Identifying the High Risk Social Media Messages by Analyzing Text Information

Mene Bagudu, Dr. Soo-Yeon Ji, College of Arts and Science, Bowie State University, 1400 Jericho Park Rd, Bowie, MD 20715

As technology is getting advanced, new tools and methodologies that have significant societal influences on people communicating through social media (i.e., Instagram, Twitter, Facebook,etc. ) have developed. Although social media offers a convenient and useful way to connect people, it negatively influences children and teenagers and often causes significant antisocial behavior such as cyberbullying and cybercrime. Due to the negative influences, children and teenagers may suffer from emotional, mental, and behavioral problems. To address the issues, detecting suspicious anti-social behaviors as early as possible is critical. Thus, this research aims to design a method to identify whether text information in cyberspace contains negatively influential messages. Specifically, this research contains two tasks as its features: extraction and classification. The feature extraction is performed to determine significant characteristics for distinguishing influential textual information. As an initial step, numerical representations of text information are performed with a commonly used Bag-of-words model. There are emerging research studies on the relationship between emotion and the negative influences behavior cause through emotions such as anger, sadness, and frustration which may play an important role in understanding antisocial behaviors. Thus, we measure the frequency of the emotionality (positive, negative, and neutral) of words from each conversation. Then, we will perform a classification to generate a predictive model using known machine learning techniques such as support vector machine, logistic regression, and Bayesian networks. So far, we collected publicly available social media datasets and are currently working on extracting features. This study shows a method of identifying potential risks of conversation in cyberspace by constructing a predictive model with optimal accuracy by comparing their performances. We anticipate that the results of our research will be able to produce an effective model for detecting suspicious conversations via text.

Additional Abstract Information

Presenter: Mene Bagudu

Institution: Bowie State University

Type: Poster

Subject: Computer Science

Status: Approved

Time and Location

Session: Poster 5
Date/Time: Tue 12:30pm-1:30pm
Session Number: 4004