Strange behaviour of mb_detect_order() in PHP

Question

I would like to detect encoding of some text (using PHP). For that purpose i use mb_detect_encoding() function.

The problem is that the function returns different results if i change the order of possible encodings with mb_detect_order() function.

Consider the following example

$html = <<< STR
ちょっとのアクセスで落ちてしまったり、サーバー障害が多いレンタルサーバーを選ぶとあなたのビジネス等にかなりの影響がでてしまう可能性があります。特に商売をされている個人の方、法人の方は気をつけるようにしてください
STR;
mb_detect_order(array('UTF-8','EUC-JP', 'SJIS', 'eucJP-win', 'SJIS-win', 'JIS', 'ISO-2022-JP','ISO-8859-1','ISO-8859-2'));
$originalEncoding = mb_detect_encoding($str);
die($originalEncoding); // $originalEncoding = 'UTF-8'

However if you change the order of encodings in mb_detect_order() the results will be different:

mb_detect_order(array('EUC-JP','UTF-8', 'SJIS', 'eucJP-win', 'SJIS-win', 'JIS', 'ISO-2022-JP','ISO-8859-1','ISO-8859-2'));        
die($originalEncoding); // $originalEncoding = 'EUC-JP'

So my questions are:
Why is that happening ?
Is there a way in PHP to correctly and unambiguously detect encoding of text ?

Answer 1

That's what I would expect to happen.

The detection algorithm probably just keeps trying, in order, the encodings you specified in mb_detect_order and then returns the first one under which the bytestream would be valid.

Something more intelligent requires statistical methods (I think machine learning is commonly used).

EDIT: See e.g. this article for more intelligent methods.

Due to its importance, automatic charset detection is already implemented in major Internet applications such as Mozilla or Internet Explorer. They are very accurate and fast, but the implementation applies many domain specific knowledges in case-by-case basis. As opposed to their methods, we aimed at a simple algorithm which can be uniformly applied to every charset, and the algorithm is based on well-established, standard machine learning techniques. We also studied the relationship between language and charset detection, and compared byte-based algorithms and character-based algorithms. We used Naive Bayes (NB) and Support Vector Machine (SVM).