Interactive Cluster Visualization for Information Retrieval
Interactive Cluster Visualization for Information Retrieval
复制标题
用于信息检索的交互式集群可视化
DOI：
--
复制
发表时间：
1997
期刊：
影响因子：
--
通讯作者：
A. Leouski
中科院分区：
文献类型：
--
作者：
James Allan;A. Leouski
这项研究调查了集群可视化帮助用户快速识别相关文档的能力。它对检索到的文档的集群假设的真实性提供了额外的支持，并表明相关文档的集群是容易看到的。然后，这项研究展示了一种类似于相关反馈的技术的视觉效果，并展示了如何增强视觉效果，以进一步帮助用户找到相关材料。文本搜索引擎返回的排序列表声称以文档最可能相关的顺序呈现文档：第一个文档最匹配用户的查询，第二个文档最有可能有帮助，依此类推。我们感兴趣的情况是，这个简单的模型出现故障|用户无法在列表的第一个或第二个屏幕中找到足够的相关材料。特别是，我们感兴趣的是帮助搜索者在不强迫他或她费力浏览所有不相关的材料的情况下，搜索排名靠前的列表中的所有相关材料。我们的方法是基于文档聚类和可视化的组合。我们已经观察到，当文档被聚集并且它们的关系被可视化地显示时，相关文档通常在可视化中聚集在一起。在这项研究中，我们调查了与这一观察结果相关的几个假设：1.聚类有助于分离相关和不相关的文档。这一假设对我们的工作至关重要，但一点也不新奇或令人惊讶
This study investigates the ability of cluster visualization to help a user rapidly identify relevant documents. It provides added support for the truth of the Cluster Hypothesis on retrieved documents and shows that clustering of relevant documents is readily visible. The study then shows the visual eeect of a technique similar to relevance feedback and shows how to enhance that eeect to further help the user locate relevant material. A ranked list returned by a text search engine purports to present the documents in the order they are most likely to be relevant: the rst document is the best match for the user's query, the second is the next most likely to be helpful, and so on. We are interested in situations where this simple model breaks down|where the user is unable to nd enough relevant material in the rst or second screens of the list. In particular, we are interested in helping a searcher nd all of the relevant material in the top ranked list without forcing him or her to wade through all of the non-relevant material. Our approach is based on a combination of document clustering and vi-sualization. We have observed that when documents are clustered and their relationships are visually displayed, the relevant documents generally clump together in the visualization. In this study, we investigate several hypotheses related to this observation: 1. Clustering is useful for separating relevant and non-relevant documents. This hypothesis is critical to our work, but not at all novel or surprising 1