The proliferation of mobile devices and the fact that high-bandwidth connectivity is not available everywhere, has led to the development of hoarding algorithms. The aim of these algorithms is to select and prefetch data, mobile users might access when they are weakly connected in the future. The data is selected based on context information of users, such as their location. In this paper, we present a hoarding approach for semi-structured information items and introduce our Bounded Path Search algorithm (BPS) for the computation of the hoard list. We show by means of experimental evaluation that BPS outperforms existing hoarding techniques that use standard graph search algorithms by a factor of 2.7 in terms of hoard cache hit ratio.