Online rating systems are unable to provide objective evaluations of products due to the presence of water armies and malicious raters. Therefore, it is crucial to establish a reputation measurement model based on rating behavior for the healthy development of online rating systems. Existing user reputation measurement methods only take into account the difference between the user's rating information and product quality, regardless of user rating behavior patterns. Combining user rating bias and behavior patterns, a new reputation measurenment algorithm was proposed, and the algorithm considered not only the extremes of user rating frequency, but also the total number of user ratings. The extensive experimental results for two empirical datasets show that the accuracy of the new algorithm for identifying random ratings can be improved by up to 17% compared to the classical algorithm, and has better performance for solving cold start and robustness problems.