Whether planning a trip or researching medical symptoms, many complex and exploratory search tasks involve intense cross referencing of multiple sources. One has to determine the popularity and trustworthiness of competing options and avoid missing valuable but less common options. Traditional search systems leave this task to users, while automatic result clustering approaches often fail to produce structures coherent enough to meet these needs. We introduce a novel approach that builds coherent topical landscapes from search results. By combining the rich but sparse representations of knowledge bases with word vector models that are shallow but dense in coverage, we produce structures more coherent than alternative approaches. We instantiate this in a prototype interactive system and explore which elements users find useful. Across three information seeking scenarios, we found that participants valued both the structure and distribution information provided by our system for complex exploratory searches and, surprisingly, simple searches as well.