Try using pentagrams rather than trigrams for copyvio Markov chains.

CHANGELOG

@@ -15,7 +15,8 @@ v0.2 (unreleased):
 - Added copyvio detector functionality: specifying a max time for checks;
   improved exclusion support. URL loading and parsing is parallelized to speed
   up check times, with a multi-threaded worker model that avoids concurrent
-  requests to the same domain. Fixed assorted bugs.
+  requests to the same domain. Improvements to the comparison algorithm. Fixed
+  assorted bugs.
 - Added support for Wikimedia Labs when creating a config file.
 - Added and improved lazy importing for various dependencies.
 - Fixed a bug in job scheduling.

earwigbot/wiki/copyvios/markov.py

@@ -30,7 +30,7 @@ class MarkovChain(object):
     """Implements a basic ngram Markov chain of words."""
     START = -1
     END = -2
-    degree = 3  # 2 for bigrams, 3 for trigrams, etc.
+    degree = 5  # 2 for bigrams, 3 for trigrams, etc.
 
     def __init__(self, text):
         self.text = text