With the development of the Internet, the users share information usingWeb applications. Because of this reason, there is lots of information on the Web. The information includes not only high quality information, but also useless one. With the phenomena, the recommendation system appears on the Web. Existing information recommendation systems on the Web have known problems. One famous problem is coldstart. We tackle the cold-start problem for a new item in recommendation system. To alleviate cold-start for a new item, we use method for identifying representative reviewers in raters group and recommendation algorithm based on category correlations. The representative reviewers mean the users who represent their raters group. Namely, the ratings of the reviewers can represent the average ratings of other users. If there are the ratings for new items rated by the representative reviewers, then we can consider the ratings rated by many other users. We predict the ratings of these reviewers for a new item. To predict ratings, we use the recommendation algorithm based on the category correlations. This algorithm can draw the prediction results without ratings since the algorithm uses category information. We propose the prediction results of the representative reviewers as the representative ratings for a new item. We propose the algorithm to alleviate cold-start for a new item and show the reliability of our approach through tests.