Link-Spam-Resistant Domain Ranking: A Comparative Evaluation of PageRank, HITS, and TrustRank Under Simulated Link-Farm Attacks Using Common Crawl Web Graphs

Link-based ranking is still an important part of web information retrieval, but it remains vulnerable to coordinated link-farm attacks. This study examined the resistance of PageRank, HITS, and TrustRank to link spam using simulated attacks on domain-level web graphs extracted from Common Crawl. A quantitative, comparative, and controlled experimental design was used. Five independent base graphs were prepared, with each graph containing 100,000 legitimate domain nodes. Four link-farm structures were tested: star-centered, bounded-density clique, bipartite, and mixed. The attacks were applied at four intensity levels: 1%, 3%, 5%, and 10% of the legitimate graph size. Across all conditions, the experiment produced 480 attacked graph instances and 1,440 verified algorithm-level runs. The algorithms were compared using spam target promotion, spam score share, top-1000 spam infiltration, legitimate-domain displacement, ranking stability, execution time, peak memory use, and iteration count. The results showed that PageRank was the most vulnerable algorithm. It had the highest spam target promotion, highest spam score share, measurable top-1000 spam infiltration, greater legitimate-domain displacement, and lowest ranking stability. HITS showed the strongest overall resistance and computational efficiency. It recorded near-zero spam score share, zero top-1000 infiltration, zero legitimate-domain displacement, highest stability, fastest runtime, and fewest iterations. TrustRank also blocked top-1000 infiltration and legitimate-domain displacement, although it allowed higher target promotion than HITS. The bounded-density clique structure caused the most damage to PageRank, while higher attack intensity increased spam impact in most cases. The findings show that link-spam resistance depends on the ranking algorithm, attack structure, attack intensity, and evaluation metric used. A single ranking score or attack model is not enough to evaluate resistance to link-farm manipulation.