STIMULATE: Generating Coherent Summaries of On-Line Documents: Combining Statistical and Symbolic Techniques

本项目提出了一个多层次的摘要过程，其特点是：(1)综合使用统计和符号技术得出的关于文件的知识；(2)使用语言生成将这些信息重新表述为指示性和信息性摘要；(3)跨相关文章集的摘要；以及(4)开发独立于领域的摘要技术。拟议的研究包括文件分析和摘要生成。文档分析涉及将文档分割成多个主题，使用统计技术来识别每个部分中的关键内容，并构建词汇语义表示以供摘要生成器使用。生成侧重于整合来自几个数据来源的相关信息，展示来自输入文章的对比信息，以及使用规划运算符将来自不同表述的信息组合成一个简明的文本。这项研究包括用于衡量分析器和生成器的覆盖率和稳健性的评估技术，包括概念和语言知识。通过与哥伦比亚大学信息获取研究中心的合作，拟议的跨域动态生成、连贯的摘要研究最终将成为数字图书馆界面的一部分，该界面根据用户的兴趣和需求过滤文档。

This project proposes a multi-level process of summarization featuring: (1) integration of knowledge about a document derived with both statistical and symbolic techniques; (2) the use of language generation for reformulating this information into both indicative and informative summaries; (3) summarization across sets of related articles; and (4) development of techniques for domain independent summarization. The proposed research comprises both document analysis and summary generation. Document analysis involves segmenting a document into topics, using statistical techniques to identify key content within each section, and constructing a lexical semantic representation for use by the summary generator. Generation focuses on integration of relevant information from several data sources, on presentation of contrasting information from the input articles, and on the use of planning operators to combine information from the separate representations into a single concise text. The research includes evaluation techniques for measuring both coverage and robustness of the analyzer and generator, including conceptual and linguistic knowledge. Through collaboration with Columbia's Center for Research on Information Access, the proposed research on dynamically generated, coherent summarization across domains will ultimately be part of a digital library interface that filters documents according to a user's interests and needs.