A click-through rate model of e-commerce based on user interest and temporal behavior

Abstract

In the advertising and marketing of e-commerce platform, click rate prediction is directly related to the revenue of e-commerce platform. In this paper, we propose an advertising click-through rate prediction model based on user interest and temporal behavior. First, in response to the phenomenon that traditional recurrent neural networks (RNNs) have the advantage of processing temporal data but ignore the time interval of input sequences, this paper designs a structure of time-weighted gated recurrent units (TW-GRU) to predict users’ short-term interest and long-term interest. In addition, we introduce auxiliary losses to supervise the extraction of user interest features and the learning of TW-GRU. Secondly, we introduce an attention mechanism-based gated recurrent unit A-GRU in the model to enhance the impact of interests associated with target advertisements. Finally, for the potential high-dimensional implicit information of non-temporal features and behaviorally hidden user interests, this paper proposes a click-through rate prediction model that fuses user interests with implicit generalization features. Experiments show that our model can effectively capture users’ interests and interest update process associated with target products from users’ historical behaviors, and then can effectively improve the click-through rate prediction accuracy of advertisements.