Input Data for Task 2

SemEval23 Task 5: Clickbait Spoiling
1

I have a clarifying question:

Does the input data for Task 2 contain the “tags” field or not? E.g. if I want to use the spoiler type (phrase, passage, multi) in my solution for task 2 do I need to implement a solution to task 1?

It would make sense if the tags are provided for task 2 to separate the two tasks but the task website reads This field is only available in the training and validation dataset (not during test) which suggests otherwise.

Thanks in advance!
Simon

2

Dear Simon,

Sorry for the ambiguity.
The field tags is only missing for task 1. For task 2, it is available and can be used.

I have updated the documentation accordingly.

Best Regards,

Maik

1 Like
3

Dear Maik,

can you check whether the “tags” field is available in the “task-2-spoiler-generation-validation” dataset?

The data I load from the input.jsonl seems to not contain this feature.

Data before processing:
DatasetDict({
    validation: Dataset({
        features: ['uuid', 'postId', 'postText', 'postPlatform', 'targetParagraphs', 'targetTitle', 'targetDescription', 'targetKeywords', 'targetMedia', 'targetUrl'],
        num_rows: 800
    })
})
1 Like
4

Dear Simon,

Thank you for notifying us!
This was indeed a mistake in the validation dataset, that this field was missing.
I have now fixed this.

The tags field is now available in the task-2-spoiler-generation-validation dataset (it was already correct for the task-2-spoiler-generation-test).

Can you please test again?
It should work now.

Thanks and best regards,

Maik

5

Cool, I saw that your submission was now successful.
Nice, thanks for trying again!

Best regards,

Maik