Hugging Face's logo

Datasets:
Aureliolo
/
game-review-claims

Tasks:
Text Classification
Languages:
multilingual
Tags:
steam
game-reviews
aspect-based-sentiment-analysis
DOI:
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
ambiguous: bool
app_id: int64
confidence: string
end: int64
index: int64
ironic: bool
language: string
polarity: string
produced_by: string
review_id: string
split_wrong: bool
start: int64
subject: string
subset: string
taxonomy: string
also: list<item: struct<polarity: string, subject: string>>
  child 0, item: struct<polarity: string, subject: string>
      child 0, polarity: string
      child 1, subject: string
acceptable: bool
by: string
to
{'acceptable': Value('bool'), 'app_id': Value('int64'), 'by': Value('string'), 'index': Value('int64'), 'review_id': Value('string'), 'subject': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              ambiguous: bool
              app_id: int64
              confidence: string
              end: int64
              index: int64
              ironic: bool
              language: string
              polarity: string
              produced_by: string
              review_id: string
              split_wrong: bool
              start: int64
              subject: string
              subset: string
              taxonomy: string
              also: list<item: struct<polarity: string, subject: string>>
                child 0, item: struct<polarity: string, subject: string>
                    child 0, polarity: string
                    child 1, subject: string
              acceptable: bool
              by: string
              to
              {'acceptable': Value('bool'), 'app_id': Value('int64'), 'by': Value('string'), 'index': Value('int64'), 'review_id': Value('string'), 'subject': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Game review claims

48,923 claims from 25,242 Steam reviews of 69 games, each labelled with its subject and whether it is praise, a complaint or neutral. A claim is one point a reviewer makes. Part of SteamGauge.

The dataset contains no review text and no author information. fetch_text.py fetches the labelled reviews from Steam's public review endpoint and adds each claim's text.

Files

  • claims.jsonl: one label per claim.
  • second-readings.jsonl: a second label, from a different model, for some claims.
  • gold.jsonl: 200 claims labelled by one person.
  • acceptable.jsonl: for some gold claims, whether a subject other than the gold one is also acceptable, judged by the same person.
  • reviews.jsonl: one row per labelled review.
  • fetch_text.py: adds the text to the claims. Python standard library only.

Fields

claims.jsonl, second-readings.jsonl and gold.jsonl:

Field Meaning
app_id, review_id The Steam game and review.
index The claim's position among its review's claims.
start, end The claim's UTF-8 byte offsets in the review text.
language The review's language, as Steam gives it.
subject What the claim is about.
polarity praise, complaint or neutral.
also Further subjects the claim is about, each with its own polarity.
ironic The claim means the opposite of what it says; polarity is what it means.
confidence The labeller's confidence in the subject: high, medium or low.
ambiguous The category rules do not settle which subject the claim is about.
split_wrong The claim is cut in the wrong place.
produced_by The model that wrote the label, or a person.
taxonomy The category sheet the label was written against: a hash of the sheet, or core-5 or core-6. Null where it was not recorded.
subset How the claim was drawn (below).

subset:

  • random: a random draw of the game's reviews. Only these rows estimate how often a subject comes up.
  • declined: claims a trained reader did not answer.
  • mined: claims found by keyword for rarely mentioned subjects.
  • retrieved: claims found by meaning for rarely mentioned subjects.
  • multilingual: claims from reviews in languages other than English.

acceptable.jsonl: app_id, review_id, index, subject, acceptable, by.

reviews.jsonl: app_id, review_id, language, created and updated (Unix time), voted_up, votes_up, votes_funny, weighted_vote_score, comment_count, steam_purchase, received_for_free, written_during_early_access, refunded, primarily_steam_deck, playtime_at_review_minutes, and the SHA-256 (text_sha256) and length in bytes (text_bytes) of the review text the labels were written against.

Text 

python fetch_text.py --data . --to claims-with-text.jsonl

--claims gold.jsonl or --claims second-readings.jsonl does the same for another label file. Reviews deleted or edited on Steam since labelling are left out.

Labels

Written by Claude models from a category sheet, one game at a time, without the game's name: claude-fable-5-1 (42,985), claude-opus-5-5 (4,330), claude-opus-5 (1,608).

21,449 claims have a second label from a different model, in second-readings.jsonl. On those claims the two labels agree:

Field Agreement Cohen's kappa
subject 86% 0.84
polarity 94% 0.91
ambiguous 70% 0.36

The category sheet is taxonomy.rs in the SteamGauge repository.

Licence

CC BY 4.0. Review text is not included.
Downloads last month
37
Total file size:
35.2 MB

Models trained or fine-tuned on Aureliolo/game-review-claims