metadata

annotations_creators:
  - crowdsourced
  - expert-generated
language_creators:
  - expert-generated
languages:
  - en-US
licenses:
  - mitmultilinguality: null
  - monolingual
pretty_name: ''
size_categories:
  - 10K<n<100K
source_datasets:
  - original
task_categories:
  - text-classification
task_ids:
  - sentiment-classification

NewsMTSC dataset

NewsMTSC is a high-quality dataset consisting of more than 11k manually labeled sentences sampled from English news articles. Each sentence was labeled by at least five human coders. The dataset is published as our EACL 2021 paper NewsMTSC: (Multi-)Target-dependent Sentiment Classification in News Articles.

Dataset

Files

The dataset consists of three splits. We suggest to use the splits as follows:

train.jsonl - For training.
devtest_rw.jsonl - To evaluate a model's classification performance on a "real-world" set of sentences, i.e., the set was created with the objective to resemble real-world distribution as to sentiment and other factors mentioned in the paper.
devtest_mt.jsonl - To evaluate a model's classification performance only on sentences that contain at least two target mentions. Note that the mentions were extracted to refer to different persons but in a few cases might indeed refer to the same person.

For most projects, we suggest to use train.jsonl for training and devtest_rw.jsonl for evaluation. More information can be found in our paper.

Format

Each split is stored in a JSONL file. In JSONL, each line represents one JSON object. In our dataset, each JSON object consists of the following attributes. When using the dataset, you most likely will need (only) the attributes highlighted in bold.

sentence_normalized: a single sentence
primary_gid: an identifier that is unique within NewsMTSC
targets: one or more targets

Each target in targets consists of:

Input.gid: an identifier that is unique within NewsMTSC
from: the character-based, 0-indexed position of the first character of the target's mention within sentence_normalized
to: the last character of the target's mention
mention: the text of the mention
polarity: the sentiment of the sentence concerning the target's mention (2.0 = negative, 4.0 = neutral, 6.0 = positive)
further_mentions (optional): one or more coreferential mentions of the target within the sentence. Note that the coreferences were extracted automatically and thus might not be incorrect. Our annotators labeled sentiment concerning the main mention, which might not be identical to the sentiment of coreferences.

An example looks like this:

{
   "primary_gid":"allsides_1192_476_17_— Judge Neil M. Gorsuch_126_139",
   "sentence_normalized":"But neither side harbored any doubts, based on the judge’s opinions, other writings and the president who nominated him, that Judge Gorsuch would be a reliable conservative committed to following the original understanding of those who drafted and ratified the Constitution.",
   "targets":[
      {
         "Input.gid":"allsides_1192_476_17_— Judge Neil M. Gorsuch_126_139",
         "from":126,
         "to":139,
         "mention":"Judge Gorsuch",
         "polarity":6.0,
         "further_mentions":[
            {
               "from":116,
               "to":119,
               "mention":"him"
            }
         ]
      }
   ]
}

Contact

If you want to get in touch, feel free to contact Felix Hamborg. If you find an issue with the dataset or model or have a question concerning either, please open an issue in the repository.

Web: https://felix.hamborg.eu/
Mail: felix.hamborg@uni-konstanz.de
Repository: https://github.com/fhamborg/NewsMTSC

How to cite

If you use the dataset or parts of it, please cite our paper:

@InProceedings{Hamborg2021b,
  author    = {Hamborg, Felix and Donnay, Karsten},
  title     = {NewsMTSC: (Multi-)Target-dependent Sentiment Classification in News Articles},
  booktitle = {Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2021)},
  year      = {2021},
  month     = {Apr.},
  location  = {Virtual Event},
}