Paper: Target Inference in Conclusion Generation

This is the code for the paper Target Inference in Conclusion Generation.

Milad Alshomary, Shahbaz Syed, Martin Potthast and Henning Wachstmuth

  @InProceedings{alshomary:2020,
    author =              {Milad Alshomary, Shahbaz Syed, Martin Potthast and Henning Wachstmuth},
    booktitle =           {The 58th annual meeting of the Association for Computational Linguistics (ACL) },
    month =               jul,
    publisher =           {ACL},
    site =                {Seattle, USA},
    title =               {{Target Inference in Conclusion Generation}},
    year =                2020
  }

---

Preprocessing

All scripts for preprocessing the data are in the preprocessing folder.

Target Identification

To tag targets in premises and conclusions, we train a sequence tagger on the IBM dataset. The code is in target_identification/claim_target_tagger.py. A trained model ready to be used is under target_identification/models/target_tagger_model.pt. The preprocessed IBM dataset that the model was trained over is under target_identification/data/ibm_ds

Target Inference

Preprocessed and tagged corpora is under target_inference/data along with the knowledge base of targets used in our approach.

Ranking approach

The code for training ranking models is under target_ranking/ranking_targets.py. The trained models are under target_inference/models

Target embedding learnin approach

• Code for training the triplet neural network is under target_inference/siamese-triplet
• targets_inference_experiment.py contains all experiments performed for target inference.