A Novel Multi-Class Technique for Suicide Detection in Twitter Dataset

Abstract

Suicide is not an individual phenomenon, but it is also affected by factors of social and environmental significance, it is a serious problem for public health around the globe. Unlike conventional retrospective studies carried out via self-reported surveys and questionnaires, the reliable identification of suicidal symptoms from Twitter tweets attained for more than a year period using different online web-blogging sites as reference are analysed in this paper. To recognize tweets containing suicidal thoughts, three set of features are used to train the dataset by using base and ensemble classifiers; proposed baseline Rotation Forest (NROF) algorithm and Maximum probability voting decision method is applied on seven different labelled classes related to suicide communication and class exhibiting suicidal ideation, this enhanced model achieved an F-measure: 0.82 (including suicidal contents for all seven classes) and 0.79 for suicidal ideation class. The results obtained are summarized by focusing on predictive principal components of classes with suicidal communication to offer awareness to the languages that are used to expel suicidal ideation in Twitter.