The paper introduces LexRank, a stochastic graph-based method for computing the relative importance of textual units in Natural Language Processing, particularly for text summarization. The method uses eigenvector centrality in a graph representation of sentences, where the connectivity matrix is based on intra-sentence cosine similarity. LexRank outperforms centroid-based methods and other systems in the DUC 2004 evaluation, demonstrating its effectiveness in extractive text summarization. The authors also discuss the advantages of LexRank over centroid-based methods, such as better handling of information subsumption and resistance to noisy data. Experimental results show that LexRank performs well on various datasets, including noisy data, and compares favorably with other summarization systems.The paper introduces LexRank, a stochastic graph-based method for computing the relative importance of textual units in Natural Language Processing, particularly for text summarization. The method uses eigenvector centrality in a graph representation of sentences, where the connectivity matrix is based on intra-sentence cosine similarity. LexRank outperforms centroid-based methods and other systems in the DUC 2004 evaluation, demonstrating its effectiveness in extractive text summarization. The authors also discuss the advantages of LexRank over centroid-based methods, such as better handling of information subsumption and resistance to noisy data. Experimental results show that LexRank performs well on various datasets, including noisy data, and compares favorably with other summarization systems.