This paper studies the top-k possible shortest path (kSP) queries in a large uncertain graph, specifically, given two vertices
, a kSP query reports the top-k possible shortest paths from
. Different from existing solutions for uncertain graph problems, we adopt the
semantics have been studied widely in relational databases, the graph structure leads to more complexity in computational model. The main contribution of this paper lies in developing lower and upper bounds for the probability that one path is the shortest path in an uncertain graph, based on which, a filter-and-refine framework is proposed to answer kSP queries. We demonstrate the superiority of our method by extensive experiments.