Aaron Shear has an interesting post discusses 950 penalty which he finds to have "a direct correlation with a seemingly coincidental patent that was recently released." He says that:
if you read through this patent it seems to me that it was designed to detect and penalize content generated from basically a content generation system. The sole purpose of these content generators is to mix-up content and make it seem themed and relevant for search engines.
He also produces an excerpt from the patent:
An example of the cluster bit vectors are as follows, using the above phrases: TABLE-US-00001 Monica purse Cluster Bill Clinton President Lewinsky designer ID Bill Clinton 1 1 1 0 14 President 1 1 0 0 12 Monica 1 0 1 1 11 Lewinsky purse 0 0 1 1 3 designer.
This leads Aaron to comment that:
Google could take this sample and run it against other known examples of content that are themed. If this particular sample is so far beyond the normal threshold it could easily trigger such a filter. The most interesting part of spam is usually it is targeted to mostly competitive subjects, thus the sample size is very large and easy to target.