Sentiment Analysis using Social and Topic Context for Suicide Prediction

In many fields, analysing large user-generated microblogs is very crucial and drawing many researchers to study. However, processing such short and noisy microblogs is very difficult and challenging. Most prior studies use only texts to find the polarity of sentiment and presume that microblog site is independent and distributed identically, ignoring networked data from microblogs. Consequently, not satisfied with performance motivated by emotional and sentimental sociological approaches. This paper proposes a new methodology that incorporates social and topic context to analyze sentiment on microblogs by introducing the meaning of structure similarity into social context. Unlike from previous research employing direct relations from user and by suggesting a new method to quantify structure similarity. In addition, to design the microblog semantic relation, topic context is introduced. The Laplacian matrix of these graph produced by these context combines social and topic context and Laplacian regularization is applied to the microblogging sentiment model. The Experimental results on the two datasets show that, the suggested model had reliably and substantially outperformed the baseline methods that is helpful for suicide prediction.