基于图的描述逻辑高效非标准推理技术的研究

Description logic is viewed as a formalism for representing knowledge representation systems, and providing the semantic foundation of OWL, which is a W3C recommended standard language for the semantic web. With the development of the Semantic Web and Knowledge Graph, description logics have attracted widely attention in academia and industry. In real applications, evelopment and evolution of ontologies are complex and error-prone. Thus, logical conflicts can easily occur in ontologies and lead to logical inconsistencies. Reasoning with inconsistent ontologies would then generate unacceptable conclusions. Therefore, non-standard reasoning in inconsistent ontologies has become an important issue for researchers. In this project, we study ontology debugging and ontology revision which guide to find the reason of inconsistency and eliminate the inconsistency finally. We further study the inconsistency-tolerance semantics which uderpins the valid reasoning of inconsistent ontologies. We first give theories for graph-based representation and provide different graph transformation rules for different description logics. In the meantime, we also consider the effect of partially ordered structure. In order to improve the efficiency of non-standard reasoning algorithm, a parallel reasoning method is introduced into our approach. Finally, based on the theoretical work and algorithms, we implement a graph-based prototype system for ontology non-standard reasoning.

描述逻辑是一类基于逻辑的形式化知识表示体系，为万维网组织W3C推荐的标准本体语言OWL提供语义基础。随着语义Web和知识图谱技术等相关技术的发展，描述逻辑吸引了众多研究人员的关注。在实际应用中，描述逻辑本体的构建与维护是一项复杂的过程，此过程常导致本体出现不一致性问题，从而使得标准推理服务失效，因而，面向不一致本体的非标准推理是研究中须面对的课题。项目将研究描述逻辑非标准推理中的本体修复技术，找出并解决本体中的不一致性问题，同时还将研究描述逻辑不一致容忍语义，实现在不一致存在情况下的有效推理。项目将研究描述逻辑本体到图的转换规则，考虑偏序结构对图转换的影响；为提高不一致推理的效率，将应用并行计算技术改进传统的推理算法，研究通用的并行非标准推理框架。最后，在前面研究的基础上，实现一个基于图的本体非标准推理原型系统。

描述逻辑本体中的不一致性问题将导致推理的失效，为解决该问题，我们提出了一种基于图的交互式本体匹配修复方法，该方法基于图的推理技术来完成对未判断匹配的自动更新，设计了基于图的影响函数来寻找最合适的匹配提供给专家判断，能在交互式过程中有效地减少专家判断匹配的次数，提供本体调试的效率，该方法能为知识图谱的构建提供调试工具；在信息过载的背景下，从拥有共同主题的多篇文档中挖掘核心概念和语义连接是一项重要的信息抽取任务，为有效提高信息抽取的效率，我们基于预定主题挖掘主题词，并改进文档的排序算法，借助于三元组实例过滤算法去除开放域数据中的噪声，最终提高了文档概念图主题概念覆盖率以及关系实例的兼容性。项目研究还涉及情感分析领域，通过时间使用者敏感的主题模型学习突发的情感感知主题；基于情感轮的文本情感分布学习，有效的利用情感先验知识，提高了算法的效率。在国家自然科学基金的支持下，我们共指导了九名硕士在读研究生，研究方向为大数据背景下的自然语言处理方法。