Click-through rate prediction based on feature interaction and behavioral sequence

• Published in: International journal of machine learning and cybernetics Vol. 15; no. 7; pp. 2899 - 2913
• Main Authors: Wang, Yingqi, Ji, Huiqin, Yu, Junyang, Han, Hongyu, Zhai, Rui
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2024; Springer Nature B.V
• Subjects: Artificial Intelligence; Behavior; Complex Systems; Computational Intelligence; Control; Deep learning; Design; Engineering; Interaction models; Mechatronics; Neural networks; Original Article; Pattern Recognition; Prediction models; Recommender systems; Robotics; Systems Biology; User behavior; Attention mechanism; Click-through rate; Behavior sequence; Feature interaction
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-023-02072-5

Click-through rate prediction is one of the hot topics in the recommendation and advertising systems field. The existing click-through rate prediction models can be classified into feature interactions and behavior sequences. Feature interaction models form new feature combinations by fusing different features. The behavior sequence models capture the user’s interests by considering the historical behavior and using an attention mechanism to model the relationship between the target item and the behavior sequence. However, the existing click-through rate prediction techniques either ignore both aspects or only consider one, limiting prediction performance. In order to solve the above problems, we propose a click-through prediction model (CFIBS) that combines feature interaction and behavioral sequence in this paper. Firstly, the Global-Local Gate Module and Post-LN Informer are proposed to extract the user’s interests from user behavior sequences to improve training efficiency. In addition, we introduce auxiliary losses to supervise the extraction of user interest features. Secondly, in the interest update layer, we introduce an attention mechanism based gated recurrent unit to enhance the relationship between interest representation and the target item. Finally, for non-temporal features, we propose a Multi-Cross Layer to increase the nonlinear ability of the model. Experiments show that our model can effectively improve the click-through rate prediction accuracy of advertisements. The codes will be available at https://github.com/jihuiqin2/sequence_ctr .