Friendbook is a semantic-based friend recommendation system for social networks that recommends friends based on users' lifestyles rather than social graphs. It uses sensors in smartphones to discover users' lifestyles from daily activities and measures lifestyle similarity between users. Users are recommended as friends if their lifestyles are highly similar. Lifestyles are extracted from "life documents" of daily activities using Latent Dirichlet Allocation. Friendbook also incorporates feedback to improve recommendation accuracy. It was implemented on Android smartphones and evaluated on small and large-scale tests, finding recommendations accurately reflected real-life friend preferences.