This paper proposes an information recommendation system focusing on personal classification space based on bookmarks. This study handles the problem where the standard collaborative filtering method cannot offer recommendations with a better level of precision. This paper improves the collaborative filtering algorithm for users of social bookmark services. In this research, the classification-based method is proposed as the recommendation method, and Recall- DCG and Precision-DCG are proposed as evaluation scales. A user’s bookmarks are placed on his/her own classifying space made of tags. These bookmarks are transformed into a degree of similarity for recommendations. The degree is used to compare the personal classifying space with another’s space. Comparison with previous studies confirms the superiority of the method based on space classification. In particular, where the cos distance with the distribution weight added is used as similarity between items. The cos distance with the matching weight added is designed with the expectation of the effect of striking up the characteristics of other information. This proposed method shows a significant superiority in almost all experiments performed.