Implements the Aho-Corasick string search algorithm


Keywords
aho, aho-corasick, corasick, match, search, string, substring, multi-string, sql-clr, sql-server, string-search
License
MIT
Install
Install-Package AhoCorasick -Version 2.0.258

Documentation

AhoCorasick

Version Build status Coverage Status netstandard2.0 net40

This is an implementation of the Aho-Corasick string matching algorithm for .NET (netstandard2.0 and net40) and SQL Server (SQL CLR). Mostly ported from xudejian/aho-corasick in CoffeeScript.

Usage

var ac = new AhoCorasick("a", "ab", "bab", "bc", "bca", "c", "caa");
var results = ac.Search("abccab").ToList();

Assert.AreEqual(0, results[0].Index); // index into the searched text
Assert.AreEqual("a", results[0].Word); // matched word
// ...

or

var results = "abccab".Contains("a", "ab", "bab", "bc", "bca", "c", "caa").ToList();

Custom char comparison

You can optionally supply an IEqualityComparer<char> to perform custom char comparisons when searching for substrings. Several implementations with comparers that mirror StringComparer are included.

var results = "AbCcab".Contains(CharComparer.OrdinalIgnoreCase, "a", "ab", "c").ToList();

SQL CLR Functions

There are also several SQL CLR user defined functions that can be used to perform fast substring matching in Microsoft SQL Server. To use this:

  1. Make sure you have enabled CLR integration
  2. Execute AhoCorasick.SqlClr_Create.sql

For one-off queries, you can use the functions that rebuild the trie on each query, e.g.

select top(100) * from Posts P
where dbo.ContainsWords((select Word from Words for xml raw, root('root')), P.Body, 'o') = 1

The words to match are always supplied as XML where the values are taken from the first attribute of all elements directly beneath the root node. Be careful to select the word column as the only or first column otherwise you'll end up matching the wrong words. The XML in the example above looks like this:

<root>
  <row Word="Aachen" />
  <row Word="Aaliyah" />
  <row Word="aardvark" />
  ...
</root>

Here's more about FOR XML.

The last parameter in the function indicates the culture to use since there is no way to use SQL Server collations in SQL CLR code. Values can be:

Value Character comparison
c Current Culture
n Invariant Culture
o or Empty Ordinal
Culture name, e.g. "de-de" Specific .NET Culture

The culture identifier can be suffixed by :i indicating case-insensitive matching.

Static objects

The function in the example above has the problem that the trie is rebuilt for each query even though the input always stays the same. To overcome this problem, there are a number of functions to manage the creation and destruction of static objects whose handles can be saved in SQL variables. Example:

declare @ac nvarchar(32);
set @ac = dbo.CreateAhoCorasick((select Word from Words for xml raw, root('root')), 'en-us:i');
select * from Posts P
where dbo.ContainsWordsByObject(P.Body, @ac) = 1;

This is a lot faster than the first example because the trie is created only once and then reused for each row in the query. The handle (@ac) is a hash value generated from the words to match and the culture. The corresponding object is saved in a static dictionary. You can list the currently active objects using dbo.ListAhoCorasick(), remove all objects using dbo.ClearAhoCorasick() or remove only one object using dbo.DeleteAhoCorasick(@ac).

Getting all matches

The examples above only checked if the words occurred in the queried texts. If you want to get the matched words and the indexes where they occur in the queried texts you can use the supplied table-valued functions. For example:

declare @ac nvarchar(32);
set @ac = dbo.CreateAhoCorasick((select Word from Words for xml raw, root('root')), 'o');
select top(100) * from Posts P
cross apply dbo.ContainsWordsTableByObject(P.Body, @ac) W

This will return a table such as this:

ID Body Index Word
1 What factors related... 5 factor
1 What factors related... 6 actor
1 What factors related... 5 factors
...

Word boundaries

There are also functions that return only matches occuring at word boundaries: dbo.ContainsWordsBoundedByObject() and dbo.ContainsWordsBoundedTableByObject(). Word boundaries here are the same as \b in regexes, i.e. matches will occur as if words were specified as \bword\b.

Forcing parallelism

Although these kinds of queries lend themselves very well to parallel execution, SQL Server tends to overestimate the cost of parallel queries and builds non-parallel plans most of the time where user defined functions are involved. You can force a parallel plan by using a trace flag (more about this here):

declare @ac nvarchar(32);
set @ac = dbo.CreateAhoCorasick((select Word from Words for xml raw, root('root')), 'en-us:i');
select * from Posts P
where dbo.ContainsWordsBoundedByObject(P.Body, @ac) = 1
OPTION (RECOMPILE, QUERYTRACEON 8649)

Parallel operators are identified by a yellow badge with two arrows in the query plan.

Performance

Here's a benchmark searching for ~5000 words (average length 7) in ~250,000 texts (average length ~900):

SQL AhoCorasick
560s 7s

The SQL query used was this:

select * from Posts P
where exists (select * from Words W where CHARINDEX(W.Word, P.Text) > 0)

But I can simply use full-text search

No. The CONTAINS predicate can only search for a single literal or variable at a time. You can't use it in a join or subquery to search for a column value of a table in the query, i.e. this won't work:

select * from Posts P
where exists (select * from Words W where CONTAINS(P.Text, W.Word))

If you know of a way to make this work using FTS (perhaps using a cursor?) let me know.