redirect 404 to similar urls

Oh man, oh man!

What you're asking for is not simple and need you to have a powerful computer, but the results are simply amazing.

Here's what I'd suggest to do:

• For 404 proper handling, you have the ErrorDocument redirection in vhost configuration. Mine looks like this: ErrorDocument 404 /404.php ;
• When having a 404, Apache will call /404.php with all the arguments (which bad URL and so on, dump $_SERVER to see this). You have to test if there are only two expressions in the URL / i.e. http://mysite.com/(expr1)/(expr2)/
• If not, then do a classical 404.
• If yes then do a SOUNDEX search with MySQL (in your 404 Php file). See query sample here.
• Then, in this "special" 404 case, do a suggestion, like google does, i.e.: "did you mean /action/story-name-action/? if so, click on the link".

This a hard work, but it's both interesting and shows your skill. Very few websites do this (I just know google actually).

Here's a demo on my French table that could give you an overview of how it works:

mysql> SELECT * FROM job WHERESOUNDEX( description ) LIKE SOUNDEX('Machiniste cinéma');+-------+--------------------+| id    | description        |+-------+--------------------+| 14018 | Machiniste cinéma  |+-------+--------------------+1 row in set (0.06 sec)mysql> SELECT * FROM job WHERESOUNDEX( description ) LIKE SOUNDEX('Mchiniste cinéma');+-------+--------------------+| id    | description        |+-------+--------------------+| 14018 | Machiniste cinéma  |+-------+--------------------+1 row in set (0.06 sec)mysql> SELECT * FROM job WHERESOUNDEX( description ) LIKE SOUNDEX('Machnste cinema');+-------+--------------------+| id    | description        |+-------+--------------------+| 14018 | Machiniste cinéma  |+-------+--------------------+1 row in set (0.06 sec)mysql> 

Unless you are very sure of the URL the user really wanted to navigate to, using rewrite / redirecting to a specific URL is a very bad idea.

Taking your example, suppose you want to handle every case where two letters may have been dropped, with 17 characters in the last part of the URL, that's 17*16 = 272 combinations, while it may be possible to match multiple 'false' urls with one regex, you're stil going to need a lot of rewrite rules.

A better solution would be, to implement 404 handler using PHP (since you included that tag in your q), to generate a list of (say) the top 10 URLs whose paths have the shortest levenstein distance from the requested path, along with a default link and supporting text. (There are mysql based implementations - try Google for URLs). NB handler should still return a 404 status - NB HTML content must be more than a minimum length to suppress MSIE's 'friendly' error message.

If you know what the possible correct URLs could be, you can use:

levenshtein($givenURL, $possibleURL)

Example from PHP docs, comments removed for brevity:

$input = 'carrrot';$words  = array('apple','pineapple','banana','orange',                'radish','carrot','pea','bean','potato');$shortest = -1;foreach ($words as $word) {    $lev = levenshtein($input, $word);    if ($lev == 0) {        $closest = $word;        $shortest = 0;        break;    }    if ($lev <= $shortest || $shortest < 0) {        $closest  = $word;        $shortest = $lev;    }}echo $shortest == 0 ? "Exact match found: $closest\n" : "Did you mean: $closest?\n";

Outputs:

Input word: carrrot
Did you mean: carrot?

This is good when you think people may have omitted a letter or put an extra one in, but it may fall short when people genuinely don't know how to spell a word and came up with something creative!

If you prefer the soundex() route, take a look at the metaphone() function.

I like the idea of using metaphone() alongside levenshtein() or similar_text(), as it returns a phonetic representation of the word, and you still want to see how similar it is to your original.

Examples:

metaphone('name') = NMmetaphone('naaaaaameeeeeeee') = NMmetaphone('naiym') = NMmetaphone('naiyem') = NYM

While a lot of misspellings will return an identical match, the last example shows that you really still want to find the closest match with something like levenshtein()

For efficiency, if you use a different 404 file where the rewrites tried to match this pattern and failed, than you use for the rest of the site, it really shouldn't really be a massive overhead.

If you're getting the same 404 from the same referrer a lot, (and can't get them to change the link) it might be worth just putting a static rewrite in for that case.