Hugging Face's logo

Datasets:
pavelshpagin
/
SupportBench

Tasks:
Text Classification
Question Answering
Languages:
English
Spanish
Ukrainian
Size:
10K<n<100K
Tags:
technical-support
case-extraction
multilingual
chat
telegram
benchmark
License:
Dataset card Files
xet
 Community
1
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:    TypeError
Message:      Couldn't cast array of type
struct<name: string, pretty_name: string, version: string, benchmark: string, source: string, lang: string, domain: string, description: string, stats: struct<total_messages: int64, with_text: int64, with_replies: int64, reply_rate: double, with_media: int64, with_media_files: int64, unique_senders: int64, qa_exchanges: int64, total_reactions: int64, top_reactions: struct<👍: int64, ❤: int64, 🎄: int64, 😁: int64, 🤣: int64, 🔥: int64, 🙏: int64, 👏: int64, 👎: int64, 👌: int64>, media_types: struct<photo: int64, video: int64, image: int64, webpage: int64, document: int64, archive: int64, poll: int64>, first_ts: int64, last_ts: int64>, docs_urls: list<item: string>>
to
{'name': Value('string'), 'version': Value('string'), 'benchmark': Value('string'), 'pretty_name': Value('string'), 'source': Value('string'), 'lang': Value('string'), 'domain': Value('string'), 'description': Value('string'), 'stats': {'total_messages': Value('int64'), 'with_text': Value('int64'), 'with_replies': Value('int64'), 'reply_rate': Value('float64'), 'with_media': Value('int64'), 'unique_senders': Value('int64'), 'first_ts': Value('int64'), 'last_ts': Value('int64')}}
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                         ^^^^^^^^^
                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 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2690, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2227, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 2251, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 494, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/iterable_dataset.py", line 384, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 265, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/packaged_modules/json/json.py", line 120, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2272, in table_cast
                  return cast_table_to_schema(table, schema)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2224, in cast_table_to_schema
                  cast_array_to_feature(
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 1795, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.12/site-packages/datasets/table.py", line 2092, in cast_array_to_feature
                  raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
              TypeError: Couldn't cast array of type
              struct<name: string, pretty_name: string, version: string, benchmark: string, source: string, lang: string, domain: string, description: string, stats: struct<total_messages: int64, with_text: int64, with_replies: int64, reply_rate: double, with_media: int64, with_media_files: int64, unique_senders: int64, qa_exchanges: int64, total_reactions: int64, top_reactions: struct<👍: int64, ❤: int64, 🎄: int64, 😁: int64, 🤣: int64, 🔥: int64, 🙏: int64, 👏: int64, 👎: int64, 👌: int64>, media_types: struct<photo: int64, video: int64, image: int64, webpage: int64, document: int64, archive: int64, poll: int64>, first_ts: int64, last_ts: int64>, docs_urls: list<item: string>>
              to
              {'name': Value('string'), 'version': Value('string'), 'benchmark': Value('string'), 'pretty_name': Value('string'), 'source': Value('string'), 'lang': Value('string'), 'domain': Value('string'), 'description': Value('string'), 'stats': {'total_messages': Value('int64'), 'with_text': Value('int64'), 'with_replies': Value('int64'), 'reply_rate': Value('float64'), 'with_media': Value('int64'), 'unique_senders': Value('int64'), 'first_ts': Value('int64'), 'last_ts': Value('int64')}}

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.

SupportBench

A multilingual benchmark for evaluating case extraction from real-world tech support group chats.

SupportBench contains 60,000 messages across 6 datasets in 3 languages (English, Spanish, Ukrainian), spanning 6 technical domains. All messages are sourced from public Telegram support groups.

Datasets

Dataset Language Domain Messages Users Reply% Media
Ardupilot-UA Ukrainian UAV / Drones 10,000 319 51.8% 1,440
MikroTik-UA Ukrainian Networking 10,000 205 55.4% 884
Domotica-ES Spanish Smart Home / HA 10,000 530 58.3% 736
NASeros-ES Spanish NAS / Networking 10,000 761 46.5% 495
Tasmota-EN English IoT Firmware 10,000 1,237 31.5% 903
AdGuard-EN English Ad-blocking / VPN / DNS 10,000 954 46.2% 1,049

Structure 

SupportBench/
├── ua_ardupilot.json     # Ukrainian drone/UAV support
├── mikrotik_ua.json      # Ukrainian networking support
├── domotica_es.json      # Spanish smart home support
├── naseros.json          # Spanish NAS/networking support
├── tasmota.json          # English IoT firmware support
├── adguard_en.json       # English ad-blocking/VPN/DNS support
├── manifest.json         # Benchmark metadata and stats
└── README.md

Message Format

Each JSON file contains {"meta": {...}, "messages": [...]}. Each message:

{
  "id": "tg_tasmota_12345",
  "group_id": "tasmota",
  "ts": 1700000000000,
  "sender": "user_a1b2c3d4e5",
  "body": "My Sonoff Basic won't flash via serial...",
  "reply_to_id": "tg_tasmota_12340",
  "grouped_id": null,
  "media_type": "photo",
  "media_path": null,
  "webpage_url": null,
  "reactions": null,
  "views": 42,
  "forwards": 0
}

Fields

Field Type Description
id string Unique message ID (tg_{group}_{telegram_id})
group_id string Dataset/group name
ts int Unix timestamp in milliseconds
sender string Anonymized sender (user_{sha256[:10]})
body string Message text with Unicode emoji preserved
reply_to_id string|null ID of parent message
grouped_id int|null Album group ID for multi-media posts
media_type string|null photo, video, image, document, pdf, archive, audio, webpage, poll
media_path string|null Relative path to media file (only in ua_ardupilot)
webpage_url string|null URL from link preview
reactions object|null Emoji reaction counts (only in ua_ardupilot)
views int|null View count
forwards int|null Forward count

Notes

  • All sender IDs are irreversibly anonymized via SHA-256 hashing
  • ua_ardupilot has richer metadata (reactions, media paths) from a separate export pipeline
  • AdGuard EN is a topics-based supergroup; reply_to_id often points to the topic root rather than the actual parent message (46.2% resolved within 10K window)
  • Tasmota spans 3.5 years of IoT firmware support history (1,274 days)

Intended Use

  • Case extraction: identify problem-solution pairs from unstructured chat streams
  • Thread reconstruction: reconstruct conversation threads from reply chains
  • Cross-lingual transfer: evaluate whether case-mining generalizes across languages
  • Q&A retrieval: mine question-answer pairs from technical discussions

Citation 

@inproceedings{supportbench2026,
  title={SupportBot: Continuous Case Mining for Grounded Technical Support},
  author={Shpagin, Pavel},
  booktitle={Proceedings of EMNLP 2026},
  year={2026}
}

License

CC BY 4.0. All messages are from public Telegram groups. Sender identities are irreversibly anonymized.
Downloads last month
28
Total file size:
53.7 MB