What constitutes relevant information to an individual may vary widely under different contexts. However, previous work on pervasive information systems has mostly focused on context-aware delivery of application-specific information. Such systems are only able to operate within narrow application domains and cannot be generalized to handle other heterogeneous types of information. To fill this gap, we propose a context-aware system for information integration that can handle arbitrary information types and determine their relevance to the user's current context. In contrast to existing model-based approaches to context reasoning, we log user interaction and perform usage mining using OLAP to discover context-dependent preferences for different information types. This allows us to build a more generic and adaptive system that automatically selects the most relevant content and presents it to the user in a succinct manner that supports ease of consumption and comprehension