New WSO Moderation Efforts [Updates Here]
As many of you already saw the thread talking about moderation on WSO, I thought it would be helpful to start a new thread that gives everyone updates on where we are relative to the plan I posted there... as you can see, this morning we already sent a lot of updates live. We will be keeping a close eye on all of these.
Below are some details on what is coming and a rough timeline. So as to discourage trolls, I will be intentionally vague on where certain thresholds are + the exact rules. All of these are subject to change/removal as we see how it impacts everyone and how some of them get abused (which we know will eventually happen).
UPDATES THAT ARE LIVE [10/3/22]:
- Posting to the Off Topic forum now requires a certain type of profile...so any members outside this "allowed range" will lose the ability to post to the off topic forum (this is a tiny Band-Aid, not a perfect measure... but it's a start to limit the members that tend to debate politics and/or are the most extreme). No, it won't stop most of the users that like to debate, but it is likely to limit the obvious trolls. [DONE]
- New members will also NOT be able to post to the Off topic forum for an unspecified amount of time + until they earn a certain number of bananas (again, this one is minor, mostly as a way to discourage troll users from creating new accounts to start shit-posting right away again after they have just been blocked). [DONE]
- New automated unpublishing rules based on a certain criteria (+time based) to be reviewed by mods. This one should help more meaningfully since it is automated but we will need to keep an eye on and see what the flow looks like... I suspect that it will help get the worst comments and threads off the site faster until it can be reviewed later. This requires us to build in some new features to our current moderation table + make sure our permissions are dialed in on mod role but we think we can have it live by mid week [DONE]
- Autoblock members that hit certain criteria. This has been all manual in the past, so it would be good to automate this so that the worst actors eventually lose their account if they are not adding enough value [DONE]
UPDATES COMING LATER THIS WEEK [10/3/22]:
- (likely by Thursday) New violation notice will go live and be used to help remove controversial topics faster. Like @m1 mentioned, religion and politics are good examples : ""Violation Magnet: Hi user, while your post itself may not be directly against our official terms and conditions, it has been removed because it will lead to a high number of violations (or already has). As such, to save WSO from going bankrupt and having a team of thousands of moderators, we had to remove your post. We appreciate your contributions and hope you understand this decision. Best Regards, WSO Mod Team" [THIS IS NOT TOO HARD AND IS JUST A WAY TO MAKE SURE PEOPLE KNOW WHY SOMETHING WAS REMOVED...WE MAY NOT NEED TO DO THIS TOO OFTEN IF THE CHANGES ALREADY IMPLEMENTED ABOVE HELP REDUCE VIOLATIONS]
- Hive Moderation which is used by other large content businesses. By using their API, we can potentially use their AI mod tools as a first line of defense to either flag more or just autoblock content that scores too high on various metrics: https://docs.thehive.ai/docs/classification-text This will require more dev work to integrate with our system, but I'm hoping we can go live with something before end of October [SOME TECHNICAL WORK NEEDED W HIVE TEAM SINCE THERE ARE CHARACTER LIMITS THAT WE NEED TO WORK AROUND. THIS WILL LIKELY PUT ALL NEW COMMENTS AND POSTS IN A QUEUE SO AS NOT TO DELAY POSTING BUT TO GET THEM REVIEWED WITHIN ~5-10 MINUTES AFTER]
UPDATES COMING IN 2+ WEEKS:
- Team of 5 volunteer mods that can flag and remove content and recommend bans based on user history - hoping to have a few ready to go by end of next week so the automated flagging/unpublishing doesn't drown us
- Better visibility of moderation for comments. Instead of just unpublishing comments that are against our T&C, we will try to "grey them out" and/or have some sort of indicator that a comment was removed...this way, if there are replies nested under that comment it won't make the discussion very confusing...especially in topics that are controversial (still TBD whether they will just be removed entirely)
ON THE TABLE BUT MAY TAKE LONGER:
- Allowing users to "mute" specific other users that they don't like or are offensive according to them
- Potentially requiring certification status to post anonymously. I'm open to this but I want to see how everything trends with the above changes into November before we make this one.
That's all for now,
ps - have one call scheduled on Monday w a potential mod but would like more. If you have a long history on WSO of avoiding all political/controversial topics and focusing on helping members, I'd love to speak with you. [email protected]
- Hive moderation (using API) testing is complete and ready for roll out to production server on Monday Oct 10, 2022. Scoring explanation here: https://docs.thehive.ai/docs/classification-text
- Posts scoring a 1 will be flagged automatically but stay up by default. This will allow mods to review it to determine if it should stay live or get removed. Mod will review and either send violation notice which then unpublishes the content OR will clear the flag which will keep it up.
- Posts automatically flagged at level 2 or 3 will be unpublished and sent to flag table for review and user will be notified that their post had a potential violation and was removed. Mod will review and either send violation notice OR will clear the flag which will republish it.
- This will not be perfect but it's really more as a line of first defense...I tested it an it does a decent job at flagging obvious violations but it misses nuance (as expected)
- New MOD ROLE IS READY. Actually called "Content Reviewer" - we had to create a new role in order to make sure permissions were dialed in and that the new content reviewers only had the right level of permissions. We are still working on documentation to help the mods, but I could still use some volunteers here (only a few I've had calls with).
- We have increased shifts of internal mods to help us deal with the 200% increase in the # of potential violations already hitting the flag table with the new systems already in place. I think it's working to get violations off faster, but it's not perfect
- New Violation Magnet PM created (now LIVE) so mods can send violation notices to members that post threads to incite debate, get a reaction or prove a political point...
COMING SOON [Posted this 10/6/22]
These next set of initiatives (after Hive on Monday and what we've already rolled out) have me the most excited because I think they will make a meaningful difference long term and are more WSO specific...
- Better transparency...greying out but leaving "frame" of comments that have been removed so responses to those violations are less confusing and there is more clarity when something has been taken down because of a violation
- "No bump forever" flag on threads will no turn into "No bump Forever + No Anonymous" which means two things
- 1. Any new comments will not bump the thread up to the top of the recent activity trackers [how it is now]
- 2. Any new comments after flag is applied will force participants to post under their username [this is new]. Even though admins can see what members are posting even if they use anonymous feature, the goal here is for people not to hide behind the anonymous feature on the most controversial/sensitive topics in order to try and encourage more respectful debate, especially in these threads
- Adjustment to Silver Banana weightings from Topics and comments in the Off Topic forum...tbd if we can do this, but should help lower the benefit of riling up your "team" and getting a bunch of SBs as a result at the expense of another member
Thank you for your patience and help with these ideas...
- Hive moderation ("first line of defense") API is LIVE. This means every new post or comment is run through this API and scored. See above for explanation
- New Moderator ("Content Reviewer") role is READY. I will start granting this new role this week to a few select members
- No Bump forever flag applied to threads has now become "No bump forever + No Anonymous flag"... this means once this flag is applied to a thread, everyone must use their username to post and can't do it behind an additional layer of anonymity. While this is not a huge difference, we think it ay help reduce violations on the margin and lead to a slightly more respectful tone on average...
- NOT live but still high priority = leaving a placeholder for moderated comments so people know when something was moved (not just the person with violation notice)
- Re-weighting of SBs granted in off topic forum
...After this week I'd like to take an honest assessment of the changes that are currently live and see if we think the new system is doing a good job (or not) and where we may be able to improve things further.
- LIVE = now there is a placeholder for moderated comments so people know when something was moved (not just the person with violation notice)
- Hive moderation ("first line of defense") API is adjusted so that only level 3 violations is automatically unpublished (to reduce false positives). Removed potential violation PM notice for level 1 and level 2 violations since this was causing confusion. They are still auto-flagged and reviewed by mod team but the user will only be notified if their content is unpublished (level 3 potential violation or actual PM violation)
- New Moderator ("Content Reviewer") role has been granted to 3 members. I would still like an additional 3 more mods to help us since the number of flags is now ~3-4x what it was prior to these changes. Potential violations still sitting for too long in the flag table.
- Fixed anon posting issues in the mobile app related to some of these changes...new restriction was not being applied correctly there.
- Re-weighting of SBs granted in off topic forum still to come...
...still very much in the learning and tweaking stages.
Thanks for your patience,
- Removed any automatic unpublishing from Hive moderation. There was still too many false positives (w level 3) so Hive is still checking every piece of content, but it's just flagging potential violations and we're leaning on mod team to get violations off the site (some won't be immediate)
- To deal with dramatic increase in flags, we have brought on 2 additional staff members to mod team (internal to WSO) to help have better coverage of the flag table throughout the day + evenings + weekends
- New tool in mod arsenal to help deal with troll members... top secret
- Re-weighting of SBs granted in off topic forum still to come...likely in ~2 weeks
Overall, I think we're doing a much better job than before. We've blocked ~3 members in the last 2 weeks that refused to reduce their violations... some that have been around for years but were creating 80%+ of the work... others we've warned and so far they've been better / more helpful.
Thank you for your continued patience...it feels like we're starting to turn a corner.
UPDATES THAT ARE LIVE [10/27/22]:
- Boosted data for the mods on the flag table to make better decisions (types of violations flagged by Hive, etc)
- Started violation tracking by user so that we can build a history of violations and start automating some rules for repeated offenders that should decrease the mod burden over time
- We are still going to re-weight the SBs granted in off topic forum but it is still to come...still likely in ~2 weeks
- In other news, with better mod team we are also now reviewing every company database submission BEFORE publication to help reduce fakes (and incentives for fakes). we will also introduce more sanity checks that should increase the signal: noise ratio dramatically in coming weeks...
Overall, I feel like the efforts are paying off...the conversations overall seem to be much more respectful - even in some violation magnet threads that were left up. I'm hoping this continues but I understand we'll need to stay on top of it.