How Minds is dealing with SPAM

Spam and the Internet, name a more iconic duo.

We've had our own challenges with spam that you are no doubt well aware of. What may be less obvious is what we've been doing to combat spam on Minds. Let's talk about what changes you can expect to see very soon.

What we're up against

But before I get into the changes, a quick recap on the scope of the spam battlefield on Minds. This will give context behind the changes we've got planned. As we'll see, spam isn't just one thing, so a one-size-fits-all solution won't cut it.

• Link spam: Content intended to attract clicks, most often dropped into comments of popular posts. These guys suck, but at least they're obvious. They know 99.9% of people aren't gonna click their links and rely on a massive volume of spam content to fish for a few clicks. Because of this, they're obvious everywhere and one of the most-seen types of spam on the Internet. On Minds included.
  
  Impact: There's visible spam in comments of popular posts, gross.
   
• SEO spam: Content with backlinks intended to improve the Google search rankings of the pages they link to. Generally, this content is very obviously spam when it's found, but finding it isn't always easy. For these spammers, it's not so important that anyone on Minds actually sees the content – in fact, they'd rather go unnoticed. They're banking that Google will see the backlinks to their scuzzy websites and improve their rankings in search results, since Google favors sites with more backlinks. Thanks, Google.
  
  Impact: Risks Minds getting de-ranked or blacklisted in public search results; Creates cruft on the site, in search results, latest feeds, etc.
   
• Influence spam: Link and SEO spam are common across the Internet, but social media apps like Minds enable a new type of spam: Content that is intended to manipulate public opinion by making a particular opinion (often related to public policy) or person (politician or influencer) appear more popular than it really is. Minds is not and will never be in the business of restricting the expression of points of view, but we do want to make it easy for people to understand which points of view are being expressed by individuals and not sock puppet accounts. Having multiple accounts on Minds is A-OK, but using them to mislead, or to manipulate voting mechanisms like Minds jury undermines the value of public discourse. As you might guess, this spam is amongst the hardest to detect, as it's intended to blend into the community.
  
  Impact: Public opinion can be manipulated; Voting mechanisms like Minds jury are vulnerable to sybil attack.
   
• Rewards spam: Inauthentic content and engagement spam that's coordinated to exploit the Minds token rewards program. This spam is designed to blend into the site and look like real content, but rather than expressing the point of view of an individual and being rewarded with engagement from other individuals, it's designed to collect inauthentic engagements from sock puppet accounts and siphon rewards.
  
  Impact: Public feeds like Discovery, Search, and tag feeds are cluttered with low-quality content no one actually likes; High-quality content gets fewer token rewards because inauthentic accounts skim off the rewards pool.

As if what we're up against wasn't difficult enough, the Minds mission assumes values that make things even harder in the war on spam.

• Free speech: We believe elevated discourse requires the defense of free expression, for everybody. So what's the line between genuine expression and inauthentic spam? It's not always black and white, and for our defense of expression to be credible we can't limit speech in the gray zone.
   
• Privacy: Unlike some other social apps, Minds doesn't require that people use their real names. We believe pseudonymity is important for enabling free expressions since not everyone is comfortable with that level of exposure. Further, we don't run sophisticated ad trackers or other invasive behavior-tracking technology that might let us formulate identities with big data. As you might imagine, this commitment to privacy poses challenges for combating spam and determining individuality.
   
• Open source: We open source all of the code behind Minds, and while it might be tempting to imagine spammers as dummies, they're often very smart and capable of understanding our public code. So if, for example, we notice that spammers always post on Wednesdays at 3:00 pm from Linux machines and we publish code that automatically marks users that match that profile as spam, they'll read the code and change their tact.
   
• Inclusion: For better public discourse, everyone needs a voice. Minds is a community open to all. You get to freely express yourself no matter your opinions, your identity, your popularity, or your status.
   
• Rewards: We believe quality content should be rewarded, so we built a token-based rewards system. And because we don't want to be the arbiters of what's quality content, the rewards are distributed based on engagement, effectively a democratic vote. But this creates a valuable incentive for manipulators to exploit. In other words, we'd have less spam if we didn't have a democratic rewards program. But we do have one, and we don't want to lose it.

What we're doing about spam

While it hasn't been obvious, in the background we've been working on solutions for all of these problems and we're ready to pull the trigger on rolling out many. We don't expect the war on spam will ever really end, but we now have effective tools to deal with spam at scale.

• Auto-reporting link spam: We've developed data models that can identify obvious link spam at scale. We've been testing the models internally and the accuracy is high – we can detect ~95% of link spam comments. Link spam gets automatically flagged into the report queue for initial admin review and public jury adjudication upon appeal.
  
  Timeframe: Live now!
   
• Auto-blocking link spam: While auto-flagging spam is useful, it's even better if the spam never makes it to the comments in the first place. We can use the same spam detection data models to process comments as they're being made, in real time. When a suspected spam comment is detected, we can ask the poster to complete a CAPTCHA or two-factor authorization before the comment is posted.
  
  Timeframe: Soon
   
• No-indexing SEO spam: A subtle way we deal with SEO spam is by indicating to search engines to ignore channels and content pages from accounts that haven't yet proven themselves. We're not satisfied with the solution we have now, as our criteria for "proving" oneself is not currently robust. But with our new capabilities to detect spam in comments, we can turn our models to other content (blogs especially) for similar automatic detection of spam. This will allow us to default trust new accounts with indexing in search engines, while no-indexing based on a more targeted and intelligent heuristic.
  
  Timeframe: Interim solution now; Improvements later
   
• Filtering public feeds: If spammers want to post and subscribe to each other on Minds… maybe they can and that's OK. But it becomes not OK when inauthentic content gets into public feeds, in the way of people who don't subscribe to them – think Discovery feeds, trending tags pages, and search results. While we're as wary as anyone of impinging on freedom of speech based on machine learning data models, especially with content that skirts the line between spam and questionably authentic, we feel giving users tools for filtering spam out of public feeds can drastically improve the general experience on the site and reduce the incentive for generating spam in the first place. We think an automated filter like this should be optional so that if you want the raw, unfiltered Minds, it's just a button tap away.
  
  Timeframe: Soon
   
• Ignoring inauthentic engagements for rewards: Similar to our automatic spam detection, we've gotten good at detecting engagement from inauthentic accounts. As we said before, maybe it's fine for these accounts to publish sus content and upvote it – far be it from us to infringe their expression. But it's not what the token system is designed to reward, and removing those inauthentic engagements from the daily calculations in rewards drastically reduces the negative community costs of their inauthentic engagement.
  
  Timeframe: Soon
   
• Proving uniqueness: Currently, we implement pretty soft checks to validate a new account is from a real person. (1) We require accounts to verify their email address in order to interact on Minds, and (2) we require accounts to verify a unique phone number in order to participate in rewards. But to be frank, email verification is easily gamed, with no shortage of free email providers and bot scripts capable of operating them. We've learned the hard way that even a unique phone number is not as rare as you might think; while you could probably imagine it's possible to buy a handful of phone lines with enough money, that's just the tip of the iceberg. In some countries, phone numbers are purchasable en masse for cheap, invalidating their effectiveness at proving uniqueness.
  
  For now, we've disabled phone verifications in countries where we know this circumvention is possible, but we're not satisfied with that solution. For starters, it's still not super effective, but also we don't want to exclude entire countries from rewards and other Minds features. We need a better solution.
  
  We're currently building out a new uniqueness verification flow that involves using the Minds mobile app. It's designed to enable anyone to anonymously prove they're human. When we roll this out, it'll enable us to replace phone verification to join rewards. We'll ask everyone who wants to be part of the rewards program to re-verify. This will enable us to better verify that engagements are unique, and enable people in all countries (excepting OFAC-sanctioned countries) to participate. Additionally, we're considering adding badges to channels and/or avatars that let everyone see who's verified that they're a unique human.
  
  Timeframe: In progress now

Checks and balances: Involving the Minds jury

You might be thinking, duh just do it already. But we're treading carefully here. We're confident in the automatic spam detection we've developed, but we need the community behind it too. And importantly, we need failsafes for when the automation gets things wrong. As we've discussed, there's often gray area, and one person's spam may be another person's inspirational dank meme.

Thankfully, we have the Minds Jury, a decentralized and randomized system for the community to adjudicate appeals on moderation decisions. What counts as spam on Minds ultimately isn't our decision – it's yours. So we'll be making upgrades to the jury in the coming months, to better deal with an increase in volume of appeals and also to better inform jurors when they're making their decisions to uphold or overturn moderation actions.

Further, we have some work to do in service of maximum transparency. We haven't yet open sourced our spam detection algorithms – we're still figuring out how we can expose them to public scrutiny without giving spammers a playbook for getting around them. We also want to make sure that any automated moderation action is publicly disclosed. If an account gets flagged as suspicious, we want to let that account know and give them options to appeal to the Minds jury.

Stay tuned for more. And as always, your feedback helps us make better decisions.

Provided you're not spam.