In this paper, we propose an efficient approach to identify the opinion leader from group discussion. This approach is able to recognize the opinion leader without analyzing semantic and syntactic features, which may cost a lot more computing effort. We firstly propose algorithms to evaluate the degree of participation and the emotion expression from the speaking of each member during group discussion. Moreover, by conducting lab-scale experiment, a well-trained model, which is tested on single dataset as well as on cross dataset, is obtained to recognize the opinion leader. Finally, we conduct a field experiment to evaluate the proposed system in a real world setting. The results show that the accuracy of opinion leader identification could achieve to 94.68% on Berlin dataset, 76% on Youtube data and 73.33% on live group discussion. Thus, with this simple and efficient system, opinion leader can be successfully identified in various conditions.