An Investigation of Explicit Indicators for Identifying Cyberstalking Incidents Towards Sexism using Keyword-Assisted Topic Model

This research explores the use of explicit indicators to identify cyberstalking incidents, a growing concern in the digital age. As the web and computer technology have grown, cyberstalking is now a major problem that affects an estimated 1.3 million people a year in the US alone. This research evaluates the comparative legitimacy of several predictors, such as those pertaining to an individual’s background, sociodemographic characteristics, risk, protective domains, and sexism, that are linked to allegations of cyberstalking and victimisation. The results draw attention to the distinctive elements that contribute to adult individuals from different nations engaging in cyberstalking regarding sexism. The outcomes of the investigation can help guide the improvement of initiatives to prevent and cope with cyberstalking, giving organisations and governments’ significant data to develop preventative plans and effectively deploy resources. In this research, keyword-assisted topic model (KATM), a machine learning-based system, has been developed to detect cyberstalking incidents. By combining keywords and attributes, KATM is able to identify potential instances of cyberstalking in online communication. The established automated topic coherence metrics, such as normalised pointwise mutual information (NPMI), UCI-coherence, and UMass-coherence across the SemEval-2023 Task 10 dataset, are used to evaluate the topics obtained from the established topic model KATM and a few other topic models as benchmark, including latent Dirichlet allocation (LDA) and embedded topic model (ETM). The evaluation outcomes reveal that KATM consistently surpasses its counterparts in terms of topic coherence, excelling in all three metrics.