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(ln(t))
time, t
is the largest amount of a word's common prefixes. The final complexity is O(n * ln(t))
, absolutely t > 2
, so n * ln(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.
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[]
{
"hers",
"his",
"she",
"he"
};
for (String key : keyArray)
{
map.put(key, key);
}
// Build an AhoCorasickDoubleArrayTrie
AhoCorasickDoubleArrayTrie<String> acdat = new AhoCorasickDoubleArrayTrie<String>();
acdat.build(map);
// 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 aSomeObject
as value to a keyword. - Store the
AhoCorasickDoubleArrayTrie
to disk by callingsave
method. - Restore the
AhoCorasickDoubleArrayTrie
from disk by callingload
method.
In other situations you probably do not need a huge wordList, then please try this:
acdat.parseText(text, new AhoCorasickDoubleArrayTrie.IHit<String>()
{
@Override
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 repesents 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
http://www.apache.org/licenses/LICENSE-2.0
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.