknown user-item interactions. Furthermore, the SFSGNN can effectively address the data sparsity issue by utilizing the prediction results as augmented data for training the CTR prediction models. Extensive experiments conducted on three real-world datasets demonstrate the effectiveness of SFSGNN in alleviating data sparsity in CTR prediction.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SFSGNN: Soft Feature Selection Graph Neural Network for Alleviating Data Sparsity in Click-Through Rate Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Junming Zhou,
  • Chao Chang,
  • Weisheng Li,
  • Ronghua Lin,
  • Zhengyang Wu,
  • Yong Tang

摘要

Click-Through Rate (CTR) prediction, which estimates the probability of a user clicking on a particular item, plays an important role in online advertising and recommender systems. However, data sparsity remains a significant challenge in achieving accurate CTR prediction results. Fortunately, semi-supervised node classification techniques offer a potential solution to this problem. To this end, this paper proposes a semi-supervised node classification model named Soft Feature Selection Graph Neural Network (SFSGNN), which is designed for alleviating data sparsity in CTR prediction. Firstly, the integration of key modules, including node generation, graph construction, and soft feature selection networks, enables SFSGNN to effectively leverage the limited known user-item interaction data to predict the unknown user-item interactions. Furthermore, the SFSGNN can effectively address the data sparsity issue by utilizing the prediction results as augmented data for training the CTR prediction models. Extensive experiments conducted on three real-world datasets demonstrate the effectiveness of SFSGNN in alleviating data sparsity in CTR prediction.