Abstarct: In Interactive Question Answering (IQA) systems, users will ask their questions in a natural language and receive related responses. Because of the interactions in these systems, users can raise descxriptions about their questions and increase accuracy. Human conversations have many features, one of which is the control of the conversation. In these conversations, control may be exchanged between people. Such interactions have a mixed-initiative feature. In these conversations, at any one time, a person may take the initiative and change the direction of the conversation. Mixed-initiative feature is an important and effective aspect of helping people work together to solve problems and make people as a team more productive. The purpose of this research is to detect the conversation controller in interactive question answering systems using statistical techniques. This method, unlike other methods, which mainly uses semantic and grammatical analysis, uses statistical methods without semantic and cognitive analysis and is therefore independent of the language. In order to detect the controller in this way, it is only necessary to prepare the databaxse and train the system with this databaxse. In this research, a databaxse is designed for this purpose. To detect the controller, methods such as euclidean distance, hamming, minkowski, naive bayes, k-nearest neighbour, k-means, support vector machine, cosine similarity, and TF-IDF have been used. The best performance in this system is the support vector machine model with 91.49%.