Echo Chambers in the Age of Algorithms: An Audit of Twitter's Friend Recommender System

TL;DR

This study provides the first empirical audit of Twitter's Who-To-Follow friend recommendation system during the 2022 U.S. midterms. Using automated audit accounts and an observational baseline, it reveals that the recommender drives dense, interconnected networks that are structurally echo-chamber-like but ideologically more diverse than both social-endorsement growth and real-world followers. The results show lower exposure to false or misleading election content under the algorithmic growth condition, suggesting that recommender systems alone are not the sole driver of online echo chambers. The findings highlight the need to study social factors alongside algorithms to understand and promote healthier information environments on social platforms.

Abstract

The presence of political misinformation and ideological echo chambers on social media platforms is concerning given the important role that these sites play in the public's exposure to news and current events. Algorithmic systems employed on these platforms are presumed to play a role in these phenomena, but little is known about their mechanisms and effects. In this work, we conduct an algorithmic audit of Twitter's Who-To-Follow friend recommendation system, the first empirical audit that investigates the impact of this algorithm in-situ. We create automated Twitter accounts that initially follow left and right affiliated U.S. politicians during the 2022 U.S. midterm elections and then grow their information networks using the platform's recommender system. We pair the experiment with an observational study of Twitter users who already follow the same politicians. Broadly, we find that while following the recommendation algorithm leads accounts into dense and reciprocal neighborhoods that structurally resemble echo chambers, the recommender also results in less political homogeneity of a user's network compared to accounts growing their networks through social endorsement. Furthermore, accounts that exclusively followed users recommended by the algorithm had fewer opportunities to encounter content centered on false or misleading election narratives compared to choosing friends based on social endorsement.

Figures (6)

• Figure 1: Audit accounts were created in matched pairs to compare network growth strategies where both accounts initially followed a single U.S. politician (seed account) running for office during the 2022 midterms. Seed accounts were chosen as the Democratic and Republican candidates from five Senate races. Across the five Senate races, 20 accounts were created in the audit (two candidates per race, two audit accounts per candidate).
• Figure 2: Network measurements of the audit accounts' personal networks, grouped by network growth method and seed party. A) Density B) Reciprocity C) Number of strongly connected components after removing the ego node D) Number of weakly connected components after removing the ego node. Error bars represent the standard error (n=5 audit account networks in each group except the Democratic Algorithmic group where n=4).
• Figure 3: Characteristics of the accounts followed by the audit accounts, grouped by network growth method and seed party. A) Median number of friends B) Median number of followers C) Median account age in years D) Percent of accounts that are verified by Twitter. Error bars represent show the standard error (n=5 audit account networks in each group except the Democratic Algorithmic group where n=4).
• Figure 4: Political makeup of the of the personal network of each audit account (columns 1, 2, 4, and 5) and each comparison group (columns 3 and 6). Users are classified as left-leaning (blue), right-leaning (red) or neutral (orange) as described in section \ref{['data:partisanship']}
• Figure 5: The average partisanship over time of each audit account's set of friends. Left leaning accounts are assigned the value of -1, neutral or apolitical accounts have a value of 0 and right leaning accounts have a value of 1.