Neural Tensor Network for Multi- Label Classification

Wenxing Hong, Wenjing Xu, Jianwei Qi, Yang Weng

Abstract: The difference of multi-label classification from traditional classification is that an instance may associate a set of labels simultaneously. In recent study, some scholars have proposed that the information which derives from the query instance's nearest neighbors, can be useful when predicting the labels of the query instance. On the basis of their research, we propose a new approach to multi-label classification, which employs neural tensor network (NTN) to explore the relations among the labels of neighbors and classify the query instance with these correlations. This method utilizes the correlations and interdependencies between labels and leverages the potential of data. Experiments on real data show that our method can achieve good performance in multi-label classification.