Abstract: In order to solve the problem of information overload in the post bar, a method of information filtration was proposed based on the users commenting behavior. After analyzing the properties of the recommended posts, the importance of an individual user was evaluated by the PageRank algorithm, in which the weight of replies to the posts among users and the weight of reply intervals were taken into consideration. The users with a high PageRank score were then taken as a cluster center in k-means clustering. The similarity between two groups of users (one from the clustering analysis and the other from the recommending system) was calculated by a collaborative filtering algorithm. The posts with high correlations to the users were presented as the recommended results. Experimental results show that the proposed method performs better than the recommending methods in use.