/bridge_latent_knowledge_selection_gap_for_conversation

Codes for EMNLP2020 Paper "Bridging the Gap between Prior and Posterior Knowledge Selection for Knowledge-Grounded Dialogue Generation"

bridge_latent_knowledge_selection_gap_for_conversation

Codes for EMNLP2020 Paper "Bridging the Gap between Prior and Posterior Knowledge Selection for Knowledge-Grounded Dialogue Generation"

Based on Sequential Knowledge Transformer (SKT)

The code and dataset can be found in \url{https://github.com/bckim92/sequential-knowledge-transformer}

TODO:

The author is lazy. Please concat chenxiuyi2017@ia.ac.cn

System Requirements

  • see the env.sh

Running Experiments

  • for SKT, SKT+PIPM and the variants of PIPM in the paper
    • see train.sh
  • for models with KDBTS
    • see train.teacher.sh for the first stage
    • see train.student.sh for the second stage

pretrained checkpoints

  • for results in table 2 in the paper
    • see inference.sh