In one embodiment, the system receives unlabeled data from one or more sensors. The one or more sensors are associated with a computing device of a user. The system clusters the unlabeled data into one or more unlabeled clusters. The system determines, based on one or more rules, at least one label for at least one unlabeled cluster of the one or more unlabeled clusters. The system expands, by a natural language processing algorithm, the at least one label for the at least one unlabeled cluster to produce one or more semantic terms associated with the at least one unlabeled cluster. The system generates, based on the one or more semantic terms, a personalized graph associated with one or more behavioral patterns of the user.