HTML::Diff Perl module compares two strings of HTML and returns a list of a chunks which indicate the diff between the two input strings, where changes in formatting are considered changes.
HTML::Diff does not strictly parse the HTML. Instead, it uses regular expressions to make a decent effort at understanding the given HTML. As a result, there are many valid HTML documents for which it will not produce the correct answer. But there may be some invalid HTML documents for which it gives you the answer you're looking for. Your mileage may vary; test it on lots of inputs from your domain before relying on it.
$result = html_word_diff($left_text, $right_text);
Returns a reference to a list of triples [< flag >, < left >, < right >]. Each triple represents a check of the input texts. The flag tells you whether it represents a deletion, insertion, a modification, or an unchanged chunk.
Every character of each input text is accounted for by some triple in the output. Specifically, Concatenating all the < left > members from the return value should produce $left_text, and likewise the < right > members concatenate together to produce $right_text.
The is either 'u', '+', '-', or 'c', indicating whether the two chunks are the same, the $right_text contained this chunk and the left chunk didn't, or vice versa, or the two chunks are simply different. This follows the usage of Algorithm::Diff.
The difference is computed on a word-by-word basis, "breaking" on visible words in the HTML text. If a tag only is changed, it will not be returned as an independent chunk but will be shown as a change to one of the neighboring words. For balanced tags, such as < b > < /b >, it is intended that a change to the tag will be treated as a change to all words in between.