The theory of structural balance in signed networks plays an important role in information dissemination. Source localization is crucial for suppressing the spread of negative information, however, existing research on source localization has overlooked the theory of network structural balance. In this paper, we leverage the theory of structural balance in signed networks to develop a framework for source localization. This framework, called Direction Induced Search and Source Identification based Label Propagation (DISLPSI), consists of two main components: the Direction Induced Search (DIS) algorithm constructs a relaxed-induced search tree, and the Source Identification based Label Propagation (LPSI) algorithm validates the propagation source within the search tree. Meanwhile, in order to improve the proximity between the relaxed direction-induced search tree and the real propagation path, we propose the Adaptive Observation Node Selection (AONS) algorithm. Extensive simulation experiments demonstrate that the original LPSI algorithm performs very poorly in signed networks, proving that LPSI cannot be directly applied to signed networks. Conversely, the DISLPSI framework is effective in accurately localizing the source and robustness against the structural balance. Furthermore, the observer nodes selected by the AONS algorithm significantly enhance the source localization. We also find that the source localization framework is more advantageous in homogeneous networks than in heterogeneous networks.