Typing entities in structured sources such as Wikipedia has been well studied to construct English knowledge bases automatically. However, there still remain two tough challenges in typing entities in Chinese-pedia. The first one is that structured information from Chinese-pedia cannot assign entities fine-grained types due to its inaccuracy and coarseness. The other challenge is the incompletion of Chinese-pedia, which means we can only use limited attribute fields to type entities. In this paper, we propose a novel Hierarchical Neural System (HNS) to infer fine-grained types for entities in Chinese-pedia. The HNS contains three main models which are hierarchical attention model, feature fusion model and hierarchical classification model. The hierarchical attention model extracts features from entity description based on a bi-LSTM network with hierarchical attention mechanism to break the limitation of inaccurate Chinese-pedia. To deal with the incompletion of Chinese-pedia, the feature fusion model is presented to obtain type features from multi-source such as descriptions, info-boxes, and categories. Through this model, we fuse all the features from different sources together and reduce the features to low-dimensional and dense vectors. Finally, the hierarchical classification model is designed to infer fine-grained types for entities in Chinese-pedia with features obtained from the other two models. The experiments illustrate that HNS outperforms the start-of-art work by 15.6% on f1-score. © Springer International Publishing AG, part of Springer Nature 2018.