Sensitive Content Recognition in Social Interaction Messages

Online social networks are a predominant medium for social interaction where people communicate in a way similar to what they do in real life. User communication comes mainly in forms of textual data which are rich in personal information, opinions and sentiments. The automatic recognition of sensitive content in texts is quite important for a number of reasons. In this work, we study the dimensions of sensitive content recognition and we examine the performance of various machine learning methods for sensitive data recognition in text. Understanding the key features of sensitive content can assist in formulating more efficient user-centric interaction frameworks too that secure users’ privacy, promote users' inclusion and enhance the diversity awareness of the online society. Also, another part of this work focuses on the models’ explainability where the integration of LIME and SHAP offer insight on features that are consistent and robust predictors of sensitive content