Improving deep classification by centroid-based candidate selection strategy
Li He, Junwu Tan, Yan Jia, Weihong Han, Shuang Tan · 2011
According to the Large Scale Hierarchical Classification Workshop of the ECIR 2010 [6], we know that the performance of classification for the large scale hierarchy, such as Open Directory Project (ODP), is still lower. Obviously, we need to improve the performance of hierarchical classification urgently. While a deep classification method [1] was proposed to make the problem tractable, the candidate selection method might be a bottleneck in the deep classification. In this paper, we used a centroid-based classification algorithm as the candidate selection strategy to enhance the deep classification. We conducted experiments with all the Chinese categories on the Open Directory Project. The experimental results show that the proposed method improved the performance of classification through selecting candidate accurately.