Hey Community!!
We have recently updated the article with information on Content Moderation Widget!

Thanks a lot for this new feature! I have to start using it ASAP 😀

At this point, I have a few questions:

1. When we write the Code of Conduct for the AI, what kind of tone should we use, and who should we address in the text? Is it the AI, the community member, or does it not really matter?

2. About the tags — what does “SFW” mean?

3. The Moderation AI works with topics (opening posts) and replies. If the AI thinks that a member’s reply to a topic is suspicious and marks it as “Pending” (for example) because there are no moderation tags in the reply, how can we know why that decision was made?

Hi ​@revote,

Thank you for reaching out to us with your query. Tagging the PM ​@Graeme Rycyk for more inputs on these, Thank you.

Hey ​@revote,

My name is Graeme and I am the Product Manager for the Moderation AI Agent. First, thanks for your questions here.

1. Your Code of Conduct can be written like you were posting your community rules to your community but it can also be written like training or onboarding documentation for a new member of the community team. The ideas is it operates the same way as your human community moderators would moderate on your community.

2. SFW stands for "safe for work," which means that the content is appropriate and is often used to indicate that something is suitable to view in a professional or public setting.

3. The Moderation Agent uses Moderation Tags to give your community team a fast and easy way to understand what moderation decisions have been made and why. Further to this when a post is Trashed and Reported a more detailed explanation of the reported reason is also included.

If you have any other questions, please feel free to reach out.

All the best,

Graeme

Where these tags can be seen? Is there a new field in the Control? Currently, moderation tags appear in the opening post.

Hey ​@revote,

The Tags that the Moderation Agent adds are visible in the “Moderator Tags” section of the Moderate Topic page in Control.

Home > Content > Overview > [Topic]

See this example here:

If you need anything please let me know.

Cheers,

Graeme

 

Maybe I dont how to ask, but your example looks like tags from to opening post.

Will there be same kind of field next to each and every comment?

Hey ​@revote,

Yes they do look like the Public Tags, but Moderator Tags are only visible inside Control, if you go to the moderation view of a topic you will see a input field labeled Moderator Tags, like in the close up screenshot in my earlier post or see a full screen annotated image below. I hope that helps clear this up.
 

If you need anything please don’t hesitate to ask.

Cheers,

Graeme

Thanks, but you are still not answering my question. I dont know how to ask differently 😃

Hi ​@revote,

Sorry about that. I think I understand the crossed wires here, by comments, I think you are referring to Replies, I took “comments” more broadly and thought you were referring to all posts.

So Moderator Tags are not applied to Replies, only Topics. However, we will be extending and improving functionality here to give more context on the status of Replies, I cannot give a firm date on this yet.

Note: When a Reply is Trashed there will be added context in the “Report” feature to elaborate on the reason the reply was Trashed.

I hope this clears up things.

Cheers,

Graeme

Okay, so how this actually works?

(Sorry, I was not able to delete those tables here)

Hi ​@Graeme Rycyk , 

I am testing this tool in our sandbox - seems to work well even in Finnish language 💪🏻

However I have found some cases where synonyms like nasty, mean, vicious, unkind were not always recognised by the tool. How can we teach the AI?

And another thing came up:

 

The upper comment/reply contains a phone number > it was trashed.
The lower one contained an email address > it was automatically moderated.

Why the logic is different? Is it because email is more easily recognized as an email?

Hey ​@Suvi Lehtovaara,

An easy way to solve this would be to add words you want to ensure are always caught into your Code of Conduct.

Do let me know if you have any other questions or if this doesn’t work.

Kind regards,

Graeme

Hi ​@Suvi Lehtovaara,
Mohammed Fahd here, I’m the POC from documentation for this article. We have recently received a request from you to access the Overview of Control Home article. Apologies for the inconvenience as the link was wrongly tagged. 
We have now updated the article link - Overview of Control Home. Please let me know if you have any trouble accessing it. Thank you

 

 

Thanks, this is helpful post.

So there is no Moderation tags in replies, so we dont know what and why actions are made.

 

 

Thanks, this is helpful post.

So there is no Moderation tags in replies, so we dont know what and why actions are made.

Yes, that’s true ​@revote. And if there’s no context in the post, the reason may stay unclear.

 

 

Thanks, this is helpful post.

So there is no Moderation tags in replies, so we dont know what and why actions are made.

At this moment yes, this is correct, but this is an improvement we intend to make in the near future.

Kind regards,

Graeme

Also, I didn’t know that AI deletes content automatically. I thought it just hides content that goes against the code of conduct.

Why is this a problem?

If you think about the DSA, the Digital Services Act: if we limit user communication by deleting content or even just part of it, the DSA states that the user has the right to make a note about the decision (the decision to delete content). And we are obligated to handle that note.

In theory, the decision to delete might be wrong. That means we would have to restore the content.

But since there is no version history in replies, restoring the content is not possible.

We need to consider whether we should start using Moderation AI at this point.

Hey again ​@revote,

So AI Moderation doesn’t delete any content, it only Trashes and adds a Report reason for further context for content that was trashed. This is how content is removed from public viewing in our communities. If mistakes are made by the AI Moderator, say by mistakenly Trashing content, this can always be reversed by the human-moderator. Our “Notify Author” flow can be applied by your community team to adhere to the DSA even if using AI Moderator. Custom views can be set up to enable your community team to easily create a workflow to track the moderation action of the AI Moderator.

I would be happy to jump on a call to discuss things and take more questions, if you would prefer.

Kind regards,

Graeme

Okay, good. Thanks.

I just looked this screenshot (lower example) and ​@Suvi Lehtovaara´s comment:

 

The upper comment/reply contains a phone number > it was trashed.
The lower one contained an email address > it was automatically moderated.

It is good to know that it doesnt delete content automatically. It just hides content.

To be precise, the email address was deleted. I mean, at least in our case we could not find the original content from Control.

I can confirm, AI moderation deletes content automatically. Here´s example, it deleted phone number and then it published my test topic:

But same time I have to say that this works perfectly 👌 I have tested with several different ways and AI works nicely. Nice job!

Btw, I didnt find phrase for this text, to translate in Finnish:

Based on my test, the AI didn´t delete the email address. It just hid the post.

But it deleted phone number when I tested it with separate post 😀