Exploiting Behavioral Consistence for Universal User Representation

User modeling is critical for developing personalized services in industry. A common way user to learn representations that can be distinguished by their interests or preferences. In this work, we focus on universal representation model. The obtained are expected contain rich information, and applicable various downstream applications without further modifications (e.g., preference prediction profiling). Accordingly, free from the heavy work of training task-specific models every task as previous works. specific, propose Self-supervised Modeling Network (SUMN) encode behavior data into representation. It includes two key components. first one a new learning objective, which guides model fully identify preserve valuable information under self-supervised framework. other multi-hop aggregation layer, benefits capacity aggregating diverse behaviors. Extensive experiments benchmark datasets show our approach outperform state-of-the-art unsupervised methods, even compete with supervised ones.