SBIR Phase I: Concise Visualization of a Document Collection via Conceptual Clustering

SBIR 第一阶段：通过概念聚类对文档集合进行简洁可视化

• 批准号: 0060524
• 负责人: Raul Valdes-Perez
• 金额: $ 9.94万
• 依托单位: Vivisimo, Inc.
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2001
• 资助国家: 美国
• 起止时间: 2001-01-01 至 2001-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0060524&HistoricalAwards=false
• 关键词: SBIR; Phase; Concise; Visualization; Document

This Small Business Innovation Research (SBIR) Phase I project will spin off NSF-sponsored basic research on knowledge discovery at Carnegie Mellon University computer science department. The result will be commercial software that can convey the contents of hundreds or thousands of documents on one computer screen with minimal clicking and scrolling. This capability will enhance information needs as diverse as search, overviewing, and browsing, and alleviate the problem of information overload, which today confronts all retrievers of computer-based textual information. The basic approach is a new form of conceptual clustering that emphasizes the human describability of the resulting document clusters. The techniques combine classical hierarchical clustering with results from the PI's research on data-driven knowledge discovery, which focused on generating very concise and contrastive descriptions of a large number of classes (here, document clusters). The overall goal is to replace the tedious long ranked list display of matching documents, which is nearly universal, but which forces users into repeated and inefficient clicking, backtracking, and scrolling.The potential market opportunities include any domain where more than a few dozen relevant matches are returned for typical information queries, such as web searches, news, patents, scientific research abstracts, proprietary corporate information, and, generally the content delivered by the numerous vendors of specialized information services.

这个小企业创新研究 (SBIR) 第一阶段项目将在卡内基梅隆大学计算机科学系开展由 NSF 资助的知识发现基础研究。 其结果将是商业软件可以通过最少的点击和滚动在一台计算机屏幕上传达数百或数千个文档的内容。 这种能力将增强搜索、概述和浏览等多样化的信息需求，并缓解当今所有基于计算机的文本信息检索者所面临的信息过载问题。 基本方法是一种新形式的概念聚类，强调生成的文档聚类的人类可描述性。 这些技术将经典的层次聚类与 PI 对数据驱动的知识发现的研究结果相结合，该研究的重点是生成大量类（此处为文档集群）的非常简洁和对比的描述。 总体目标是取代乏味的长排名的匹配文档列表显示，这种显示方式几乎是普遍存在的，但却迫使用户重复且低效的点击、回溯和滚动。潜在的市场机会包括任何典型信息查询返回几十个相关匹配项的领域，例如网络搜索、新闻、专利、科学研究摘要、专有公司信息，以及通常由众多供应商提供的内容 的专业化信息服务。