People usually use trust and reputation to cope with uncertainty which exists in the nature and routines. The existing approaches for inferring trust rely on homogeneous relations. In other words, trust is just inferred by a homogeneous relation. In this paper, we present a new model for inferring trust using heterogeneous social networks; we use relation extraction to make a trust relation from the other relation such as friendship and the college relation and then introduced an algorithm to infer trust using extracted relation. In order to get higher performance, we extend relation extraction problem by proposing a genetic algorithm. This algorithm is more scalable, interpretable and extensible in comparison with prior ones. We also present a new algorithm for inferring trust in a social network. Using these methods, we conclude higher accuracy for trust values. Our claims are evaluated by experimental results.