Closed eveliao closed 6 years ago
Yeah, for the Seq2Seq model we need to pre-define the order of the output labels. For the label order problem, you can refer to our latest work: https://arxiv.org/pdf/1809.03118.pdf
@ypengc7512 Hi, would you please open the source code of the latest work (seq2set), as you mentioned above. Many thanks.
In the paper, you mentioned that "In addition, the proposed models are trained using the maximum likelihood estimation method and the cross-entropy loss function, which requires humans to predefine the order of the output labels. " but this doesn't explain for me... It means that order matters because we use seq2seq model(instead of the mentioned 'MLE and cross-entropy'), which is the intrinsic property of model. The prediction order should follow the true order. Thus human-defined order has impact. Is the reason that high-frequency labels are more likely the root label in a hierarchical category taxonomy and thus more likely to have correlations with other labels?