Mega-simple Naive Bayes classification of posts, using NLTK's built in NaiveBayesClassifier.
Features are binary word occurrences. Posts are tokenized and lemmatized, and stopwords are removed, all using standard NLTK/wordnet modules.
Performance in 10-fold cross validation classifying posts flagged with self-harm/suicide content vs. unflagged posts:
Fold 0
Accuracy = 0.9067796610169492
Precision = 0.4901185770750988
Recall = 0.6424870466321243
F-score = 0.5560538116591929
Type 1 err = 0.5098814229249012
Type 2 err = 0.03687867450561197
##########
Fold 1
Accuracy = 0.8865348399246704
Precision = 0.42356687898089174
Recall = 0.689119170984456
F-score = 0.52465483234714
Type 1 err = 0.5764331210191083
Type 2 err = 0.03314917127071823
##########
Fold 2
Accuracy = 0.8945386064030132
Precision = 0.45110410094637227
Recall = 0.7409326424870466
F-score = 0.5607843137254902
Type 1 err = 0.5488958990536278
Type 2 err = 0.02767017155506364
##########
Fold 3
Accuracy = 0.880357983984927
Precision = 0.4109195402298851
Recall = 0.7447916666666666
F-score = 0.5296296296296297
Type 1 err = 0.5890804597701149
Type 2 err = 0.0276056338028169
##########
Fold 4
Accuracy = 0.8813000471031559
Precision = 0.4166666666666667
Recall = 0.78125
F-score = 0.5434782608695653
Type 1 err = 0.5833333333333334
Type 2 err = 0.0238230289279637
##########
Fold 5
Accuracy = 0.8492699010833726
Precision = 0.3476190476190476
Recall = 0.7604166666666666
F-score = 0.47712418300653586
Type 1 err = 0.6523809523809524
Type 2 err = 0.02701115678214915
##########
Fold 6
Accuracy = 0.83843617522374
Precision = 0.3295711060948081
Recall = 0.7604166666666666
F-score = 0.4598425196850394
Type 1 err = 0.6704288939051919
Type 2 err = 0.02738095238095238
##########
Fold 7
Accuracy = 0.8704663212435233
Precision = 0.38108882521489973
Recall = 0.6927083333333334
F-score = 0.4916820702402958
Type 1 err = 0.6189111747851003
Type 2 err = 0.03325817361894025
##########
Fold 8
Accuracy = 0.8657560056523788
Precision = 0.3798449612403101
Recall = 0.765625
F-score = 0.5077720207253886
Type 1 err = 0.6201550387596899
Type 2 err = 0.025921658986175114
##########
Fold 9
Accuracy = 0.8496701225259189
Precision = 0.3432098765432099
Recall = 0.7239583333333334
F-score = 0.4656616415410385
Type 1 err = 0.6567901234567901
Type 2 err = 0.030867792661619105