In this paper, we design a tag ranking method to provide multi-level keyword suggestion. The suggested keywords are used to effectively filter query results, which helps users to perform query specialization in social tagging systems. Besides, error-tolerant set containment queries are used to support various degrees of query generalization. We propose an index structure, which aggregates similar tag sets into clusters. A bounding mechanism is provided to efficiently deal with query processing for error-tolerant set containment queries on tag sets. These strategies can be used to support generalizations of a query. A systematic performance study is performed to show the effectiveness and the efficiency of the proposed methods.