- Total number of reviews: 115
- Each review has been annotated two times by different annotators.
- Total number of annotations: 164
{article_review_name}.txt - file with collected text from all files from the review
{article_review_name}_{reviewer}.tsv - file with annotation of the review
Stores all sentences from the review. If sentence is not a review, then all columns will be None except ann and text
| Column name | Description |
|---|---|
| side | side of the argument. |
| opponent | opponent of the argument. |
| round | Number of round |
| number | Number of argument in the round |
| attacks | Number of attacking argument from the previous round |
| ann | Type of argument (0-not an argument, 1-author, 2-reviewer) |
| text | Text of argument\sentence |
You can use the dataset already split into train\val\test subsets. (./dataset/sentence/)
OR you can prepare a dataset by yourself using a script ./src/dataset/prepare.py
You can use save_annotated_text_html func from ./src/utils.py
Example of result represented in ./assets/admsci5030125_boyarkin.html:
Krippendorff's alpha for the dataset is 0.81±0.19 [link]
Available models:
You can train models using ./src/model_training.py script. To choose a model you need to uncomment line with a desired model.