Item Recommendation Based on Monotonous Behavior Chains

In the field of e-commerce,recommendation systems can accurately provide users with products and services of potential interest,thereby enhancing users’online shopping experience.“Explicit”feedback and“implicit”feedback are mostly studied in two relatively independent research fields.In the actual interaction process between users and commodities,there is a kind of signal between the two Monotonic dependence,that is,sparse and reliable explicit signals must imply dense and noisy implicit signals.In this paper,a special“monotonic behavior chain”structure is proposed to constrain the two signals,and a series of usercommodity interaction behaviors is mapped into a user-commodity multi-stage binary interaction diagram.The two feedback signals were combined and the complete interaction was simulated between the user and the product.Then a depth model framework GAERE was proposed based on the graph auto-encoder,which converts the matrix completion problem of the traditional recommendation system into the problem of graph link prediction.Four realistic data sets were applied to evaluate the effectiveness of the proposed method.The model shows competitiveness on standard collaborative filtering benchmarks.In addition,the application of graph convolutional network was further explored to process graph structure data in recommendation system from the perspective of user behavior intention.