Date of Award

5-10-2023

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Information Science

First Advisor

Nitin Agarwal

Abstract

YouTube serves as a primary information source for many users, with its recommendation algorithm playing a vital role in video discovery and viewership on the platform. It determines what users are exposed to and is responsible for a significant portion (70%) of the content users engage with on the platform. It is therefore crucial to scrutinize recommendation systems to understand potential algorithmic biases that may spread disinformation. Previous studies have shown that the recommendation algorithm favors a small number of videos, creating mild ideological echo chambers. This study aims to investigate the extent to which YouTube's recommendation algorithm spreads disinformation by analyzing the Cheng Ho narrative. Cheng Ho was a Chinese Muslim naval admiral in the 15th century, known as the "Chinese Columbus," and symbolized China's peaceful ascendancy to power. To achieve this aim, a list of 50 videos on Cheng Ho was gathered by using relevant keywords with YouTube's search API. These videos served as the seed for recommendations, with 58,825 unique videos collected through five depths of recommendations. To determine the relevance of the recommendations to the Cheng Ho narrative, I computed the topic drift on the recommendation depths and discovered that the recommendations led us further away from the original topic. Furthermore, observing the eigenvector centrality values of videos within the recommendation network of different depths, I saw the evolution of influential videos as their relevance to Cheng Ho diminished. The results showed how YouTube’s recommendation system discards the topics of the seed videos by subtly introducing a new but still pro-China topic in the network through influential videos. This new topic is about economic growth and religious freedom in China targeting Indonesia’s younger demographic by focusing on current events and pop culture. This study sets the stage for further research in analyzing bias in recommendation algorithms, their exploitation by information actors, their impact on mis/disinformation propagation, and their effect on user consumption.

Download

Included in

Social Media Commons

Share

COinS