Code for NAACL 2018 paper "Multi-task Learning of Pairwise Sequence Classification Tasks Over Disparate Label Spaces" by Isabelle Augenstein, Sebastian Ruder, Anders Søgaard
Switch branches/tags
Nothing to show
Clone or download
isabelleaugenstein - Renamed models the way we named them in the paper
- More documentation to clarify which functions correspond to which sections of the paper
- Removed spurious code
Latest commit fb13002 Apr 4, 2018

README.md

mtl-disparate

Code for NAACL 2018 paper "Multi-task Learning of Pairwise Sequence Classification Tasks Over Disparate Label Spaces" by Isabelle Augenstein, Sebastian Ruder, Anders Søgaard

Note that this is research code and will not be maintained to e.g. ensure compatibility with more recent library versions.

Requirements:

  • Tensorflow 1.5
  • Numpy 1.12.1
  • sklearn 0.18.1
  • scipy

Steps to run:

  • run data/download_data.sh to download and extract data
  • preproc/data_reader.py tests if all the data readers work
  • preproc/fnc_data_splits.py to split the FNC training dataset into a training and dev set
  • main.py trains models

Datasets

SemEval 2016 Task 6 Stance detection

Fake News Challenge (FNC)

Multi-NLI

SemEval 2016 Task 4 Subtask B Topic-based Twitter sentiment analysis

SemEval 2016 Task 5 Subtask 1 Slot 3 Aspect-based sentiment analysis

Clickbait Challenge 2017