The document discusses detecting mental disorders on Twitter. It proposes a framework to collect tweets from community portals and patients with bipolar disorder or borderline personality disorder. Features are extracted from tweets, including TF-IDF, LIWC, and pattern of life features. Classifiers like random forest are trained and evaluated on the data using 10-fold cross validation and limited data tests. The results show pattern of life models can correctly detect disorders with high precision and low selection bias compared to TF-IDF and LIWC models.