Does exposure to like- and non-like-minded information lead to political incivility? Few previous studies have investigated this question, and the results have been mixed. There are two conflicting possibilities: (i) if individuals are frequently exposed to like-minded political information, their preexisting beliefs are reinforced and they are more likely to use uncivil language, and (ii) if individuals are frequently exposed to non-like-minded information, they often feel negative emotions and therefore are more likely to be uncivil. To evaluate these two competing hypotheses, the present study analyzes data from Japanese Twitter using a semi-supervised machine learning method. The results show that individuals who are exposed to non-like-minded information are more prone to political incivility.