Factored Latent-Dynamic Conditional Random Fields for single and multi-label sequence modeling

Highlights：

• We propose a single and multi-label generalization of LDCRF (Morency et al., 2007), called the Factored LDCRF.

• FLDCRF unifies the concepts of LDCRF and Dynamic CRFs (DCRF, Sutton et al., 2007) and extends the CRF family.

• The single-label variant of FLDCRF (FLDCRF-s) outperforms state-of-the-art models, viz., CRF, LDCRF, LSTM and LSTM-CRF across 5 experiments over 2 different datasets.

• The multi-label variant of FLDCRF-m outperforms state-of-the-art single-label, viz., CRF, LDCRF, LSTM and LSTM-CRF, and multi-label, viz., Coupled CRF, Factorial CRF and multi-label LSTM models on the multi-label sequence tagging experiment.

• We compare FLDCRF and LSTM model families not only on the test data, but also across several other modeling aspects, e.g., model selection, consistency and computation times.