碎片化知识聚合方法研究

Massive fragmentized knowledge from the World Wild Web have three features: isolated location, incomplete content, distorted structure, which lead to “cognitive overload” and “disorientation” for learner during knowledge acquisition and cognitive learning. This project proposed a novel knowledge fusion method by combining faceted topic taxonomies and cognitive relations -- knowledge forest, making the knowledge fragments from multiple isolated sources to be ordered and structured. Based on knowledge forest, knowledge fragments are fused with three steps: faceted taxonomy construction, fragmentized knowledge association and cognitive relation mining. The key issues to be addressed include: formalization of knowledge forest and modeling of the fusion process, construction and optimization of faceted topic taxonomy, mining association between faceted topic taxonomy and knowledge fragments, mining cognitive relation between knowledge topics, visual navigating and searching of knowledge forest. A demonstration system for fragmentized knowledge fusion is developed for evaluating the effectiveness of the proposed methods by empirical experiments in an e-learning environment...This project will promote the interdiscipline fusion of computer science and cognitive science, and produce a new knowledge fusion pattern based on cognitive learning. The proposed methods have great application prospects in web search engine, digital library, e-Learning, public opinion analysis and so on. The proposed method can help to create a bran-new knowledge acquisition pattern based on the navigation of knowledge forest. The pattern can “see the minute details and also see the major issue,” overcoming the limit of “seeing the minute details but missing the major issue.”

互联网海量碎片化知识因位置分散、内容片面、结构无序对人们的知识获取与认知学习造成了认知过载、学习迷航等新挑战。提出一种融合主题分面树与认知关系的新型知识聚合方式——知识森林，实现多源分布的碎片化知识的结构化、有序化；提出“主题分面树生成—碎片化知识装配—认知关系挖掘”三阶段碎片化知识聚合思路，研究知识森林的形式化表示、知识主题分面树生成与优化、碎片化知识装配、主题间认知关系的挖掘、知识森林可视化导航与检索等5个关键问题，研制碎片化知识聚合原型系统，并以开放知识源和MOOC为对象，在e-Learning中开展实证测试。. 本研究有助于计算机科学与认知科学的交叉融合，产生新的认知学习和知识聚合模式，在搜索引擎、数字图书馆、e-Learning、舆情监控等典型领域具有广阔应用前景，有助于建立全新的基于知识森林导航的“既见树木、又见森林”的知识获取模式，克服只见树木、不见森林的局限性。

本项目旨在将多源、异构、无序的碎片知识聚合成结构化、有序化且符合人类认知特点的组织形式，缓解海量碎片知识因位置分散、内容片面、结构无序三个特性引发的认知过载与学习迷航问题。.1）提出一种融合主题分面树与认知关系的新型知识聚合方式——知识森林。主题分面树使用具有上下位关系的分面组织碎片知识以减轻认知过载问题；认知关系对知识主题进行组织来应对学习迷航问题；实现多源分布的碎片知识的结构化、有序化聚合。.2）研究“分面树生成—知识装配—认知关系挖掘”三阶段的知识森林构建机理；突破知识主题分面树生成、碎片知识装配、知识主题间认知关系挖掘、知识森林可视化等知识森林构建关键技术。.3）研究基于知识森林的碎片知识检索和检索答案解释技术。在知识森林图存储的基础上，提出基于图模式匹配的碎片知识检索方法，设计基于自动查询精化的解释模型，解决检索过程中的Why/Why-Not问题。.4）研制出知识森林构建工具及知识森林导航系统，并以开放域知识源和MOOC为对象，开展面向e-Learning的实证测试。.取得的主要成果包括：在IEEE TIP、Pattern Recognition、VLDB Journal、TKDE、Neural Computation、IEEE VAST、WWW、ICDE、DASFAA等国际期刊与会议上发表学术论文60多篇；申请20多项国家发明专利；获得省部级一等奖2项，国家教学成果二等奖1项；研制出碎片知识融合系统及相关工具，构建200门课程的知识森林数据集；培养20多名硕博研究生。.本项目有助于产生新的认知学习和知识聚合模式，在搜索引擎、数字图书馆、e-Learning、舆情监控等典型领域具有广阔应用前景；有助于建立全新的基于知识森林导航的“既见树木、又见森林”的知识获取模式，克服只见树木、不见森林的局限性。