Limit the number of allowed entities/mentions in a message

[BoberMod]

I faced a new spam type when the first message contained mass-mention of chat members and it was edited to ad-message immediately after that. Telegram keeps mention notifications even after message is edited, so users go to chat from notification and see ad message.

The classifier doesn't detect such messages as spam, because they contain a lot of random text (usernames) even if the original message contains spam too:

Note: Message without mentions is added to spam samples, and has 99% detection.

Screenshot of how it looks

Each mention in the message is counted as a separate entity of the type mention. I suggest a feature request to allow limiting the number of entities by type or specifically restricting mention entities.

I think it would be useful to block/limit any entity type because spam also contains telegram cashtags ($USD) and hashtags.

Message JSON from Telegram API

  {
   "update_id": 936949643,
   "message": {
    "message_id": 1881484,
    "from": {
     "id": 155807040,
     "is_bot": false,
     "first_name": "BoberMod",
     "username": "BoberMod",
     "language_code": "ru",
     "is_premium": true
    },
    "chat": {
     "id": 155807040,
     "first_name": "BoberMod",
     "username": "BoberMod",
     "type": "private"
    },
    "date": 1736530089,
    "text": "Ищу энергичных партнёрοв для запуска прибыльных криптοпρоектов. У вас есть желание заρабатывать пассивнο? Ηапишите мне!\n\nИщу энергичных партнёрοв для запуска прибыльных криптοпρоектов. У вас есть желание заρабатывать пассивнο? Ηапишите мне!  @vladimiir49 @AliLit6062 @j_evgenyyyy @kravtsov_dya @Charger69 @Militant_Hamster @Bearded_alex @deshik80 @vovazlv @melkosofter @Vyacheslav_Voznyy0 @andyvers @Leonid_ur5yar @Andrey_911_psg @S_bobo",
    "entities": [
     {
      "offset": 242,
      "length": 12,
      "type": "mention"
     },
     {
      "offset": 255,
      "length": 11,
      "type": "mention"
     },
     {
      "offset": 267,
      "length": 12,
      "type": "mention"
     },
     {
      "offset": 280,
      "length": 13,
      "type": "mention"
     },
     {
      "offset": 294,
      "length": 10,
      "type": "mention"
     },
     {
      "offset": 305,
      "length": 17,
      "type": "mention"
     },
     {
      "offset": 323,
      "length": 13,
      "type": "mention"
     },
     {
      "offset": 337,
      "length": 9,
      "type": "mention"
     },
     {
      "offset": 347,
      "length": 8,
      "type": "mention"
     },
     {
      "offset": 356,
      "length": 12,
      "type": "mention"
     },
     {
      "offset": 369,
      "length": 19,
      "type": "mention"
     },
     {
      "offset": 389,
      "length": 9,
      "type": "mention"
     },
     {
      "offset": 399,
      "length": 14,
      "type": "mention"
     },
     {
      "offset": 414,
      "length": 15,
      "type": "mention"
     },
     {
      "offset": 430,
      "length": 7,
      "type": "mention"
     }
    ]
   }
  }

Check for entities also allows to update/unify LinksCheck function, because each URL in the message is also an entity of url type.

tg-spam/lib/tgspam/metachecks.go

Lines 17 to 32 in 60b5c3b

	func LinksCheck(limit int) MetaCheck {
	return func(req spamcheck.Request) spamcheck.Response {
	links := req.Meta.Links
	if links == 0 {
	links = strings.Count(req.Msg, "http://") + strings.Count(req.Msg, "https://")
	}
	if links > limit {
	return spamcheck.Response{
	Name: "links",
	Spam: true,
	Details: fmt.Sprintf("too many links %d/%d", links, limit),
	}
	}
	return spamcheck.Response{Spam: false, Name: "links", Details: fmt.Sprintf("links %d/%d", links, limit)}
	}
	}

Bot API documentation: https://core.telegram.org/bots/api#messageentity

[umputun]

I like the idea of this new checker

[BoberMod]

I'll try to implement it myself and submit the PR.

[umputun]

I'll try to implement it myself and submit the PR.

Cool. I don't think we want to reimplement LinksCheck because currently it is a part of a library that does the job on any text, not just on TG meta info. It seems to work and feels like a more universal approach to me.

[umputun]

This issue has been addressed in PR #262 (now merged).

Added the ability to limit the number of mentions (@username) in a message with a new '--meta.mentions-limit' parameter. When the limit is exceeded, the message will be treated as spam.

• Default value is -1 (check disabled)
• Set to 0 to allow no mentions
• Set to a positive number to allow that many mentions
• It counts both 'mention' and 'text_mention' entity types from Telegram

The feature follows the same pattern as the existing links limit.