Datasets:

lukeslp
/

steam-co-review-network

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
title: string
id: string
subtitle: string
description: string
isPrivate: bool
keywords: list<item: string>
  child 0, item: string
licenses: list<item: struct<name: string>>
  child 0, item: struct<name: string>
      child 0, name: string
unavailable_full_graph: struct<github_pointer_path: string, pointer_bytes: int64, expected_bytes: int64, expected_sha256: st (... 91 chars omitted)
  child 0, github_pointer_path: string
  child 1, pointer_bytes: int64
  child 2, expected_bytes: int64
  child 3, expected_sha256: string
  child 4, status: string
  child 5, historically_claimed_nodes: int64
  child 6, historically_claimed_links: int64
available_files: list<item: struct<github_path: string, huggingface_path: string, bytes: int64, sha256: string, nodes (... 121 chars omitted)
  child 0, item: struct<github_path: string, huggingface_path: string, bytes: int64, sha256: string, nodes: int64, li (... 109 chars omitted)
      child 0, github_path: string
      child 1, huggingface_path: string
      child 2, bytes: int64
      child 3, sha256: string
      child 4, nodes: int64
      child 5, links: int64
      child 6, min_weight: int64
      child 7, max_weight: int64
      child 8, records: int64
      child 9, fields_per_record: int64
      child 10, positions: int64
sources: struct<metadata: string, reviews: string>
  child 0, metadata: string
  child 1, reviews: string
historical_input_revision_hashes_recorded: bool
verified_on: timestamp[s]
to
{'verified_on': Value('timestamp[s]'), 'available_files': List({'github_path': Value('string'), 'huggingface_path': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'nodes': Value('int64'), 'links': Value('int64'), 'min_weight': Value('int64'), 'max_weight': Value('int64'), 'records': Value('int64'), 'fields_per_record': Value('int64'), 'positions': Value('int64')}), 'unavailable_full_graph': {'github_pointer_path': Value('string'), 'pointer_bytes': Value('int64'), 'expected_bytes': Value('int64'), 'expected_sha256': Value('string'), 'status': Value('string'), 'historically_claimed_nodes': Value('int64'), 'historically_claimed_links': Value('int64')}, 'sources': {'metadata': Value('string'), 'reviews': Value('string')}, 'historical_input_revision_hashes_recorded': Value('bool')}
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 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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
              title: string
              id: string
              subtitle: string
              description: string
              isPrivate: bool
              keywords: list<item: string>
                child 0, item: string
              licenses: list<item: struct<name: string>>
                child 0, item: struct<name: string>
                    child 0, name: string
              unavailable_full_graph: struct<github_pointer_path: string, pointer_bytes: int64, expected_bytes: int64, expected_sha256: st (... 91 chars omitted)
                child 0, github_pointer_path: string
                child 1, pointer_bytes: int64
                child 2, expected_bytes: int64
                child 3, expected_sha256: string
                child 4, status: string
                child 5, historically_claimed_nodes: int64
                child 6, historically_claimed_links: int64
              available_files: list<item: struct<github_path: string, huggingface_path: string, bytes: int64, sha256: string, nodes (... 121 chars omitted)
                child 0, item: struct<github_path: string, huggingface_path: string, bytes: int64, sha256: string, nodes: int64, li (... 109 chars omitted)
                    child 0, github_path: string
                    child 1, huggingface_path: string
                    child 2, bytes: int64
                    child 3, sha256: string
                    child 4, nodes: int64
                    child 5, links: int64
                    child 6, min_weight: int64
                    child 7, max_weight: int64
                    child 8, records: int64
                    child 9, fields_per_record: int64
                    child 10, positions: int64
              sources: struct<metadata: string, reviews: string>
                child 0, metadata: string
                child 1, reviews: string
              historical_input_revision_hashes_recorded: bool
              verified_on: timestamp[s]
              to
              {'verified_on': Value('timestamp[s]'), 'available_files': List({'github_path': Value('string'), 'huggingface_path': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'nodes': Value('int64'), 'links': Value('int64'), 'min_weight': Value('int64'), 'max_weight': Value('int64'), 'records': Value('int64'), 'fields_per_record': Value('int64'), 'positions': Value('int64')}), 'unavailable_full_graph': {'github_pointer_path': Value('string'), 'pointer_bytes': Value('int64'), 'expected_bytes': Value('int64'), 'expected_sha256': Value('string'), 'status': Value('string'), 'historically_claimed_nodes': Value('int64'), 'historically_claimed_links': Value('int64')}, 'sources': {'metadata': Value('string'), 'reviews': Value('string')}, 'historical_input_revision_hashes_recorded': Value('bool')}
              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.

Steam Co-Review Network

A filtered graph of 25,996 Steam games and 861,232 weighted links, supplied alongside a separate 82,928-game catalog. Links describe retained shared reviewer counts between games. The catalog is larger than the graph; its games are not all graph nodes.

Available files

Counts and SHA-256 checksums were verified on September 14, 2026. The data bytes are unchanged by this documentation correction.

Artifact	GitHub filename	Hugging Face filename	Contents	Bytes
Filtered graph	`steam_network.json`	`steam_network_full.json`	25,996 nodes; 861,232 links	42,249,061
Game catalog	`steam_all_2005.json`	`steam_all_2005.json`	82,928 packed game records, 2005–2025	6,482,810
Stored layout	`steam_force_layout.json`	`steam_force_layout.json`	8,905 positions keyed by game ID	257,523

The Hugging Face filename steam_network_full.json is a historical name for the same filtered graph available on GitHub as steam_network.json. It does not contain the previously advertised 48,362-node / 33,041,298-link graph.

The graph's SHA-256 is:

512eafbf312743aa7768a9c1c0fc17b49089bb85640c9bfe8e5265e7d8d9dd2d

SHA256SUMS covers the available data files using this platform's filenames. DATA_PROVENANCE.json records the cross-platform names, checksums, measured counts, and the identity of the unavailable historical object.

Unavailable historical full graph

GitHub's steam_network_full.json is only a 135-byte Git LFS pointer, not readable JSON. The referenced 1,423,482,375-byte object could not be recovered from the inspected copies. Its SHA-256 is f73197543ac056b56ffdb4ec73bdc94684721b47c46072d42b0cc9479f2d502d. The pointer and Git history are retained as provenance. The old 48,362-node / 33,041,298-link figures are historical claims about that missing object, not measurements of the current download. Do not substitute the smaller graph and label it as a recovered full graph.

Read the graph

After downloading the graph, use Python's standard library. Set the filename to steam_network_full.json for the Hugging Face download only:

import hashlib
import json
from pathlib import Path

path = Path("steam_network.json")  # GitHub; see the Hugging Face alias above
raw = path.read_bytes()
assert hashlib.sha256(raw).hexdigest() == "512eafbf312743aa7768a9c1c0fc17b49089bb85640c9bfe8e5265e7d8d9dd2d"
graph = json.loads(raw)
assert len(graph["nodes"]) == 25996
assert len(graph["links"]) == 861232

Each node contains a Steam app ID (id), title, release year, rating label, positive-review ratio, review count, price, genre/tag indices, and developer. Graph genre/tag indices refer to meta.genres and meta.tags.

Each link has integer source and target indices into nodes, plus an integer weight. Verified weights range from 5 to 694,377. Node IDs and undirected pairs are unique, and every endpoint is in range in the distributed graph.

Filtering and interpretation

The graph's stored metadata records top_k=50, min_shared=5, max_user_games=75, and min_game_reviews=50. The corresponding v2 builder retains the union of each game's top-50 neighbor selections. This is not a maximum-degree guarantee: a game may also be selected by many other games.

The historical builders cap the games retained per reviewer and prune partial user/edge accumulators to limit memory. Those operations discard information, so edge weights represent retained co-review counts rather than guaranteed complete audience overlap. The graph is a filtered analytical snapshot, not a complete Steam social graph, ownership network, or measure of all players.

Catalog and layout

Catalog rows use this nine-field format:

[name, year, ratio, reviews, price, ratingIdx, genreIdxs, tagIdxs, developer]

The catalog's ratings, genres, and tags lookup tables are top-level arrays. Packed catalog rows do not include Steam app IDs; a direct ID join with graph nodes is unavailable. Rating labels are derived Steam-style categories, not an independently verified copy of every game's current store rating.

The stored layout contains 8,905 positions. Its metadata records an edge-weight threshold of 100 and 80 iterations. Current compute_layout.py uses different defaults and outputs additional community information absent from this file; it is not evidence of an exact rebuild of the stored layout. Match positions to graph nodes by Steam app ID, not by array index.

Sources and reproduction limits

Game metadata: FronkonGames Steam Games Dataset. The packaged catalog identifies a January 2026 input snapshot.
Review source: artermiloff Steam Games Reviews 2024. Historical project notes describe an upstream corpus of 128 million reviews across roughly 80,000 games, covering 2012–June 2024. Those are source-corpus figures, not counts measured from this filtered graph or proof that every upstream review contributed to it.

Exact historical input revision hashes are not recorded in this repository. The graph and catalog are derived outputs; downloading a newer upstream snapshot would create a new version, not recover the missing historical object.

The Python files preserve the processing approach. They require external raw review CSVs and enriched metadata at the paths configured in each script:

build_network_v2.py: filtered co-review construction.
enrich_data.py: nine-field catalog construction and graph enrichment.
build_network_full.py: a less-filtered historical build; its missing output has not been reconstructed and its processing still includes caps/pruning.
compute_layout.py: layout and optional graph trimming; review its input and output paths before running because it can overwrite a network file.
build_all_games.py: legacy ten-field catalog writer, incompatible with the current packed catalog and notebook; do not use it to replace this catalog.

The raw review corpus is not included. No new graph was generated for this correction. Schema and checksum checks establish the distributed artifact's identity and internal consistency, not the accuracy of every upstream record.

Distribution

License and author

CC-BY-4.0. Retain source attribution and consult each upstream source's terms when reusing its original material.

Luke Steuber — lukesteuber.com

Downloads last month: 126

Total file size:

49 MB