In recent years, the rapid development of mobile information technology has made online social network service enter the stage of rapid development. Users could exchange information on social networks. But those unreliable users who have strong influences on the network tend to make the public generate emotional deviation and information biases, which seriously affects the internet safety. Therefore, it is extremely urgent to research the user trust degree measurement model on social networks. Integrated with the network link structure, emotion and heat topics of the user nodes on social networks to identify and evaluate the information resources and user trust degree on it. Also, combined the real social network data to do a comparison experiment between the UTD and four kinds of machine learning methods, so as to confirm the scientific nature of the model. The experiential results showed that the PSB user trust degree measurement method we mentioned is better than other methods on the aspects of accuracy, specificity and sensitivity. The reliable users tend to have positive emotions and the information they post, the time they created the accounts are longer as well as they are easier to be followed and mentioned by other users comparing with unreliable users. However, the information the unreliable users post tends to be contained more labels and website addresses.