Model design to look for patterns related to suicide in social networks

Abstract

Many people with suicidal ideation use social forum platforms to post or discuss information about this complex topic. The key objective of our study is to design and evaluate a model to find patterns linguistically related to suicide. We address the detection of suicidal ideation through machine learning by applying it to the social network Twitter. To do this, we use different linguistic processors to obtain characteristics of each tweet and then catalog them using unsupervised classifiers. Finally, this information is used by 7 types of supervised learning (Naive Bayes, KNN, MLP, SVM, Decision Tree, Adaboost y Random Forest) and perform a comparative analysis of the classifiers using evaluation parameters, mainly accuracy. Our experiment shows 42 classification results, as well as sequential and parallel processing time data from the best-supervised machine learning, Random Forest.