Our purpose is to implement a Visual Analytics tool for exploring topic variants in text corpora. The overlapping bi-clustering methods extract multiple topics from the documents, but the interpretation of the results remains difficult. We make the assumption that bi-cluster overlaps are articulation points between high-level topics, and their multiple variants and viewpoints. We propose to extract and visualize a hierarchical structure of bi-cluster overlaps, allowing to explore the corpus and to discover unsuspected viewpoints.