Skip to content

Latest commit

 

History

History
57 lines (28 loc) · 2.41 KB

README.md

File metadata and controls

57 lines (28 loc) · 2.41 KB
Raw

Rewritten utterances

This directory collects the files with utterances rewritten automatically by the approaches presented in [Mele2021].

For the rewriting, we used the labels predicted with the BERT classification at Step 1 and BERT context classification at Step 2 (files *_S1_BERT_S2_BERT_Context.tsv) as well as the ground-truth labels (files *_GT.tsv).

Methods and corresponding files

Standard method

If u_i is classified as FT, it is enriched with context extracted from the first utterance. If u_i is classified as PT, it is enriched with context from the previous utterance.

  • classif_results_strategy_Standard_S1_BERT_S2_BERT_Context.tsv: using the predicted labels.

  • classif_results_strategy_Standard_GT.tsv: using the ground-truth labels.

Enriched method

Similar to Standard, but if u_i is classified as PT, it is enriched with context from the previous enriched utterance.

  • classif_results_strategy_Enriched_S1_BERT_S2_BERT_Context.tsv: using the predicted labels.

  • classif_results_strategy_Enriched_GT.tsv: using the ground-truth labels.

Last SE method

Always propagating the context extracted from the last seen SE utterance.

  • classif_results_strategy_LastSE_S1_BERT_S2_BERT_Context.tsv: using the predicted labels.

  • classif_results_strategy_LastSE_GT.tsv: using the ground-truth labels.

First and last SE method

Similar to Last SE, but the context is extracted from both the first utterance and the last seen SE utterance.

  • classif_results_strategy_FirstAndLastSE_S1_BERT_S2_BERT_Context.tsv: using the predicted labels.

  • classif_results_strategy_FirstAndLastSE_GT.tsv: using the ground-truth labels.

First or last SE method

Similar to Standard. If u_i is classified as FT, it is enriched with the context extracted from the first utterance. If u_i is classified as PT, then the context is extracted from the last seen SE instead of the previous utterance.

  • classif_results_strategy_FirstOrLastSE_S1_BERT_S2_BERT_Context.tsv: using the predicted labels.

  • classif_results_strategy_FirstOrLastSE_GT.tsv: using the ground-truth labels.

Format

utteranceId \t utterance

References

[Mele2021] Ida Mele, Cristina Ioana Muntean, Franco Maria Nardini, Raffaele Perego, Nicola Tonellotto, Ophir Frieder, Adaptive utterance rewriting for conversational search, IProcessing & Management, Volume 58, Issue 6, 2021, https://doi.org/10.1016/j.ipm.2021.102682.