Publication Date

8-2022

Date of Final Oral Examination (Defense)

6-23-2022

Type of Culminating Activity

Thesis

Degree Title

Master of Science in Computer Science

Department

Computer Science

Supervisory Committee Chair

Francesca Spezzano, Ph.D.

Supervisory Committee Member

Edoardo Serra, Ph.D.

Supervisory Committee Member

Nasir Eisty, Ph.D.

Abstract

Wikipedia is a free Internet-based encyclopedia that is built and maintained via the open-source collaboration of a community of volunteers. Wikipedia’s purpose is to benefit readers by acting as a widely accessible and free encyclopedia, a comprehensive written synopsis that contains information on all discovered branches of knowledge. The website has millions of pages that are maintained by thousands of volunteer editors. Unfortunately, given its open-editing format, Wikipedia is highly vulnerable to malicious activity, including vandalism, spam, undisclosed paid editing, etc.

Malicious users often use sockpuppet accounts to circumvent a block or a ban imposed by Wikipedia administrators on the person’s original account. A sockpuppet is an “online identity used for the purpose of deception.” Usually, several sockpuppet accounts are controlled by a unique individual (or entity) called a puppetmaster. Currently, suspected sockpuppet accounts are manually verified by Wikipedia administrators, which makes the process slow and inefficient.

The primary objective of this research is to develop an automated ML and neural-network-based system to recognize the patterns of sockpuppet accounts as early as possible and recommend suspension. We address the problem as a binary classification task and propose a set of new features to capture suspicious behavior that considers user activity and analyzes the contributed content. To comply with this work, we have focused on account-based and content-based features. Our solution was bifurcated into developing a strategy to automatically detect and categorize suspicious edits made by the same author from multiple accounts. We hypothesize that “you can hide behind the screen, but your personality can’t hide.” In addition to the above-mentioned method, we have also encountered the sequential nature of the work. Therefore, we have extended our analysis with a Long Short Term Memory (LSTM) model to track down the sequential pattern of users’ writing styles.

Throughout the research, we strive to automate the sockpuppet account detection system and develop tools to help the Wikipedia administration maintain the quality of articles. We tested our system on a dataset we built containing 17K accounts validated as sockpuppets. Experimental results show that our approach achieves an F1 score of 0.82 and outperforms other systems proposed in the literature. We plan to deliver our research to the Wikipedia authorities to integrate it into their existing system.

DOI

https://doi.org/10.18122/td.1994.boisestate

Download

Share

COinS