Testing Message Filters Offline

English Help & Support
, , , ,
1

When testing message filters in my Twitch chat while my stream is offline, I have noticed that some of the filters work, while others do not.

Are only some of the filters active when offline and the rest turn on once online?

Thanks.

2

Hello,
Can you provide me with the name of your channel?

As well as the filters that didn’t work during your off-line tests?

3

Channel name is: llamalanettv

Here are my findings:

  • Capital letters filter: permitted percentage 100%, characters minimum 50, type of action - warning. Findings: When sending a single message with 51 of the same capital letters in a row, the message looks like it is sent for the user, but others in the chat do not see the message (I believe twitch is catching it on their end and must have a filter in place for this). But when sending a single message with 51 random capital letters in a row, all users can see the message in chat and a warning is not given.

  • Repeat message filter: repeat words at 20, repeat letters at 30, number of characters minimum 29, type of action - warning. When sending a single message with 21 of the same words repeated, a warning is not given but when sending a single message with 31 repeated letters, a warning is given.

  • Spam message filter: interval 10 secs, maximum messages 5, type of action - warning. When sending 6 messages within 10 secs, a warning is not given.

  • Special characters filter: special characters at 100%, characters minimum at 60, type of action - warning. When a single message with 61 exclamation marks in a row is sent, the message looks like it is sent for the user, but others in the chat do not see the message (I believe twitch is catching it on their end and must have a filter in place for this).

  • Unicode characters filter: unicode characters at 10%, characters minimum at 5, type of action - warning. When a message generated from lingojam(dot)com/ItalicTextGenerator is sent, a warning is not given (I may be testing this functionality wrong).

  • Advanced filters: fake donation, deleted message, and /me filters are active, type of action - deleting the message. /me messages are deleted. When sending a “XYZ has donated” or “deleted message” or “message deleted by a moderator” message, the messages are not deleted (I may be testing this functionality wrong).

4

At present, there’s a problem: if 100% is set, the margin of error (1%) makes detection non-positive.

=> The problem will be corrected by tomorrow.

Generally, it is recommended to set 70% / 80%.

We have a limit on the size of words (5 characters).

=> We’re going to reduce this limit by tomorrow so that we can include “smaller” words.

For the moment, this protection is only available online.

As for the capitalization system, the 100% with margin of error is causing problems.

=> Will be corrected at the same time as for capital letters.

Likewise for capital letters, we recommend 70% / 80%.

According to our history, your test was a 4-character word, which is below the limit :slight_smile:

Has donated" messages are exclusive if /me is used. The simple message with no action is easy to spot :slight_smile:

For “deleted message”, only “< message deleted >”, “< link deleted >” are deleted.

These protections are quite old, and we plan to update them in the future to include many more combinations and types of misleading messages.

Regarding the warning system configuration:

You’ve set the “No action” type for the first 3 options, I strongly recommend that you set at least the message deletion.

If, for example, someone posts a message with an inappropriate ASCII drawing or abusive messages in uppercase, it’s preferable that the message be deleted :slight_smile:

Thanks in any case for the many precisions !

5

I tested this again today and can confirm, when a single message with 51 random capital letters in a row is sent, a warning is now given. Awesome work, thank you.

I have updated my percentage to 80% per your recommendations. I originally misunderstood the permitted percentage setting.

I tested this again today and with repeat words set at 20 it is not picking up 2, 3, 4, 5, or 6 letter words when repeated 21 times in a row in single message.

Just to confirm, this system is intended to work with spaces in-between the words is that correct? For instance, with my current settings, it should block (once you lower the 5 minimum word count): “four four four four four four four four four four four four four four four four four four four four four” ?

Great recommendation! I did not realize that even when a message is deleted using the warning system, the user is still given the warning. This is great!

Thanks again for the help!

6

Yes, it’s a word repetition with spaces :slight_smile:

Consider the first word as a non-repetition, the other indentical words as repetitions.

So, in this case, only 20 words are counted.

You also need to pay attention to the limitation, which corresponds to the maximum number allowed, so 20 will only be triggered if 21 words are detected (i.e. 22 times the same word if we take the first word into account).

7

Ok, perfect. I tested it again with this information in mind and it works great. I also see the word size was decreased to three letters.

Thank you!

1 Like
8

This topic was automatically closed 5 days after the last reply. New replies are no longer allowed.