An Efficient Approach to Similarity-Based Retrieval on Top of Relational Databases