基于众包的知识融合关键技术研究

Knowledge Fusion is a new research area that integrates knowledge data extracted from different data sources to build a knowledge base with higher coverage and better quality. However, most of the existing works on knowledge fusion are purely machine-based, and they do not always work well due to the inherently difficult semantic heterogeneity. Meanwhile, the existing works ignore the potential and powerful ability of human on knowledge fusion. As a new computing model, crowdsourcing utilizes the wisdom of the crowd and provides an opportunity of introducing the computation power of human. However, there are few systematic studies on crowdsourcing-based knowledge fusion. To address the problems, this project proposes to study effective crowdsourcing-based techniques for knowledge fusion. The key scientific problem summarized in this proposal is crowdsourcing-based computation modelling and cost-budget aware optimization. The research problems include (1) crowdsourcing fusion task design, (2) crowdsourcing fusion task collaboration, (3) crowdsourcing fusion task selection, (4) crowdsourcing fusion task inference, and (5) crowdsourcing knowledge fusion prototype system. The objective is to establish the crowdsourcing-based knowledge fusion theory, address the key problems, build a prototype system, and make research breakthroughs.

知识融合将抽取自不同数据源的知识数据融合起来，构建覆盖更全、质量更高的知识库，其核心难点是多数据源的语义异质性。现有工作大多单纯基于机器设计融合算法，在融合效果上尚未令人满意，同时也忽略了人在知识融合中潜在而巨大的作用。作为一种新型计算模式，众包为引入人的计算能力提供了机会，但目前系统化的众包知识融合研究还十分欠缺。本项目提出研究基于众包的知识融合，发挥人在解决语义异质性上的优势，提升整体融合效果。项目拟解决的关键科学问题是众包知识融合的计算模型与优化机制，包括支持知识融合的众包任务建模问题与成本受限约束下的融合效果优化问题。围绕关键科学问题，项目拟研究：(1) 众包知识融合任务设计；(2) 众包知识融合任务协作；(3) 众包知识融合任务选择；(4) 众包知识融合任务推理；(5) 原型系统研制。目标是突破众包知识融合基础理论，解决关键技术，构建原型系统，取得国内外领先的研究成果。

本项目提出研究基于众包的知识融合，发挥人在解决语义异质性上的优势，提升整体融合效果。其目标是解决众包知识融合的计算模型与优化机制关键科学问题，包括支持知识融合的众包任务建模问题与成本受限约束下的融合效果优化问题。通过项目的执行，课题组完成了预定的目标，并取得了以下三方面的成果：(1) 研究了众包实体收集的计算模型与优化机制，通过自适应的工人选择与动态激励方案等技术，大幅度地提升了众包实体收集的质量；(2) 研究了众包关系发现的计算模型与优化机制，通过基于博弈论的任务协作机制，在保证准确率的前提下，显著地降低了众包成本；(3) 进行了众包知识融合的系统构建，并与企业合作进行了示范应用。课题组在理论上取得了突破，发表了一系列高水平的论文，并积极推动项目成果转化，与企业联合申请专利。