How to identify spam on tumblr

gistfile1.pl

init Score;

##### most telling features #######

Score += const * 10th_decile( vector_of_likes$Date ); 

Score += const * another_functional_of( vector_of_likes$Date );

Score += const * number of likes;   #interaction term with % reblogs / number of written posts ... if the written posts start being hacked from somewhere else there are ways to identify that too

Score += const * avg_speed_of like_submission;


Score += const * is_there_an_ad_on_the_blog;


######



##### most telling feature overall: hashtag squatting #####

@hashtags = qw( tags_on most_recently_liked_post );
init squattercount;

foreach (@hashtags) {
    squattercount = number_of_likes_with the_same_hashtag;
    }


Score += const *  squattercount ;
#do this for a few randomly spread Likes

#the squattercount is the best screening mechanism, i.e. if you're figuring out which accounts to run this regression on





###### names and photos are telling #########

run SVM on photo;

run SVM on name;

if (the kernelised photo is north of the hyperplane) { Score += const; }

if (the kernelised name is north of the hyperplane) { Score += const; }

######


###### pure laziness play ##########

if (TumblrMessage == default "I love here :)") { Score += const; }
#interaction terms with default message + every other term
#interaction terms with default title + every other term

Score += concave_fn( percent of posts that are reblogs ) * const;
#actually the fn. should be nearly linear up to 80% and then starkly nonlinear monotonic, just so long as the highest %'s are differentiated




########## put it all together #######

Do a median-polish regression on the logistic of {spam, not_spam} to get decent values for the constants.