Studying the process of network public opinion reversal is of great significance for guiding public opinion toward a positive direction. Currently, the research on opinion reversal mainly focuses on the construction of dynamic models and analysis of simulations and the results of which have a certain theory value. However, whether these models are applicable to the real social network environment has not been tested. For studying the process of public opinion reversal, we build a model according with the realities, and make an in-depth analysis of the typical case of opinion reversal. Some rules are found from the observation and statistics: the fundamental reason of public opinion reversal is the conflicting news. Spreading of news affects the opinions of the group. The news properties, including transmission rate, credibility, opinion polarity, publication date and the degree of message source determine the extent to reverse. Based on these rules, parameters of news properties are set, and a model of opinion reversal is proposed by combing the information dissemination with opinion evolution. Simulation results show that the transmission rate of news, the credibility of news, and the degree of message source have a positive influence on the margin of reversal. The influence of credibility is more dramatic than that of transmission rate. Moreover, the public opinion would be reversed more quickly and completely if the conflicting news is released more easily. The proposed model can fit the actual data, which is helpful for understanding and explaining the process of network public opinion reversal, and provides theoretical basis for guiding the network public opinion.