This paper develops a unified game-theoretical framework for data classification and network discovery based on pairwise influences in multivariate categorical choices. Data points, interpreted as individuals, are connected through a signed weighted graph and take values according to a voting rule that aggregates the influence of attractive (friend-like) and repulsive (enemy-like) neighbors. The framework consists of two complementary games: in the direct voting game, the network is exogenous and choices are endogenous; in the inverse voting game, choices are observed and the network structure is endogenous. This distinction allows the same behavioral mechanism to address both classification and discovery problems within a common strategic framework. In particular, the direct voting game yields a classification methodology that generalizes the K-nearest neighbors approach. On the theoretical side, we derive conditions for the existence of Nash equilibria and show that recognizing them is NP-complete. On the empirical side, we study three applications and show that the proposed framework improves goodness of fit in both data classification and network discovery settings.