An extremely fast implementation of Aho Corasick algorithm based on Double Array Trie.
Latest commit 56defae Oct 7, 2016 @hankcs Maven supported
Failed to load latest commit information.
src Clarify the Apache License Sep 28, 2016
pom.xml Maven supported Oct 7, 2016


Maven Central GitHub release License

An extremely fast implementation of Aho Corasick algorithm based on Double Array Trie structure. Its speed is 5 to 9 times of naive implementations, perhaps it's the fastest implementation so far ;-)


You may heard that Aho-Corasick algorithm is fast for parsing text with a huge dictionary, for example:

  • looking for certain words in texts in order to URL link or emphasize them
  • adding semantics to plain text
  • checking against a dictionary to see if syntactic errors were made

But most implementation use a TreeMap<Character, State> to store the goto structure, which costs O(lg(t)) time, t is the largest amount of a word's common prefixes. The final complexity is O(n * lg(t)), absolutely t > 2, so n * lg(t) > n. The others used a HashMap, which wasted too much memory, and still remained slowly.

I improved it by replacing the XXXMap to a Double Array Trie, whose time complexity is just O(1), thus we get a total complexity of exactly O(n), and take a perfect balance of time and memory. Yes, its speed is not related to the length or language or common prefix of the words of a dictionary.

This implementation has been widely used in my HanLP: Han Language Processing package. I hope it can serve as a common data structure library in projects handling text or NLP task.


Include this dependency in your POM. Be sure to check for the latest version in Maven Central.



Setting up the AhoCorasickDoubleArrayTrie is a piece of cake:

        // Collect test data set
        TreeMap<String, String> map = new TreeMap<String, String>();
        String[] keyArray = new String[]
        for (String key : keyArray)
            map.put(key, key);
        // Build an AhoCorasickDoubleArrayTrie
        AhoCorasickDoubleArrayTrie<String> acdat = new AhoCorasickDoubleArrayTrie<String>();;
        // Test it
        final String text = "uhers";
        List<AhoCorasickDoubleArrayTrie<String>.Hit<String>> wordList = acdat.parseText(text);

Of course, there remains many useful methods to be discovered, feel free to try:

  • Use a Map<String, SomeObject> to assign a SomeObject as value to a keyword.
  • Store the AhoCorasickDoubleArrayTrie to disk by calling save method.
  • Restore the AhoCorasickDoubleArrayTrie from disk by calling load method.

In other situations you probably do not need a huge wordList, then please try this:

        acdat.parseText(text, new AhoCorasickDoubleArrayTrie.IHit<String>()
            public void hit(int begin, int end, String value)
                System.out.printf("[%d:%d]=%s\n", begin, end, value);

or a lambda function

        acdat.parseText(text, (begin, end, value) -> {
            System.out.printf("[%d:%d]=%s\n", begin, end, value);


I compared my AhoCorasickDoubleArrayTrie with robert-bor's aho-corasick, ACDAT represents for AhoCorasickDoubleArrayTrie and Naive represents for aho-corasick, the result is :

Parsing English document which contains 3409283 characters, with a dictionary of 127142 words.
                Naive           ACDAT
time            607             102
char/s          5616611.20      33424343.14
rate            1.00            5.95
Parsing Chinese document which contains 1290573 characters, with a dictionary of 146047 words.
                Naive           ACDAT
time            319             35
char/s          2609156.74      23780600.00
rate            1.00            9.11

In English test, AhoCorasickDoubleArrayTrie is 5 times faster. When it comes to Chinese, AhoCorasickDoubleArrayTrie is 9 times faster. This test is conducted under i7 2.0GHz, -Xms512m -Xmx512m -Xmn256m. Feel free to re-run this test in TestAhoCorasickDoubleArrayTrie, the test data is ready for you.


This project is inspired by aho-corasick and darts-clone-java. Many thanks!


Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.