Learning to Recommend With Multiple Cascading Behaviors

Most existing recommender systems leverage user behavior data of one type only, such as the purchase in E-commerce that is directly related to business Key Performance Indicator (KPI) conversion rate. Besides key behavioral data, we argue other forms behaviors also provide valuable signal, views, clicks, adding a product shopping carts and so on. They should be taken into account properly quality recommendation for users. In this work, contribute new solution named short Neural Multi-Task Recommendation (NMTR) learning from multi-behavior data. We develop neural network model capture complicated multi-type interactions between users items. particular, our accounts cascading relationship among different types (e.g., must click on before purchasing it). To fully exploit signal multiple behaviors, perform joint optimization based multi-task framework, where treated task. Extensive experiments two real-world datasets demonstrate NMTR significantly outperforms state-of-the-art are designed learn both single-behavior Further analysis shows modeling particularly useful providing sparse have very few interactions.