I-Corps: Automated Summarization Technology

I-Corps项目的更广泛影响/商业潜力集中在改进人类对长而复杂文件中所载信息的处理和理解。利用美国国家科学基金会资助的数字图书馆应用信息检索和人工智能技术（如自然语言处理、机器学习和深度学习）的研究，该项目将评估这些技术（特别关注摘要）是否能在保险和法律行业带来新的商业机会。这些行业之间的一个联系是共享长而复杂的文档。将摘要方法与信息检索相集成，并进一步利用文本分析和挖掘技术，可以帮助这些企业更快、更有效地管理这些文档。更广泛地说，为长文档创建有用摘要的问题适用于许多业务，包括出版业，在这些业务中，所建议的技术可以帮助作者和读者，以及那些参与发现、搜索和管理已发布文档集合的人。I-Corps项目将探讨如何改进先进的数字图书馆方法并将其应用于商业环境。帮助管理和理解信息的一个重要服务是摘要。虽然最近出现了自动提取（例如，从文档中选择一组最重要的句子）和抽象（例如，自动准备一份摘要，就像由人类抽象人员所做的那样）总结方法，应用于新闻文章或其他简短但受欢迎的作品，但只有提取方法在应用于较长的文档时取得了真正的成功。设计和验证用于为长文档构造抽象摘要的有效技术是一个非常困难的问题，这是本项目努力的重点。此外，使这些技术能够处理各种领域中的复杂文档是一个特殊的挑战。将这些努力与大数据收集、在线云系统、可解释的人工智能方法和功能性用户界面联系起来，是必须在具有商业利益的新环境中解决的额外挑战。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is focused on improving human handling and understanding of information contained in long and complex documents. Drawing upon NSF-funded research in digital libraries which apply information retrieval and artificial intelligence techniques (such as natural language processing, machine learning, and deep learning), this project will evaluate whether such technologies (with a particular focus on summarization) can lead to new business opportunities within both the insurance and legal industry. One connection between these industries is the sharing of long, often complex, documents. Integrating summarization methods with information retrieval, and further leveraging text analysis and mining technologies, could aid these businesses in more quickly and efficiently managing such documents. More broadly, the problem of creating a useful summary for a long document is applicable to many businesses, including the publishing industry, where the proposed technology could assist both authors and readers, as well as those involved in discovering, searching, and managing published document collections.This I-Corps project will address how advanced digital library methods can be improved and applied in commercial settings. An important service to aid with managing and understanding information is summarization. While there has been recent automation of extractive (e.g., selecting a set of the most important sentences from a document) and abstractive (e.g., automatically preparing an abstract as would be done by a human abstracter) summarization methods, as applied to news articles or other short but popular works, only extractive methods have yielded much real success when applied to longer documents. Devising and validating effective techniques for constructing abstractive summaries for long documents is a very difficult problem, which is a key focus of this project's efforts. Further, making those techniques work with complex documents in a variety of domains is a special challenge. Connecting such efforts with big data collections, online cloud systems, explainable artificial intelligence methods, and functional user interfaces, are additional challenges that must be addressed in new settings that are of commercial interest.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.