【作者】 李冬梅；

【导师】 黄厚宽；

【作者基本信息】 北京交通大学 ， 计算机应用技术， 2014， 博士

【摘要】 近年来,语义Web技术不断发展,本体作为一种清晰表达语义和知识共享的方式,成为了语义Web的核心,其相关研究也得到了很大的进步。然而,在实际应用中,很难构建没有错误的本体,引起本体不一致的原因有很多。本体构建者对知识认识的不足或错误可能导致不一致,本体描述或者分类上的含混、语义上的冲突可能引起前后不一致,本体迁移或本体合并、集成等操作也很容易使本体产生不一致。在经典推理下,不一致本体可以演绎出任何结论,针对这样的本体进行推理是毫无意义的。因此不一致处理成为本体研究的关键问题,得到学术界的高度关注。本文在全面分析本体不一致问题研究现状的基础上,重点对本体不一致性的度量方法、不一致本体的诊断和修复方法、不一致本体的推理方法等若干关键问题进行了深入探讨,最后利用自行构建的林业领域本体,对研究成果进行了实证分析。主要贡献包括：(1)本体不一致性的度量方法的研究。首先提出一种基于证据理论的本体不一致性度量ETOIM方法,并通过实例验证了该方法的有效性。然后在ETOIM方法基础上,利用语法相关性,提出了另外一种更加有效的基于选择函数的证据理论本体不一致性度量SETOIM方法,SETOIM方法的实验结果比ETOIM方法的实验结果更接近“事实”。同时将这两种方法与当前具有代表性的类似算法进行了比较,结果表明我们提出的算法是有效的。(2)不一致本体的诊断和修复方法的研究。提出一种比传统的局部诊断方法效率更高的不一致本体的诊断修复方法,即证据验证诊断方法。该方法能够利用系统提供的新证据首先可将“无罪者”(“无辜者”)元素排除,缩小了诊断范围；然后再次利用新证据可将“定罪者”元素“挖”出,进一步缩小了诊断范围；最后对剩余部分进行诊断修复。这种证据验证诊断方法因诊断范围的两次缩小提高了诊断效率,另外,这种做法在“挖”出“定罪者”元素之后,便可以避免因为这些“定罪者”元素而“殃及无辜”的问题,进一步提高了诊断修复的准确率。(3)不一致本体的推理方法的研究。在研究基于线性递增策略的不一致本体推理方法的基础上,提出一种新的基于线性递减策略的不一致本体推理方法。线性递增的不一致推理适用于一致性程度高的本体,线性递减的不一致推理适用于一致性程度低的本体。但这类单纯依赖于语法相关选择函数的不一致推理,在两次线性变换之间的推理空间相差较大,容易漏失有意义的答案。我们将这两种单纯依赖于语法相关选择函数的不一致推理统称为“粗粒度”不一致推理。进而,作为“粗粒度”不一致推理的补充和完善,我们将不一致性的度量结果用于推理,给出一种“细粒度”推理策略,这种推理方法能在一定程度上提高推理的有效性。(4)设计了基于林业领域本体的林业知识资源管理服务系统框架,该系统包括林业领域本体构建模块、本体不一致处理模块和语义检索模块三部分。其中的本体不一致处理模块对本文提出的不一致性度量方法、诊断修复方法和不一致推理方法进行了逐一验证。更多还原

【Abstract】 As a clear expression of the semantic and knowledge sharing, ontology has become the core of the semantic Web. In recent years, research on ontology has been great progress. However, it is often difficult to construct an ontology which is error-free in practice. Inconsistency can occur due to several reasons, such as mis-presentation of defaults, polysemy, merging ontologies, migration from another formalism. The classical entailment in logics is explosive:any formula is a logical consequence of a contradiction. Conclusions drawn from an inconsistent ontology by classical inference may be completely meaningless. Therefore, studies on how to deal with inconsistent ontologies have become the academic focus of attention.This paper makes an overall analysis of the present situation in inconsistent ontology, and then mainly provide solutions to measuring inconsistency, diagnosing inconsistency and reasoning with inconsistency. Finally, an empirical analysis is made of the research models with forestry domain ontology constructed by us.(1) Measuring inconsistency. Firstly, this paper proposes an ontology inconsistency measures method named ETOIM and proves the correctness of the result. Secondly, a more effective selection function-based inconsistency measures method named SETOIM is presented. SETOIM improves ETOIM by using syntactic relevance selection functions, which is closer to the the "fact" than ETOIM. Compared with the typical similar algorithms in the field, the experimental results demonstrate our methods is effective.(2) Diagnosing and reparing inconsistency. We propose a more effective approach to diagnose and repair inconsistency than traditional local diagnosis. Our approach can exclude some "innocents" and some "condemners" from diagnosis range according to new evidence. Because the diagnosis range is narrowed twice, efficiency is increased significantly. Furthermore, our approach can avoid involving the innocent in the trouble by removing the "condemners", and the accuracy of diagnosis is improved.(3) Reasoning with inconsistency. Based on the linear extension reasoning algorithms, this paper proposes a framework of reasoning with inconsistent ontologies. The former is fit to the ontologies with the lower degree of inconsistency, and latter is fit to the ontologies with the higher degree of inconsistency. Both of the reasoning methods are only based on a syntactic relevance-based selection, which makes the inference space between two linear transformations larger. Hence, meaningful answers are easily losed. Both of the reasoning methods are referred to as the "coarse grained" reasoning. Accordingly, the "fine grained" reasoning based on the measure results is presented to improve the "coarse grained" reasoning, so as to increase the efficiency of reasoning.(4) Developing forestry knowledge resource management sevice system based on forestry ontology. The system consists of three parts:forestry domain ontology construction module, ontology inconsistent processing module and the semantic retrieval module. Our new approaches for dealing with inconsistencies are verified one by one in ontology inconsistent processing module.更多还原