In this paper, we propose a method which can obtain more appropriate results for XML search. Since our previous approach has a problem that large-sized XML fragments tend to have large scores in CO topics, it is not suitable for users to return the ranked results in the descendant order of their scores. To cope with this problem, we generate a final search result of XML fragments by integrating smaller-sized ones with higher scores. We apply our approach to the probabilistic scoring model which is often used in the field of XML search area.