The Impact of Assigning Weights to Semantic Resources on the Identification of Criminal Suspects on Social Media

Abstract

The identification of criminal suspects on social media has been a topic of great relevance in the analysis of this type of media. Most of the time, the methods that seek to identify these suspects use textual data made available by people on these networks (e.g. messages, comments, among others). To analyze the texts, these methods often use semantic resources such as controlled vocabularies or even simple sets composed of terms, according to the domain in question (e.g. terrorism, pedophilia, among others). The mention of one or more of these terms can raise suspicions about the people who have used them. However, some terms raise more suspicion than others. Therefore, this work seeks to investigate the impact of differentiating the level of dangerousness of the terms used by a method for identifying criminal suspects on social media and whether this can lead to better results in identifying suspects. The results obtained through experiments in the domain of pedophilia showed that differentiating the level of dangerousness of the terms provided better results in 82.5% of the experiments carried out.