Preference completion has been widely applied in multiple fields, such as social selection and recommendation systems. In these fields, each user only has a partial ranking for certain entities, and it is often unrealistic for users to provide a complete ranking for all entities. The goal of preference completion is to infer a complete preference ranking for all items for each individual based on their incomplete preference ranking. However, in many cases, users may give an unreasonable ranking or a noisy version of their preference ranking. Therefore, how to more accurately complete users' preferences in the presence of noise is a continuous research issue. On the other hand, data in preference completion problems often have multi-source, dynamic, and massive features, so how to improve the algorithm's time performance while ensuring accuracy is also a key issue.