Username Guardian is a tool for detecting toxicity in usernames. You can pass any username to the API in order to subject it to a linguistic analysis to identify any disruptive, profane, sexual or otherwise inappropriate language. With the Username Guardian you can protect your communities and eliminate the risk of a toxic username setting the tone for the ensuing discourse within the community.

Username Guardian uses a neuro-symbolic approach to dissect each username into its constituent parts, identify disruptive content and provide detailed information on what makes the username toxic. Importantly, Username Guardian can handle such text manipulation techniques as:

• leetspeak - replacing letters with digits or symbols (D1ck$uck - dick suck)
• spelling the username backwards (kcuSkciD - dick suck)
• swapping the initial letters of two words (DuckSick - suck dick, PinnySkenis - skinny penis)
• other types of self-censorship (SxckDxck - suck dick)
• intentional misspellings (SackDik - suck dick)
• flanking the username with various characters (xXSuckDickXx, oooSuckDickooo - suck dick)
• letter multiplication (SssuuuckDdickkkk - suck dick)
• use of very similar letters from other alphabets, e.g. Russian Cyrillic у, с, or р, which, when used in an English text, escape traditional filters based on the Latin alphabet (SuсkDiсk with Russian c - suck dick)
• use of letter variants with diacritics (Sȗck Dȋck - suck dick)
• and any combination thereof

This way, the Username Guardian is able to decipher even deeply hidden offensive meanings.

Then, the Username Guardian classifies toxic content according to one of the nine toxicity categories:

• Offensive - this category covers a wide range of vocabulary almost universally considered offensive, includes, but is not limited to sexist, racist and queerphobic language, for example: D34thToFurr1es (Death to Furries), sucker90000, AutistikUncle
• Sexual - words and phrases related to body parts, sexual activities and fetishes, for example: Whit3Long5chlong (White Long Schlong), CallMeSexGod, HornyItalian
• Profanity - collection of vulgar words and phrases, common profanities and curse words, for example: FolyHuck (Holy Fuck), FackThis, HOLISSSHHHIIITT (Holy Shit)
• Violence - this category is made up of words and phrases that relate to serious and unambiguous violent acts, known serial killers and torture methods, for example: xXP3D0_B34RXx (pedo bear), FreeTedKaczynski, Homicide76
• Political - vocabulary related to politics - controversial or condemned political figures, problematic world events or extremist worldviews, for example: Vl4dim1r_Put1n (Vladimir Putin), lovenazi, HolocaustWishes
• Religion - words and phrases related to offending on the grounds of religion, taboos or religious extremism, for example: Un3xp3ctdJ1h4d (unexpected jihad), AllahuAkbar2nd, JesusHas1BalSack
• Drugs - this category includes drug and medicine related vocabulary as well as slang terms related to drugs, drug use and drug users, for example: m0rfin4 (morfina), igotcrack, BongMaster
• Medical - vocabulary related to illnesses and medical procedures, mostly dealing with taboos or unpleasant imagery, for example: HIVdestruction, abortion survivor, DiarreahYumyYumy
• Other - a mixture of scatological humor, mildly violent or gaming-related violent vocabulary and other milder terms that are otherwise difficult to categorise, for example: Suic1deboy (suicide boy), IFartedAgain, Pervert_Pete

For instance, for the input DuckSick96, the pipeline of the Guardian consists of the following steps:

1. Strip numbers DuckSick
2. Split two words duck sick
3. Swap the initial letters suck dick
4. Classify the Username Toxic Username; Keywords: suck, dick; Category: sexual (‘suck dick’), profanity (‘suck dick’), inappropriate (‘suck’)

The input N00B5H1T triggers the following pipeline:

1. Decode leetspeak Noobshit
2. Split two words noob shit
3. Classify the Username Toxic Username; Keywords: shit; Categories: profanity (‘shit’)

With the Username Guardian, you can configure which types of usernames are not allowed in your community, according to your guidelines. If you run an app for kids, most likely all of the four types of toxic usernames should be ruled out. If the use case involves a dating community for adults, you can opt for enabling the usage of sexually explicit usernames but disable those that are offensive, profane, or inappropriate. Since a username can be classified into more than one category, Username Guardian will know which sexually-explicit usernames might at the same time be offensive and should be filtered out.

Username Guardian can prevent a user from registering the toxic username or it can be used to scan an already existing user base for toxic name spotting. It can help in enforcing your policies against toxic behavior and proactively prevent its breaches, resulting in increased safety and trust in your service.

There is research conducted by Samurai Labs and Riot Games independently, confirming the correlation between toxic usernames and toxic behavior: users with toxic names are much more likely to produce violent messages or to harass others. Preventing users from opting for such a username is a strong signal to your users about the rules that are being enforced and could potentially prevent them from further disruptive behavior.

The Username Guardian is an effective tool that allows community owners and moderators to effectively automate this aspect of moderation, without the need to rely on a manual process, reports by other users, or low-precision keyword methods.

Supported Languages

Language - ISO-639-1 Code

English - en
Polish - pl
Spanish - es

To detect toxicity in usernames in all available languages, use all in the language parameter. If no value is provided for parameter ‘language’, the default is en.

For more information, visit our website: https://www.samurailabs.ai/#UNM