rustrict

rustrict is a sophisticated profanity filter for Rust.

Features

• Multiple types (profane, offensive, sexual, mean, spam)
• Multiple levels (mild, moderate, severe)
• Resistant to evasion
  • Alternative spellings (like "fck")
  • Repeated characters (like "craaaap")
  • Confusable characters (like 'ᑭ' vs 'P')
  • Spacing (like "c r_a-p")
  • Accents (like "pÓöp")
  • Bidirectional Unicode (related reading)
  • Self-censoring (like "f*ck")
  • Battle-tested in Mk48.io
• Resistant to false positives
  • One word (like "assassin")
  • Two words (like "push it")
• Flexible
  • Censor and/or analyze
  • Input &str or Iterator<Type = char>
  • Plenty of options
• Performant
  • O(n) analysis and censoring
  • No regex (uses custom radix trie)
  • 4 MB/s in release mode
  • 150 KB/s in debug mode

Limitations

• English only
• Censoring removes diacritics (accents)
• Does not detect right-to-left profanity while analyzing, so...
• Censoring forces Unicode to be left-to-right
• Doesn't understand context
• Cannot add words at runtime

Usage

Strings (&str)

```rust use rustrict::CensorStr;

let censored: String = "hello crap".censor(); let inappropriate: bool = "f u c k".is_inappropriate();

assert_eq!(censored, "hello c*"); assert!(inappropriate); ```

Iterators (Iterator<Type = char>)

```rust use rustrict::CensorIter;

let censored: String = "hello crap".chars().censor().collect();

assert_eq!(censored, "hello c*") ```

Advanced

By constructing a Censor, one can avoid scanning text multiple times to get a censored String and/or answer multiple is queries. This also opens up more customization options (defaults are below).

```rust use rustrict::{Censor, Type};

let (censored, analysis) = Censor::fromstr("123 Crap") .withcensorthreshold(Type::INAPPROPRIATE) .withcensorfirstcharacterthreshold(Type::OFFENSIVE & Type::SEVERE) .withignorefalsepositives(false) .withignoreselfcensoring(false) .withcensorreplacement('*') .censorand_analyze();

assert_eq!(censored, "123 C*"); assert!(analysis.is(Type::INAPPROPRIATE)); assert!(analysis.isnt(Type::PROFANE & Type::SEVERE | Type::SEXUAL)); ```

Comparison

To compare filters, the first 100,000 items of this list is used as a dataset. Positive accuracy is the percentage of profanity detected as profanity. Negative accuracy is the percentage of clean text detected as clean.

| Crate | Accuracy | Positive Accuracy | Negative Accuracy | Time | |-------|----------|-------------------|-------------------|------| | rustrict | 90.52% | 91.49% | 90.28% | 8s | | censor | 76.16% | 72.76% | 77.01% | 23s |

Development

If you make an adjustment that would affect false positives, you will need to run false_positive_finder: 1. Run ./download.sh to get the required word lists. 2. Run cargo run --bin false_positive_finder --release --all-features

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.