Dissertation defense (April 28, 2021): Leonardo Ferreira dos Santos

Student: Leonardo Ferreira dos Santos

Title: Automatic identification of sexual predatory activities in virtual conversations in Brazil

Advisor: Gustavo Paiva Guedes e Silva

Committee: Gustavo Paiva Guedes e Silva (president), Eduardo Bezerra da Silva (CEFET/RJ), Eduardo Soares Ogasawara (CEFET/RJ), Ronaldo Ribeiro Goldschmidt (IME)

Day/Time:  April 28, 2021 / 14h.

Room: http://meet.google.com/vtf-bmvn-frp

Abstract: The use of the internet by children and adolescents provides access to a set of opportunities for their self-development. Access to information, educational material, entertainment, and socialization are some of the options available. The use of social networks is one of the main channels for socializing on the internet. By creating a public profile when joining the social network, children and adolescents can develop connections with other people and establish communication through virtual conversations. Sexual predators, in turn, make use of social networks to deceive these children and adolescents, establish a deceptive relationship for subsequent execution of various criminal activities, such as obtaining pornographic content, extortion, and the practice of sexual abuse. In this scenario, several studies have focused on the identification of sexual predators on the internet. Although it is a widely explored research domain, no studies considered the task of virtual conversations conducted in Brazil’s Portuguese language. Given the problem introduced, the present research has its primary objective to propose a method that offers significant results for identifying predatory activity in textual conversations carried out on the internet. A set of 82 anonymous predatory chats and criminal evidence present in judicial proceedings was considered the basis for the domain studies to understand sexual predatory activity in Brazil better. After the analysis of predatory conversations, a total of nineteen textual and behavioral characteristics identified served as the basis for creating the MDAP method. A data set with similar properties when compared to the data set of PAN-2012 competition validated the proposed method, using 82 predatory conversations as a basis. When compared to the state of the art candidate methods for the research domain, the results obtained prove the efficiency of the MDAP method for identifying predatory activity in textual conversations, presenting itself as an alternative to promote a safer virtual environment for children and adolescents.