We evaluate whether the contents of comments on internet articles represent public opinion on tax policy through the case of income tax on cryptocurrency investment. Internet articles related to tax policy are easily accessible to anyone, and opinions can be freely expressed in comments. Therefore, it can be regarded that the comments on the article represent public opinion on the policy. In this study, based on text data collected by surveys through random sampling, we train a transformer model to build a sentiment analyzer that can classify positive/negative opinions on the new income tax policy on cryptocurrency and obtain the positive/negative ratio for the comments. As a result, we find that the rate of negative opinions in the policy responses to the comments on the articles was 80 to 90% of the total comments. It can be seen that this shows a considerable level of bias compared to the negative rate of about 30% of the open-ended survey through random sampling. The results shed light on the process of collecting public responses to government policies such as tax policies for determining whether or not to accept comments on articles as evidence of public opinion.