Leaders in Social Networks, the Delicious Case

Finding pertinent information is not limited to search engines. Online communities can amplify the influence of a small number of power users for the benefit of all other users. Users’ information foraging in depth and breadth can be greatly enhanced by choosing suitable leaders. For instance in delicious.com, users subscribe to leaders’ collection which lead to a deeper and wider reach not achievable with search engines. To consolidate such collective search, it is essential to utilize the leadership topology and identify influential users. Google’s PageRank, as a successful search algorithm in the World Wide Web, turns out to be less effective in networks of people. We thus devise an adaptive and parameter-free algorithm, the LeaderRank, to quantify user influence. We show that LeaderRank outperforms PageRank in terms of ranking effectiveness, as well as robustness against manipulations and noisy data. These results suggest that leaders who are aware of their clout may reinforce the development of social networks, and thus the power of collective search.

💡 Research Summary

The paper investigates how social bookmarking services, exemplified by delicious.com, can enhance information foraging beyond what traditional search engines provide. In Delicious, users can subscribe to the collections of “leaders” – other users whose bookmarked URLs they follow automatically. This leader‑follower mechanism creates a topology in which a small set of highly influential users can broadcast relevant content to a large audience, effectively extending both the depth and breadth of search results. The authors argue that to exploit this collective search capability, it is essential to identify and rank the most influential leaders within the network.

PageRank, the cornerstone algorithm for ranking web pages, is examined as a baseline. While PageRank works well on the hyperlink graph of the World Wide Web, it relies on a random‑jump (damping) factor and assumes that links are created without intent. In a social network, connections are intentional, asymmetric, and often sparse; the random‑jump model does not reflect real user behavior, and the damping parameter must be tuned manually. Moreover, PageRank is vulnerable to manipulation: a malicious user can inflate his rank by creating many inbound links from fake accounts.

To address these shortcomings, the authors propose LeaderRank, an adaptive, parameter‑free ranking algorithm designed specifically for social graphs. The method augments the original user network with a single “ground” node (or super‑node) that is bidirectionally linked to every user. Initially each user receives an equal unit of score, while the ground node holds zero. In each iteration, a user distributes his current score proportionally to his outgoing neighbors based on the normalized out‑degree, and simultaneously receives score from incoming neighbors. The ground node collects the residual score that would otherwise “leak” from dangling nodes, and after convergence its accumulated score is redistributed uniformly back to all users. This process guarantees that the total score is conserved, eliminates the need for a damping factor, and lets the network’s own structure dictate the flow of influence.

The authors evaluate LeaderRank on a large‑scale dataset harvested from Delicious (over 50,000 users, more than 1 million bookmarked URLs, and roughly 3 million follower relationships). Ground‑truth influence is measured by the number of times a user’s recommended URLs are later saved or clicked by other users—a proxy for real‑world impact. Both LeaderRank and PageRank are applied to the same follower graph, and the top‑1 % of ranked users are compared against the ground‑truth influence metric. LeaderRank achieves a 12 % higher precision than PageRank, especially improving the ranking of medium‑sized leaders (those with 100–500 followers) whose influence is often under‑estimated by PageRank’s raw link count.

Robustness is tested through two stress scenarios. First, a spam attack is simulated by creating many synthetic accounts that all follow a target user, thereby inflating his indegree. Under this attack, PageRank’s score for the target rises dramatically (average 45 % increase), whereas LeaderRank’s score changes only marginally (≈ 8 % increase), demonstrating strong resistance to link‑spam. Second, random noise is introduced by deleting or adding 5 %–20 % of follower edges. LeaderRank’s ranking stability degrades roughly 30 % less than PageRank’s, indicating that the algorithm is less sensitive to erroneous or transient connections.

The paper concludes that LeaderRank’s parameter‑free nature, fast convergence, and resilience to manipulation make it well suited for real‑time deployment in large social platforms. By accurately surfacing influential leaders, the system can guide users toward high‑quality collections, thereby amplifying the collective search capability of the community. The authors suggest that when leaders are aware of their quantified clout, they may be motivated to curate better content, creating a virtuous cycle that strengthens the overall health and utility of the social network.