Flashtext2

This crate allows you to extract & replace strings very efficiently, and with better performance than using RegEx.

Its especially performant when you have a very big list of keywords that you want to extract from your text, and also for replacing many values.

How it works

The flashtext algorithm uses a trie to save all the keywords the user wants to extract, a keyword is defined as a sequence of tokens, for example "Hello world!" becomes: ["Hello", " ", "world", "!"]. And in this implementation, each node in the trie contains one token (not character!).
The tokens are split using the Unicode Standard Annex #29.

Time complexity

The time complexity of this algorithm is not related to the number of keywords in the trie, but only by the length of the document!

Quick start

use flashtext2::case_sensitive::KeywordProcessor;

fn main() {
    let mut kp = KeywordProcessor::new();
    kp.add_keyword("love");
    kp.add_keyword("Rust");
    kp.add_keyword("Hello");

    assert_eq!(kp.len(), 3);

    // extract keywords
    let keywords_found: Vec<_> = kp
        .extract_keywords("Hello, I love programming in Rust!")
        .collect();
    assert_eq!(keywords_found, ["Hello", "love", "Rust"]);

    // extract keywords with span
    let keywords_with_span: Vec<_> = kp
        .extract_keywords_with_span("Hello, I love programming in Rust!")
        .collect();
    assert_eq!(keywords_with_span, [("Hello", 0, 5), ("love", 9, 13), ("Rust", 29, 33)]);

    // replace keywords
    let mut kp = KeywordProcessor::new();
    kp.add_keyword_with_clean_word("Hello", "Hey");
    kp.add_keyword_with_clean_word("love", "hate");
    kp.add_keyword_with_clean_word("Rust", "Java");

    let replaced_text = kp
        .replace_keywords("Hello, I love programming in Rust!");
    assert_eq!(replaced_text, "Hey, I hate programming in Java!");
}

Case insensitive

The KeywordProcessor struct is defined in two modules: case_sensitive and case_insensitive. Both modules provide the same methods and signatures; however, the internal string storage differs. The case_insensitive module utilizes a case-insensitive hashmap (case_insensitive_hashmap).

use flashtext2::case_insensitive::KeywordProcessor;

let mut kp = KeywordProcessor::new();
kp.add_keywords_from_iter(["Foo", "Bar"]);

let text = "Foo BaR foO FOO";
let keywords: Vec<_> = kp
    .extract_keywords(text)
    .collect();
assert_eq!(keywords, ["Foo", "Bar", "Foo", "Foo"]);

The unicase crate accurately processes and matches keywords despite variations in case and more complex characters:

use flashtext2::case_insensitive::KeywordProcessor;

let mut kp = KeywordProcessor::new();
let tokens = ["flour", "Maße", "ᾲ στο διάολο"];
kp.add_keywords_from_iter(tokens);

let text = "flour, MASSE, ὰι στο διάολο";
let found_tokens: Vec<_> = kp.extract_keywords(text).collect();
assert_eq!(found_tokens, tokens);