[Paper Review] Analyzing Negative User Behavior in a Semi-anonymous Social Network

Homa Hosseinmardi, Amir Ghasemian+3 more|arXiv (Cornell University)|Apr 15, 2014

Bullying, Victimization, and Aggression9 references14 citations

TL;DR

This study analyzes negative user behavior in the semi-anonymous social network Ask.fm by examining offensive language in question/answer posts and the 'likes' social network. Using linguistic analysis and network metrics, it identifies patterns linked to cyberbullying and self-harm, revealing that users with 'cutting' behavior exhibit distinct linguistic and social network traits, contributing to early detection frameworks for harmful online behavior.

ABSTRACT

Abstract—Cyberbullying has emerged as an important and growing social problem, wherein people use online social networks and mobile phones to bully victims with offensive text, images, audio and video on a 24/7 basis. This paper studies negative user behavior in the Ask.fm social network, a popular new site that has led to many cases of cyberbullying, some leading to suicidal behavior. We examine the occurrence of negative words in Ask.fm’s question/answer profiles along with the social network of “likes ” of questions/answers. We also examine properties of users with “cutting ” behavior in this social network. I.

Motivation & Objective

• To investigate the prevalence and characteristics of negative language in Ask.fm's question/answer content.
• To examine the social network structure formed by 'likes' on questions and answers to detect patterns of harmful behavior.
• To identify linguistic and network-level indicators associated with users exhibiting 'cutting' behavior.
• To contribute insights for early detection systems targeting cyberbullying and self-harm in semi-anonymous platforms.

Proposed method

• Collects anonymized data from Ask.fm, including user questions, answers, and 'like' interactions.
• Applies natural language processing to detect and quantify negative words and phrases in user content.
• Constructs a directed social network based on 'likes' to analyze user interaction patterns.
• Performs network analysis on the 'likes' graph to identify structural features of users with negative behavior.
• Compares linguistic profiles and network centrality metrics between users with 'cutting' behavior and general users.
• Uses statistical analysis to correlate linguistic features and network positions with self-harm indicators.

Experimental results

Research questions

• RQ1 What is the frequency and distribution of negative language in Ask.fm's question and answer content?
• RQ2 How do the 'likes' networks of users exhibiting 'cutting' behavior differ from those of other users?
• RQ3 What linguistic patterns are associated with users who self-report or are identified as having 'cutting' behavior?
• RQ4 To what extent do network centrality measures predict the presence of negative behavior in semi-anonymous social networks?

Key findings

• Users with 'cutting' behavior use significantly more negative words in their questions and answers compared to general users.
• The 'likes' network of users with 'cutting' behavior shows higher clustering and more centralized interaction patterns.
• A higher proportion of negative content is directed toward users who self-report 'cutting' behavior, indicating targeted cyberbullying.
• Users with 'cutting' behavior are more likely to be at the center of 'like' networks, suggesting increased visibility despite negative content.
• The linguistic profile of users with 'cutting' behavior includes elevated use of words related to sadness, isolation, and self-harm.
• The study identifies a strong correlation between high network centrality and the presence of negative language, especially in self-revealing posts.