Academic literature recommendation technology based on two-layer attention network

Recently, academic literature recommendation for learners has become an important topic. Recently deep learning based models are used in literature recommendations, which follow a similar Embedding&MLP paradigm. However, in the base model and its follow-up model Deep Interest Network (DIN), users only have to click or not click on items, limiting the expression of users' specific interests. Therefore, we propose a two-layer attention model (BIH) based on DIN. BIH adds specific behaviors to the user behavior sequence and adds a behavior attention layer, which can learn the expression of user interests more accurately. Experiments on a public dataset and users' behavior logs of academic literature demonstrate the effectiveness of proposed approaches, which achieve superior performance on both Macro-F1 and Micro-F1 compared with the base model and DIN.