Harmful social interactions, such as injurious feather pecking in poultry and tail biting in swine, reduce both animal welfare and efficiency. While these traits are heritable, application of breeding is still limited due to the lack of proper genetic models and precise phenotyping methods for large groups. In the near future, large scale longitudinal data on social interactions will become available thanks to developments in computer vision and artificial intelligence. Here we present models to simulate and analyze such data, which are an extension of the classic social genetic model. Latent traits were defined to represent the tendency of individuals to be engaged in behavioral interactions, distinguishing performer and recipient effects. Binary interaction records were simulated and subsequently analyzed using generalized linear mixed models. Results show that high accuracies of estimated breeding values can be obtained (0.4-0.7), despite the low observed-scale heritability of the binomial trait (0.05-0.2). We conclude that our model can be promising for breeding value estimation for social traits in large groups.