click here to download the abstract project
click here to download the base paper
ABSTRACT
Due to multiple reasons like the nuclear family, peer pressure for fake prestige, impatience attitude, and mental stress has become a common trait in every person. With advancements in technology like the
internet and online presence, it has become a routine to be active online. Some sections of people vent out their emotions online as they have no support system in real life. It has been detected, as seen in
some instances; those suicidal tendencies ranging from mild to extreme could be from a person’s online profile activity. In our current work, we use a specific method that includes all critical criteria that could be
exhibited by a suicidal person by using Natural Language Processing (NLP) techniques. NLP interprets written language, whereas Machine Learning makes predictions based on patterns learned from experience. These textual features are passed through a robust Machine Learning framework for detecting an abrupt change in input data. Our method predicts efficiently a genuine, mentally disturbed
profile from a typical profile.
INTRODUCTION
At the moment, the task of preventing suicide is relevant, because according to statistics of the World Health Organization every year more than 800 000 people commit suicide. According to the same data, the Russian Federation is among the five countries with the highest number of suicides per 100,000
inhabitants. Since the Internet is the easiest way to distribute suicidal content in the form of various web pages dedicated to suicide, a large number of organizations are trying to solve this problem, for example, the popular social network Facebook, which detects suspicious profiles of users prone to suicide, and posts containing suicidal content. In addition to social networks, federal services are trying to solve the problem, for example, in Russia Roskomnadzor in 2016 published recommendations for disseminating
information about suicide cases in the media, which probably affects the results of search engines on this topic. In addition, since 2006 there has been a unified register that contains websites blocked in the Russian Federation. However, blocking does not happen immediately. Some people manage to visit dangerous web pages. Manual blocking of suicidal websites can hardly be called an effective measure in the fight against the spread of suicidal content. Since after several years of such blockages, the number of child suicides in Russia has risen sharply again. In addition to the «death groups» in social networks that platform developers are already actively fighting, one of the reasons may be that websites with suicidal content can create their own copies (mirrors). This article discusses the possibility of detecting such web
pages by analyzing their content in real time using machine learning algorithms. Detection occurs on the client’s side. In this way, with sufficient accuracy to identify dangerous websites visited by the user, it is possible to identify a person who is suicidal at an early stage.
