This document summarizes an approach to generate semantic user profiles from informal communication exchanges like emails, meetings, and chats. It extracts keywords, named entities, and concepts from communications to represent user profiles. Similarities between user profiles are then calculated to infer relationships. An experiment on email data found profiles based on concepts best correlated with human judgments of user similarity, outperforming profiles from keywords and entities alone. Future work involves applying the approach to organizations and connecting profiles to linked open data.