This paper investigates the importance of input optimization for topic modeling by illustrating authors’ interest modeling example. Finding authors interest is an important problem investigated for fulfilling different recommendation tasks in academics social networks. Previously state-of-art Author-Topic model used grouping of author on the basis his co-authors in a single document. Intuitively, an author has a set of co-authors in a single document (subgroup) and a set of co-authors in all documents (group, which contains subgroup as a subset).The authors tried three methods to find authors interests with respect to different inputs but with similar structure. Firstly, traditional AT (Author-Topic) model, in which each author of a subgroup is responsible for generating latent topics of that subgroup. Secondly, IAT (Inverse-Author-Topic) model, in which each word of a subgroup is responsible for generating latent topics of that subgroup and thirdly, GAT (Governing-Author-Topic) model, in which each author of a group, is responsible for generating latent topics of that group. Experimental results on the corpus downloaded from DBLP show that purposed method (GAT) with group based input outperformed baseline methods in terms of entropy measure, which shows better clustering ability of unsupervised methods. Additionally to ensure dominance of GAT all methods are applied for collaborator recommendation task and found GAT method to be the best, because of its optimal input.