Abstract: With the rapid development of social media in recent years, the access to information has been broadened, but the spreading of incredible information has been facilitated at the same time, which brings a series of negative impacts to cyber security. Compared with the traditional online media, the information in social media is more open and complicated, giving rise to great challenges to judge online information credibility for individuals. How to filter the incredible information becomes an urgent problem. In the existing research on the assessment of information credibility in social media, lots of effort has been involved in extracting the useful factors for credibility assessment, but the processing of noisy data is neglected, and a large number of useless tweets can be included in the evaluation process, resulting in the deviation of the information credibility assessment. So it is particularly important to select the significant tweets for information credibility assessment. This paper takes the topic factor and conformity of users into consideration to relieve the impact of noisy data, such as conformity retweeting, on information credibility assessment, and uses Bayesian network to establish an evaluation model for information credibility in social media. Then we verify the effectiveness of our model using a real dataset.