【作者】 张彤；

【导师】 曲明成；

【作者基本信息】 哈尔滨工业大学 ， 软件工程（专业学位）， 2020， 硕士

【摘要】 随着人类社会的不断发展进步,各行各业对于高精尖材料的需求也就日益增加,人类对于材料学的研究也在不断深入,而计算材料学领域就是人们利用计算方法在材料学上的探索。然而,随着研究的不断深入,材料学相关的研究数据规模越来越庞大,各个研究机构建立自己的材料信息管理系统,其存储结构和存储形式的多元性造成了数据结构的模糊性,调用、整合数据的复杂性。为了能够使数据具有逻辑性和可复用性,科学家们引入哲学中本体这一概念,对数据脉络和结构进行概括。领域本体主要是由极具本领域代表性的概念集、关系集构成,其中关系集中又包含了公理的概念,同时考虑到计算材料学是在材料学研究的基础上引入计算方法,则计算材料学领域的本体可以基于材料学领域本体进行扩展。为了实现这一目的,本文提出了一个改进的基于短语的主题模型框架。在这个框架中,本文提出一个新的频繁短语挖掘算法,结合词频阈值、词性等方法对原始文本数据进行分割和修饰,获取高频短语集。其次,框架中还包含了一个改进的基于短语的潜在迪利克雷分布主题模型,通过引入词性判断对短语中的词频进行二次统计,加重某一个单词对于整个短语进行主题分布概率的影响。通过这个改进的主题模型框架,获取到的具有代表性的短语集,即为计算材料学领域的概念集。对这个短语形式的概念集进行形式概念分析,获取到了该领域的关系集。通过领域专家的认证,利用获取到的属性集和概念集对材料学本体进行了概念、关系上的扩展。本文最后对收集到的9000余篇领域文献的标题和摘要进行实验和分析,实验结果与现有算法的比较结果证明了本文框架的可实践性和实用性。更多还原

【Abstract】 With the continuous development of human society,the demand for sophisticated materials in all walks of life is increasing,and human research on materials science is also deepening,and the field of computational materials science is people’s exploration in materials science by using computational methods.However,with the deepening of the research,the scale of research data related to materials science is getting larger and larger,and each research institution establishes its own material information management system.The diversity of storage structure results in the fuzziness of data structure and the complexity of data call and integration.In order to make data logical and reusable,scientists introduce the concept of ontology in philosophy to generalize the context and structure of data.The domain ontology is mainly composed of the most representative concept set and relation set in this field,among which the relation set contains the concept of axioms.At the same time,considering that computational materials science is based on the research of materials science,its domain ontology can be extended based on the domain ontology of materials science.In order to achieve this goal,an improved phrasebased topic model framework is proposed.In this framework,this paper proposes a new frequent phrase mining algorithm,which combines the word frequency threshold,part of speech and other methods to segment and modify the original text data to obtain the high frequency phrase set.Secondly,an improved phrase-based potential Dirichlet distribution topic model is included in the framework.By introducing the part of speech,the word frequency in the phrase is counted twice,and the influence of a certain word on the whole phrase topic distribution probability is aggravated.Through this improved thematic model framework,a representative phrase set is obtained,that is,a concept set in the field of computational materials science.The set of concepts in the form of this phrase is analyzed,and the set of relationships in this field is obtained.Through the authentication of domain experts,the text extends the concept and relation of the ontology of material science.In the end,this paper carries out experiments and analyses on the titles and abstracts of more than 9000 articles in the field.The comparison between the experimental results and the existing algorithms proves the practicability and practicability of this framework.更多还原