Decision information system based on rough set is an important processing form in data mining. Attribute reduction is one of the most critical issues in rough set theory. Chen et.al. proposed an attribute reduction algorithm for decision information system based on the covering generalized rough sets. In this paper, we first point out that Chen's algorithm applies to the fifth, the sixth and the seventh type of covering generalized rough sets. Then we establish the equivalent relationship between covering and the reflexive and transitive relation. Moreover, we prove that attribute reduction based on covering in this paper is a special case of the one based on general binary relation proposed by Wang et.al. As a result, a new attribute reduction algorithm with lower time complexity is designed. With these discussion, the research provides theoretical and practical significance.