String::Trigram version 0.11 ==================== INSTALLATION To install this module type the following: perl Makefile.PL make make test make install DEPENDENCIES none =============================================================================== NAME String::Trigram - Find similar strings by trigram (or 1, 2, 4, etc.-gram) method SYNOPSIS use String::Trigram; Object Oriented Interface my @cmpBase = qw(establishment establish establishes established disestablish disestablishmentarianism); my $trig = new String::Trigram(cmpBase => \@cmpBase); my $numOfSimStrings = $trig->getSimilarStrings("establishing", \%result); print "Found $numOfSimStrings similar strings.\n"; foreach (keys %result) { print "Similar string $_ has a similarity of ", sprintf ("%02f", ($result{$_} * 100)), "%\n"; } Functional Interface my $string1 = "foo"; my $string2 = "boo"; my $smlty = String::Trigram::compare($string1, $string2); print "$string1 and $string2 have a similarity of ", sprintf("%.2f", ($smlty * 100)), "%.\n"; $smlty = String::Trigram::compare($string1, $string2); print "$string1 and $string2 have a similarity of ", sprintf("%.2f", ($smlty * 100)), "%.\n"; DESCRIPTION This module computes the similarity of two strings based on the trigram method. This consists of splitting some string into triples of characters and comparing those to the trigrams of some other string. For example the string kangaroo has the trigrams "{kan ang nga gar aro roo}". A wrongly typed kanagaroo has the trigrams "{kan ana nag aga gar aro roo}". To compute the similarity we divide the number of matching trigrams (tokens not types) by the number of all trigrams (types not tokens). For our example this means dividing 4 / 9 resulting in 0.44. To balance the disadvantage of the outer characters (every one of which occurs in only one trigram - while the second and the penultimate occur in two and the rest of the characters in three trigrams each) somewhat we pad the string with blanks on either side resulting in two more trigrams ' ka' and 'ro ', when using a padding of one blank. Thus we arrive at 6 matching trigrams and 11 trigrams all in all, resulting in a similarity value of 0.55. When using the trigram method there is one thing that might appear as a problem: Two short strings with one (or two or three ...) different trigrams tend to produce a lower similarity then two long ones. To counteract this effect, you can set the module's "warp" property. If you set it to something greater than 1.0 (try something between 1.0 and 3.0, flying at warp 9 won't get you anywhere here), this will lift the similarity of short strings more and the similarity of long strings less, resulting in the '%%%' curve in the (schematical) diagram below. 1.0 simi- | % * % larity | % * # # value | % * # | * # | % * # | * # | * | % * # | | * | % # | * | *** no warp (i.e. warp == 1.0) | %# %%% warp > 1 | * ### warp < 1 |________________________________________________________ 0.0 length of string Dependency of similarity value on length of string and warp Don't hesitate to use this feature, it sometimes really helps generating useful results where you otherwise wouldn't have got any. Please be aware of that a "warp" less than 1.0 will result in an inverse effect pulling down the similarity of short strings a lot and the similarity of long ones less, resulting in the '###' curve. I have no idea what this can be good for, but it's just a side effect of the method. How is all this done? Take a look at the code. Splitting strings into trigrams is a time consuming affair and if you want to compare a set of n strings to another set of m strings and you do it on a member to member base you will have to do n * m splittings. To avoid this, this module takes a set of strings as the base of comparison and generates an index of every trigram occuring in any of the members of the set (including the information, how often the trigram occurs in a given member string). Then you can feed it the members of the other set one by one. This results in an amount of n + m splitting plus the overhead from generating the index. This way we save a lot of time at the expense of memory, so - if you operate on a great amount of strings - this might turn out to be somewhat of a problem. But there you are. There's no such thing as a free lunch. Anyway - the module is optimized for comparisons of sets of string which results in single comparisons being slower than it might be. So, if you use the "compare()" function which compares single strings in a functional interface, to be able to use the full functionality of the module and not to get into the need to program same things twice, internally a String::Trigram object is instantiated and an index of the trigrams of one of the strings is generated. In practice however this shouldn't be a big disadvantage since a single comparison or just a few won't need too much (absolute) time. METHODS new my @base = qw(chimpanzee lion emu kangaroo); my $trig = new String::Trigram(cmpBase => \@base); This is the constructor. Before we are able to do any computing of similarities, it will want the parameter "cmpBase" to point to a reference to an array of strings (the base of comparison). Everything else is taken care of unless you want to change the defaults: my $trig = new String::Trigram(cmpBase => \@base, minSim => 0.5, warp => 1.6, ignoreCase => 0, keepOnlyAlNums => 1, ngram => 5, debug => 1); PARAMETERS: * "cmpBase" - Reference to array containing strings for base of comparison. * "minSim" - Minimal similarity you are prepared to accept. Specify a value between 0 (not similar at all) and 1 (identity). Any string matching with less (equals) than "minSim" will not be returned in "getSimilarStrings()" and not counted as a match in "getBestMatch()", even if it is the only one. Default is 0, so anything even remotely matching will be returned. * "ngram" - If you do not want to use trigrams, but some other n in n-gram, use this parameter, for example ngram => 5 for 5-grams. * "warp" - Set warp attribute (see description). Default is 1.0 (no warp). * "ignoreCase" - Just that. Default is 1 (do ignore case) * "keepOnlyAlNums" - Remove any \W character before comparison. Default is 0 (keep every character). * "padding" - Set the number of blanks for padding string (see above). The number has to be between 0 and n-1 (n from n-gram). Default is n-1. * "debug" - print some debugging information to STDERR CROAKS IF ... * it gets an unknown parameter * ngram is 0 * parameter cmpBase does not point to a reference to an array * minimal similarity is out of bounds (should be 0 <= minSim <= 1) * warp is out of bounds (0 <= warp) * padding is out of bounds (should be 0 <= padding <= n-1) reInit $trig->reInit(["zebra", "tiger", "snake", "gorilla", "kangaroo"]); Give the object a new base of comparison (deleting the old one). CROAKS IF ... * parameter does not point to a reference to an array extendBase $trig->extendBase(["zebra", "tiger", "snake", "gorilla", "kangaroo"]); Add strings to object's base of comparison. CROAKS IF ... * parameter does not point to a reference to an array minSim $trig->minSim(); $trig->minSim(0.8); Get or set minimal accepted similarity (see above). CROAKS IF ... * minimal similarity is out of bounds (should be 0 <= minSim <= 1) warp $trig->warp(); $trig->warp(1.4); Get or set warp (see above). CROAKS IF ... * warp is out of bounds (should be 0 <= warp) ignoreCase $trig->ignoreCase(); $trig->ignoreCase(0); Get or set ignoreCase property (see above). keepOnlyAlNums $trig->keepOnlyAlNums(); $trig->keepOnlyAlNums(1); Get or set keepOnlyAlNums property (see above). padding $trig->padding(); $trig->padding(2); Get or set padding property (see above). CROAKS IF ... * padding is out of bounds (should be 0 <= padding <= n-1) debug $trig->debug(); $trig->debug(1); Get or set debug property. For debugging to STDERR, set to 1. getSimilarStrings my %results = (); my $numOfSimStrings = $trig->getSimilarStrings("zebrilla", \%results [, minSim => 0.6, warp => 0.7]); Get similar strings for first parameter from base of comparison. The result is saved in the second parameter (a reference to a hash), the keys being the strings and the values the similarity values. The method returns the number of found similar strings. If parameters minSim or warp are defined, those values are changed temporalily. CROAKS IF ... * second parameter is not a reference to a hash getBestMatch my @bestMatches = (); my $sim = $trig->getBestMatch("zebrilla", \@bestMatches [, minSim => 0.6, warp => 0.7]); Don't bother about all those more or less similar strings, just get the best one. This might actually be more than one, since several strings might result in the same similarity value. So the second parameter is a reference to an array, taking the best similar strings, the first parameter is the string to compare. Returns similarity of best match or 0 if there are no similar strings at all (please observe that in the case of no match the return value was -1 up to $VERSION == 0.02). If parameters minSim or warp are defined, those values are changed temporarily. CROAKS IF ... * second parameter is not a reference to an array FUNCTIONS compare my $sim = compare($string1, $string2); or my $sim = compare($string1, $string2, minSim => 0.3, warp => 1.8, ignoreCase => 0, keepOnlyAlNums => 1, ngram => 5, debug => 1); Use this if you don't want use the oo- interface. Returns resulting similarity. Note that this is not a very fast way to use the module, if you do a lot of comparisons, since internally for every call to compare() a new Trigram object is initialized (and "DESTROY"ed as it goes out of scope). CROAKS IF ... [same as using new() and getSimilarStrings()] EXPORT_OK compare() AUTHOR Tarek Ahmed, SEE ALSO * "String::Similarity" - Uses edit scripts to compute similarity between strings. * "String::Approx" - Uses Levenshtein edit distance to compute similarity between strings. * "Text::Soundex" - Uses soundex method to compute similarity between strings. For an early description of the method, see: R.C. Angell, G.E. Freund, and P. Willet. Automatic spelling correction using a trigram similarity measure. Information Processing and Management, 19(4):255--261, 1983. COPYRIGHT AND LICENCE Copyright (c) 2003 Tarek Ahmed. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.