Information consolidation: A new paradigm in knowledge search

信息整合：知识搜索的新范式

• 批准号: 245844862
• 负责人: Professor Dr. Ido Dagan, Ph.D.
• 金额: --
• 依托单位: Fachgebiet Ubiquitäre Wissensverarbeitung
• 依托单位国家: 德国
• 项目类别: DIP Programme
• 财政年份: 2014
• 资助国家: 德国
• 起止时间: 2013-12-31 至 2021-12-31
• 项目状态: 已结题
• 来源: https://gepris.dfg.de/gepris/projekt/245844862?language=en
• 关键词: Information; consolidation; new; paradigm; knowledge

Existing search tools are effective at identifying relevant documents among billions of irrelevant ones. Yet, when users search for non-trivial knowledge on a topic, rather than just having to peek at a couple documents, state of the art technology offers little help. Search engines typically provide lists of documents, but remain weak at isolating the concrete facts of interest within them, and at organizing and presenting these facts concisely and intuitively. Today, searchers have to laboriously skim through many retrieved documents, collect the relevant statements within them and consolidate the largely redundant information in order to acquire coherent knowledge. Our proposal targets what might be the next big step in information access technology: supporting users in identifying and assimilating the large set of relevant statements found within multitudes of target documents. Based on an interdisciplinary approach, combining methods from information retrieval, natural language processing and information science, we propose a novel paradigm for knowledge search. Concretely, we propose developing an automated information consolidation approach that consists of three major processes: (a) automatically extracting key statements from large document sets; (b) consolidating the information in these statements by constructing a statement graph, which specifies the inference relations between all extracted statements; and (c) enabling users to explore the consolidated information in the graph via suitable user interfaces that support effective exploration schemes. We plan to develop algorithms and methods that implement this approach and build a prototype system. As a target domain for testing our approach we chose the field of education policies, which has rich information structure and is considered of high social importance. We will apply our system to information in this domain and will conduct intrinsic as well as extrinsic evaluations, including user studies. Our consortium structure is quite uniquely suitable for realizing our vision, including scientists with complementary skills from all involved disciplines.

现有的搜索工具可以有效地从数十亿不相关的文档中识别相关文档。然而，当用户搜索关于某个主题的重要知识时，而不仅仅是浏览几个文档，最先进的技术几乎没有提供任何帮助。搜索引擎通常提供文档列表，但在分离其中感兴趣的具体事实以及简洁直观地组织和呈现这些事实方面仍然很弱。今天，搜索者不得不费力地浏览许多检索到的文件，收集其中的相关陈述，并合并大量冗余的信息，以获得连贯的知识。我们的建议的目标可能是信息访问技术的下一个重大步骤：支持用户识别和吸收大量目标文档中的大量相关语句。基于跨学科的方法，结合信息检索，自然语言处理和信息科学的方法，我们提出了一个新的知识搜索范式。具体地说，我们提出了一种自动信息整合方法，它包括三个主要过程：（a）从大文档集中自动提取关键语句;（B）通过构造语句图来整合这些语句中的信息，语句图指定了所有提取语句之间的推理关系;以及（c）使得用户能够经由支持有效探索方案的适当用户界面来探索图中的合并信息。我们计划开发实现这种方法的算法和方法，并建立一个原型系统。作为测试我们的方法的目标域，我们选择了教育政策领域，它具有丰富的信息结构，被认为是高度的社会重要性。我们将把我们的系统应用于这一领域的信息，并将进行内在和外在的评估，包括用户研究。我们的联盟结构非常适合实现我们的愿景，包括来自所有相关学科的具有互补技能的科学家。