A Click Conversion Rate Model of E-Commerce Platforms Aiming at Effective Data Sparse

Click conversion rate is an important indicator that e-commerce platforms pay attention to. The user's behaviors toward the product on the e-commerce platform, to some extent, indicates a level of love for the product. In order to precisely predict the click conversion rate, this paper suggests a click conversion rate prediction model for e-commerce platforms based on various user behaviors in heterogeneous network. First, in view of the sparseness of interaction information data between users and items on the e-commerce, we take advantage of the ability to discover comparable nodes with the meta-path based random walk method. Simultaneously, the time decay function is introduced in order to construct a sequence of user nodes and a sequence of item nodes with comparable behaviors in the near future. Second, users' interests are diverse. Given that the self-attention mechanism can calculate the degree of association between each node and its neighbors, we propose a multi-head self-attention mechanism to aggregate neighborhood information representing the central node in order to capture the user's interests while improving the model's stability. Finally, to address the phenomenon that users have multiple behaviors under the e-commerce platform, we introduce behavioral weight factors. At the same time, we fill the user and item sequences generated according to the meta-path into the heterogeneous attention network and initialize the importance of user behaviors. Then, we propose the MUB-HAN (Multiple User Behavior Heterogeneous Attention Network) model to further improve the accuracy of click conversion rate. The experiments indicate that the model can generate superior predictions of click conversion rates based on user behavior toward items in e-commerce platforms while incorporating a huge quantity of auxiliary information.