Skip to content

Latest commit

 

History

History
80 lines (60 loc) · 3.6 KB

CONTRIBUTING.md

File metadata and controls

80 lines (60 loc) · 3.6 KB
Raw

Bot structure

Telegram (server) sends a stream of events to bot. Library telegrambot4s splits it into several streams: messages, edited messages, commands, callbacks and so on. As a result bot needs a separate process handler for each stream. Censoring service is responsible for describing processes. There are two processes at now: msgProcessing and callbackProcessing.

BotService is responsible for telegram bot API.

Storage services.

  • The first was StatsService. It keeps statistics about messages and infringements.
  • ChatSettingsService is a new and just a mock. It was born when a need of separate options per chat became obvious.

Description of msgProcessing

Process is applied to 'messages' and 'edited messages' streams only for groups and super groups.

val process =
      for {
        _       <- statCounter
        es      <- inquiry.searchEvidences
        verdict <- Judgement.judge(es)
        _       <- execution.punish(verdict)
        _       <- execution.notifyCourt(verdict)
      } yield verdict
  1. Updates statistics.
  2. Inspects event. Detects evidences.
  3. Makes decision - Is a message suspicious?
  4. Applies automatic punishment.
  5. Notifies administrators and asks their will.

Stage 4 and stage 5 don't take in account a kind of guilt and restrict number of use cases. It requires to rethink process.

2. Search evidences

Looks for telegram links (via @) and web links (via http:\\). Previous implementation searched all web links and utilized white list. False positive rate was too high. Idea for a future implementation is a combination of black and white lists both for web and telegram links.

To add a new detector implement InquiryProcedure trait and append to list of procedures in Inquiry class.

3. Judgement

Gathered on stage 2 facts are applied to list of rules. The result of application is user innocent or crime supposed.

At now a mere rule is a detection of evidences in messages of newcomers.

To add a new rule implement Rule trait and append to Rule.codex list.

4. Punishment

Not implemented. Too risky. (small) Chance of false positive result exists but a way to rollback or notify administrators doesn't.

Can be utilized for other cases. For example to stop a sticker spam.

5. Notify administration

Bot replies on suspicious message and mentions all of administrators. If bot is administrator itself then it also displays buttons for fast execution.

Description of callbackProcessing

ToDo

Storage description

Bot needs statistics to make decisions about guilts. Statistics contains general counter of messages, counter of infringements and stats per chat. As storage ChronicleMap is used - extremely simple db with portable files. storage schema

StatsService class is responsible for datum writing and reading. Serialization and deserialization are implemented via protobuf 2.

At now bot uses Chronicle-Map as storage engine. It's a simple and nice embedded KV.

Run

Add 'application.conf' file to classpath with one parameter: bot-token="token_of_your_bot". You can create and manage your bots via Bot Father

Code style

Code style is absent right now. Merely scalafmt default settings because it's better than nothing.

Additional notes

Bot'd been created as simple pet-project. Some technologies had been integrated as experiments. It'll be nice to remove or replace some of them, add unification for similar parts, give suitable names. Performance doesn't matter when popularity is low.