OsPreyOnline Predatory Conversation Detection
Standing
Undergraduate
Type of Proposal
Oral Research Presentation
Challenges Theme
Open Challenge
Faculty Sponsor
N/A
Proposal
Online sexual abuse is becoming increasingly prevalent in modern society due to the increasing number of children on social media platforms and the availability of web applications such as online chatrooms and multiplayer games. These platforms can be a breeding ground for predatory behavior, which is why detecting such conversations is crucial to catching predators and protecting children. Previously, researchers have used Artificial intelligence (AI), Machine Learning(ML), and Natural Language Processing(NLP) to identify explicit remarks and patterns in conversations that are indicative of predatory behavior. However, these existing studies have limitations, including a lack of data and the failure to consider the context of the conversations. To overcome these limitations, we propose a new approach that incorporates contextual information such as the number of online users and the time of messages in a conversation. This helps improve the accuracy of AI-ML algorithms. We also employ a sampling approach to tackle the imbalance in datasets that label the majority of conversations as non-predatory. This approach has been shown to significantly improve the performance of existing AI-ML algorithms in detecting online predatory conversations. We aim to train machine learning models with a high recall to minimize the chance of missing instances of predatory behavior. This new approach could be an important step forward in the development of the prevention of online sexual behaviors and make social media a safer place for children.
Grand Challenges
Viable, Healthy and Safe Communities
OsPreyOnline Predatory Conversation Detection
Online sexual abuse is becoming increasingly prevalent in modern society due to the increasing number of children on social media platforms and the availability of web applications such as online chatrooms and multiplayer games. These platforms can be a breeding ground for predatory behavior, which is why detecting such conversations is crucial to catching predators and protecting children. Previously, researchers have used Artificial intelligence (AI), Machine Learning(ML), and Natural Language Processing(NLP) to identify explicit remarks and patterns in conversations that are indicative of predatory behavior. However, these existing studies have limitations, including a lack of data and the failure to consider the context of the conversations. To overcome these limitations, we propose a new approach that incorporates contextual information such as the number of online users and the time of messages in a conversation. This helps improve the accuracy of AI-ML algorithms. We also employ a sampling approach to tackle the imbalance in datasets that label the majority of conversations as non-predatory. This approach has been shown to significantly improve the performance of existing AI-ML algorithms in detecting online predatory conversations. We aim to train machine learning models with a high recall to minimize the chance of missing instances of predatory behavior. This new approach could be an important step forward in the development of the prevention of online sexual behaviors and make social media a safer place for children.