juliowaissman / MeOffendEs

MeOffendEs

This repository contains the data for the MeOffendEs@IberLEF task. All zip files are password-protected. Access to data will be granted to participants of MeOffendEs workshop at IberLEF 2021.

The folder scorers contains the programs for evaluating each dataset as used by Codalab competition.

OffendEs subset corpus

• Development data (trial data): File offendes-devdata.zip (100 samples)

  • devdata1.tsv --> comment_id, comment
  • devdata2.tsv --> comment_id, comment, label, influencer, influencer_gender, media
  • devgold.tsv --> comment_id, label, confidence_list

  confidence_list is a list with four probability values (the sum up 1.0), one per each possible label in this order: [NO_prob, NOM_prob, OFP_prob, OFG_prob]

• Train data: File offendes-traindata.zip (16,710 samples)

  • traindata1.tsv --> comment_id, comment, label
  • traindata2.tsv --> comment_id, comment, label, influencer, influencer_gender, media

• Test data: File offendes-testdata.zip (13,606 samples)

  • testdata1.tsv --> comment_id, comment
  • testdata2.tsv --> comment_id, comment, influencer, influencer_gender, media

OffendMex subset corpus

• Development data (trial data): File mx-devdata.zip (76 samples)

  • mx-trial-data.csv --> each row in this file is associated to a sample in the dataset, the first row (headers) should be discarded
  • mx-trial-names-csv --> Description of the information (columns) in mx-trial-data.csv
  • mx-trial-vulgarity-label.csv --> Vulgarity label for each sample in mx-trial-data.csv, this information will be only available for trial and training data
  • mx-trial-outputs.sol --> ground truth labels for samples in mx-trial-data.csv

• Training data: File mx-traindata.zip (76 samples)

  • mx-train-data-non-contextual.csv --> tweets without contextual information, each row in this file is associated to a sample in the dataset, the first row (headers) should be discarded
  • mx-train-data-contextual.csv --> tweets including contextual information, each row in this file is associated to a sample in the dataset, the first row (headers) should be discarded
  • mx-train-names-csv --> Description of the information (columns) in mx-train-data-non-contextual.csv
  • mx-train-vulgarity-label.csv --> Vulgarity label for each sample in the data files mx-train-data-contextual.csv and mx-train-data-non-contextual.csv, this information will be only available for trial and training data
  • mx-train-outputs.sol --> ground truth labels for samples in mx-train-data-contextual.csv and mx-train-data-non-contextual.csv