Accepted Paper: From Implicit to Explicit Feedbacks: A deep neural network for modeling the sequential behaviors of online users

Back to list of accepted papers


Anh Phan Tuan (Hanoi University of Science and Technology); Nhat Nguyen Trong (VCCorp Corporation); Duong Bui Trong (VCcorp Corporation); Linh Ngo Van (Hanoi University of Science and Technology); Khoat Than (Hanoi University of Science and Technology)


Choosing suitable advertisements to display on web pages still have some limitations: (1) insufficiently aggregate multiple types of user behavior, (2) do not model directly contextual information, and (3) suffer from cold start problem. Previous works show that multiple types of user behavior (implicit and explicit feedback) have distinct properties to provide a useful recommendation. However, these works exploit implicit and explicit behaviors separately and therefore ignore the semantic of interaction between users and items. In this paper, we propose a novel model namely Implicit to Explicit (ITE) which models the order of user actions directly. Intuitively, each user has to do implicit behavior (e.g., see the item) before making the explicit decision (e.g., click, purchase). Furthermore, we introduce an extended version of ITE, namely Contextual ITE (ConITE) which addresses the online advertising problem by employing a third-party source of information besides the users and the items, i.e., the contextual information of the current website showing ads. Contextual information is a useful resource to help select suitable advertisement as reported in numerous previous works, and it is expected that this knowledge can facilitate ConITE model practically. Finally, both ITE and ConITE can deal with the cold start problem by exploiting additional data such as item category, product description, etc. The experimental results show that ITE and ConITE outperform existing recommendation systems and also demonstrate the effectiveness of utilizing contextual information in ads banner system.