Suicidal Behavior Detection on Twitter Using Neural Network

In this modern age, about half of the world's population dwells on the internet. Teens are engaged very much in a virtual world such as social media like Twitter. This platform is used by teens to share their feelings, sentiments and even cryptic messages. With the growing issue of suicide at hand and as the social media sites grow in popularity, this platform is increasingly associated with cyber bullying and in some extreme cases teen suicidal. Suicide is the second leading cause of death globally among people who are 15 to 29 years of age. This study aims to develop a model for classifying suicidal tweets using Artificial Neural Network (ANN). Specially, its objective is to describe a complete process on classifying collected tweets that was gathered via Twitter Advanced Search (TAS) using suspected keywords within tweets with high potential being on a suicidal context. All in all, 5,174 tweets were gathered and in total 3,055 were English and 2,119 were Filipino or Taglish. The data gathered was then saved in an excel file for data labeling. The data collected were annotated as "non-risky" or "risky" by trained Psychologist. Based on the results of the Multi-Layer Perceptron classifier, 65% of the risky labeled data was classified correctly and 35% of the data was misclassified. On the other hand, the classifier was able to classify 91% of the non-risky tweets and only misclassifying 9%. It was found in this study using unigrams and bigrams affect the classifier as it resulted in about 75%-76% cross-validation score. However, further future improvements such as addition of data or the use of other feature transformers are necessary in order to increase the accuracy of the model.