A text-mining approach to extracting social structures in a community is proposed. All textual resources associated with the actors of the community are collected. Using the vector space model, actor-term matrix is introduced to establish a mapping between the actors and the vocabulary. The problem of learning social ties from the actor-term matrix can be casted to an unsupervised or supervised learning framework. In the unsupervised framework, the relation between two actors is represented by the similarity between their documents. Then, the problem is translated into a similarity search problem. In the supervised model, social network from incomplete relation data is proposed. It is assumed that only a small subset of relations between the individuals is known. With this assumption, the social network extraction is translated into a text classification problem. By this setting, a text classifier is used for learning the unknown relations. The proposed framework is evaluated by applying to a true FOAF.