This is actually one of the most interesting things I've read here in years. As a former software developer, I'm fascinated because it seems to suggest that something went wrong. Maybe an attempt at 'optimizing' the keywords - by running an algorithm over all the content - that went haywire and did some damage, which would be impossibly expensive to find and fix.
The fact that the word is 'marijuana' is doubly interesting because as we know, there's at least one guy who's somehow gotten 10s of thousands of bogus-looking pot photos into the database. And it's pretty obvious they weren't individually inspected and approved in any honest way. Maybe content like this is being back-doored in by means of database scripts - possibly on the sly - and the scripts occasionally misfire and touch other content.
I'm just speculating of course, but it's hard to come up with a reasonable explanation. Will be interesting to see if other reports come in.