From "Identical" to "Similar": Fusing Retrieved Lists Based on Inter-Document Similarities