Creating custom dataset from scratch

[appledora]

I was wondering if you download the training data, in original sciERC format (as shown here) and then reformat again automatically internally before training the model? Asking this, because a little confused whether to format my custom dataset like sciERC or your input data format, as shown on the Readme.md . Also, sciERC annotates sentence wise , as per the aforementioned link. How does pure handle multisentence passages? The pretrained models that are downloaded by the repo-provided Readme.md link, are also labeled as single sentence model.

[a3616001]

Hi, we downloaded the processed SciERC (link), which is in the same format as shown on the README.

In our experiments, we do not consider cross-sentence relations. All the datasets (ACE04, ACE05, SciERC) are annotated in the sentence level. So we consider each sentence individually if there are multiple in a passage. For cross-sentence relations, you can concatenate sentences and apply our approach on concatenated inputs to predict cross-sentence relations.

[appledora]

Sort of a complementary question, the processed dataset has an additional clusters node for each document. What does it refer to or is it important?

[serenalotreck]

I think those are coreference clusters, as the model that SciERC was created for also incorporates coreference resolution. I believe this pipeline doesn't do that, so they probably aren't used?

[a3616001]

The clusters field contains the coreference annotations. Our approach doesn't use those annotations, so you may ignore this field when using our code.