Zero-shot Event Detection using a Textual Entailment Model as an Enhanced Annotator

Source code for LREC-COLING 2024 paper titled "Zero-shot Event Detection using a Textual Entailment Model as an Enhanced Annotator".

Figure 1: The illustration of the difference between a textual entailment model as a frozen annotator and an enhanced annotator.

Figure 2: The general workflow of using a pre-trained TE model and keyword expansion to annotate unlabeled data.

• We propose to use a TE(Textual Entailment) model to annotate large-scale unlabeled text and use annotated data to finetune the TE model, yielding an improved TE model.
• We propose to use keywords to filter out sentences with a low probability of expressing event(s).
• We expand limited number of seed keywords using WordNet, so that we can use the TE model to annotate unlabeled text efficiently.
• The experimental results show that our method can outperform other baselines by 15% on the ACE05 dataset.

Requirements

Install the necessary packages with:

$ pip install -r requirements.txt

Data

Related data resources can be downloaded from Google Drive.

Usage

• To finetune the TE model

  ​ cd ZS_TE/train_te_model

  ​ You may modify the arguments in fine_tune.py

  ​ python fine_tune.py

• To do the event detection task

  ​ cd ZS_TE/predict_event

  ​ You may modify the arguments in pred(smoothing).py

  ​ python pred(smoothing).py

• To do the trigger classification task

  ​ cd ZS_TE/bert_classification

  ​ You may modify the arguments in bert_c.py and dataset.py

  ​ python bert_c.py