Both comparison algorithms are based on Dice's coefficient. Each message is split into two-long overlapping parts, which are then compared. Whitespace is entirely ignored. A very simple example that illustrates the difference is calculating the similarity between "aa" and "aaaa". With the Strict method they are 50% similar and with the Lenient method they are equal.

Even with the Strict method some messages can appear equal even if they aren't, such as "aba" and "bab" (since it splits them into "ab" and "ba" and compares if they occur in both messages).