The paper presents Scatter/Gather, a cluster-based document browsing method, as an alternative to ranked titles for organizing and viewing retrieval results. It systematically evaluates Scatter/Gather and finds significant improvements over similarity search ranking alone, providing evidence validating the cluster hypothesis, which states that relevant documents tend to be more similar to each other than to non-relevant documents. The system uses Scatter/Gather to cluster documents into topic-coherent groups, providing descriptive summaries to users. Users can select clusters for iterative examination, with clustering and reclustering done on-the-fly. Scatter/Gather can be applied to entire corpora, with static offline computations used to speed dynamic online clustering. The paper also discusses related work on clustering in information retrieval, the architecture of the interactive system, and two evaluations. The first evaluation compares selecting a best cluster to an equivalent cutoff in ranked retrieval results, while the second examines whether users select the best cluster. The results show that clustering significantly improves retrieval results, supporting the cluster hypothesis. The paper also presents a user study showing that users can effectively use Scatter/Gather to navigate retrieval results and select clusters with the largest number of relevant documents. The study indicates that users are able to take advantage of the benefits that clustering can provide. The paper concludes that the Scatter/Gather approach to document clustering can produce significant improvements over similarity search ranking alone, and that clustering is tailored to the characteristics of the query rather than assuming a one-size-fits-all approach. The results support the cluster hypothesis, which states that relevant documents tend to be more similar to each other than to non-relevant documents.The paper presents Scatter/Gather, a cluster-based document browsing method, as an alternative to ranked titles for organizing and viewing retrieval results. It systematically evaluates Scatter/Gather and finds significant improvements over similarity search ranking alone, providing evidence validating the cluster hypothesis, which states that relevant documents tend to be more similar to each other than to non-relevant documents. The system uses Scatter/Gather to cluster documents into topic-coherent groups, providing descriptive summaries to users. Users can select clusters for iterative examination, with clustering and reclustering done on-the-fly. Scatter/Gather can be applied to entire corpora, with static offline computations used to speed dynamic online clustering. The paper also discusses related work on clustering in information retrieval, the architecture of the interactive system, and two evaluations. The first evaluation compares selecting a best cluster to an equivalent cutoff in ranked retrieval results, while the second examines whether users select the best cluster. The results show that clustering significantly improves retrieval results, supporting the cluster hypothesis. The paper also presents a user study showing that users can effectively use Scatter/Gather to navigate retrieval results and select clusters with the largest number of relevant documents. The study indicates that users are able to take advantage of the benefits that clustering can provide. The paper concludes that the Scatter/Gather approach to document clustering can produce significant improvements over similarity search ranking alone, and that clustering is tailored to the characteristics of the query rather than assuming a one-size-fits-all approach. The results support the cluster hypothesis, which states that relevant documents tend to be more similar to each other than to non-relevant documents.