Exposure bias widely exists in recommender systems, particularly in the case of with implicit feedbacks. It seriously influences user's satisfaction of recommendations. There are a number of methods for mitigating the exposure bias from different perspectives. In this paper, we survey the publications that focus on addressing the exposure bias issue in RS with the help of causal inference ideas. We propose a simple taxonomy consisting of bias discovery, evaluation estimator, recommendation modeling, ranking algorithm for the debiasing methods in our study. Based on the taxonomy, we discuss how those methods are beneficial to recommender systems to mitigate the exposure bias using causal graph and propensity score. Finally, we conduct the challenges and point out the future research directions.