For human mobility studies across many disciplines, mobile network data serves as a primary source of human footprints with geo-referenced and time-stamped records of human communication activities. Nevertheless, the quality of mobility information provided by mobile network data is usually not satisfactory on many users. Due to the nature of human communications, individual trajectories inferred from mobile network data are often substantially incomplete , and the pattern of missing locations is not uniform over time but is highly related to communication activities. In this paper, we propose a novel hierarchical approach based on tensor factorization to reconstruct such incomplete individual trajectories. The data-driven simulation shows that, with ground-truth locations in precision of 200 meters, our approach can recover a trajectory from 10% of its known locations with a distance error below 750 meters, which outperforms the existing proposals in the literature.