Public opinion always has an important influence on the policy process. The development of social networking sites and applications has given the public more opportunities to express their views about the related policies. In cases where the coverage of the traditional hearing system challenged the policy process, how to measure accurately the public concern and attitudes regarding policies based on online public generated content using a data mining method will be very important issue in policy informatics. Our paper provides a probabilistic topic modeling approach, mainly based on Latent Dirichlet Allocation (LDA) model, to transform the complex semanteme of online public opinions into the values could be measured. A simple case could show the usefulness of the too toward policy analysts also be provided and discussed briefly.