Textual Classification on Multiple Domains by Weighted Fusion of Multiple Models

Text classification is an important task in the natural language processing while the increased number of samples from multiple domains may reduce the accuracy of the classification. In this paper, we proposed a novel method that can classify samples from multiple domains with the weighted fusion of the multiple models. For each sample, our method firstly predicts which domain it belongs to. Then, we apply a weighted fusion to the corresponding models that are trained on this domain to predict the label of this sample. The experiments on multiple domains proved that our method achieved the best performance compared with the other methods.