The proliferation of the social Web due to increased user participation poses a challenge as well as presents an opportunity to examine the collective behavior of users for various business applications. In this work, we leverage the collective knowledge embedded in the social relationships of users on the network to predict user preferences and future behavior. We extract social dimensions in the form of overlapping communities that capture the behavioral heterogeneity in directed and signed social networks. We present an extension of signed modularity, namely Structural Balance Modularity (SBM). We first propose a metric Structural Balance Index (SBI) that determines users’ degrees of affiliation towards various communities by harnessing the concept of the generalized theory of structural balance. We then incorporate SBI into the signed modularity to define SBM. It takes into account the density as well as the sign (positive or negative) of the links between users on the network. A genetic algorithm is developed that optimizes the SBM, thereby maximizing positive intra-community connections and negative inter-community connections. The discovered latent overlapping communities represent affiliations of users with similar preferences and mutual trust relationships captured by the signs of connections exerting differential effects on users’ behaviors. Thereafter, we ascertain which communities are relevant for a targeted behavior by using discriminative learning. The computational experiments are performed on Epinions real-world dataset, and the results clearly demonstrate the effectiveness and efficacy of our proposed approach.