Abstract: In the report， the relational classification model was used to formalize the correlation between labels and the effects of features on label correlation into a fractional model. A multi-label classification model based on tensor network was established by requiring the model to be able to distinguish the scores of real data and noise data. The results on multiple data sets showed that compared with traditional multi-label learning methods and existing multi-label learning methods to investigate label correlation， the method has nearly doubled the average accuracy and error rate of multi-label evaluation indicators， and has a lower computational cost.