【作者】 黄水清；

【导师】 王凯；

【作者基本信息】 南京农业大学 ， 农业经济管理， 2010， 博士

【摘要】 科学知识,是人类科学活动的产出品,而科学文献则是记录、存贮、传播科学知识的载体,人类的科学知识一般以科学文献的数量计量。普赖斯在他的名著《巴比伦以来的科学》中揭示了科学知识与科学文献的指数增长规律。二战以来,以指数方式增长的科学知识和科学文献其数量已达到惊人的地步,这种现象被称为信息爆炸或知识爆炸。以期刊论文为例,2008年国内期刊当年发表的农业经济学期刊数量大约是1989年的17倍。在知识爆炸的时代,人类其实被自身创造的科学知识所淹没。人类科学知识的总量与人类的吸收能力之间的矛盾日益突出,科技工作者只能做选择性地阅读。任何一个农业经济学研究人员都无法通读当年的农业经济学文献。选择性地阅读带来的后果必然是对知识及知识与知识之间的联系的遗漏。即便是全面、完整地阅读,对隐含在海量文献之间的知识与知识之间的内在的、隐秘的联系仅凭人类个人的智力也很难发现。客观世界是普遍联系的世界,在海量的分裂状知识或科学文献之间存在着各式各样的联系,这些联系有相当一部分仅通过常规的查询与阅读是不能得到的。1986年,美国芝加哥大学的数学教授Don R. Swanson提出了非相关文献知识发现的思想,并以充分的文献证据论证了基于非相关文献的知识发现思想的可行性与正确性。Swanson的思想在学术界产生了广泛的影响,非相关文献知识发现既涉及知识发现技术,也涉及具体的学科领域。二十多年来,非相关文献知识发现成为计算机、情报学和知识发现领域学科(如医学、生物、工程、管理)的共同研究对象,相关研究论文在各学科的学术期刊均有发表。各个学科的研究人员包括Swanson本人在以后的二十多年中对非相关文献知识发现的理论基础、算法模型及具体应用进行了持续地研究,相继发现了更多的新知识、新联系,其知识发现应用领域也从单纯的医学、生物学领域拓展到了航空航天、工程技术乃至人文科学。农业经济是一门有自己的独特研究对象、研究方法、研究目的及悠久的历史与传统的成熟科学,与其它学科一样,现时代的农业经济也面临着专业细化、方向繁多的问题。农业经济学也无法避免Swanson所谓的知识片断化与分裂化倾向。即,农业经济学的知识总量与研究人员的吸收能力之间存在着巨大的差距；农业经济学内部的专业领域划分越来越细,专业之间的知识交流变得更加困难；一个专业领域的信息可能对另一个专业领域是有价值的,领域与领域之间存在着潜在的未被发现的关联。非相关文献知识发现方法及其在农业经济学中的应用研究,就是以Swanson的非相关文献知识发现思想为理论基础,以农业经济学科为学科领域,以国内核心期刊上发表的农业经济学论文为文献源,结合中文信息自动化处理的各项技术,研究适合农业经济学的非相关文献知识发现理论、算法和应用案例,寻找隐藏在现有文献间的关联概念,并以农业经济理论和方法论证关联概念的现实性,验证新知识的正确性。本文试图在两个方面取得突破。首先,在方法论方面,在分析、总结、比较非相关文献知识发现国内外现有研究成果与算法模型的基础上,本文试图结合国内期刊论文的特点和农业经济学的实际情况,为农业经济学文献设计一套逻辑上可行、操作上简便的知识发现方法,完整解决文献集合、词表体系、分词方案、算法模型等难点问题；其次,在知识发现实践方面,本文试图依据Swanson的非相关文献知识发现思想,以国内核心期刊上发表的农业经济学期刊论文为文献对象,发现若干对与农业经济有关的关联概念,也即发现新知识。本文首先在对国内外非相关文献知识发现的研究论文进行总结、归纳、分析的基础上,详尽地探讨了非相关文献知识发现思想模式下概念与概念之间的关联方式,针对Swanson的非相关文献知识发现的A-B-C模式和Gordon的中介文献思想,分别对应地提出了单一关联、单级关联、多级关联的概念,更进一步地,创建了多重关联和复合关联模型,给出了多级关联、多重关联和复合关联模式下非相关文献知识发现的开放式与闭合式知识发现过程的路径图。并借鉴在自然语言处理领域被广泛接受的向量空间模型,依据复合关联模式下非相关文献知识发现的开放式与闭合式过程路径图,设计了复合关联模式下非相关文献知识发现的算法模型,包括开放式算法模型与闭合式算法模型。然后,本文创建了农业经济学非相关文献知识发现语料库。首先以软件手段下载、转换得到用于非相关文献知识发现的农业经济学期刊论文题录数据库,该数据库包括了1989至2008年的二十年间国内所有核心期刊上发表的属于农业经济领域的全部研究论文,共87506篇,并建立了由停用词、通用词表、专业词表、同义词表构成的农业经济学非相关文献知识发现的词表体系。其中的专业词表是专门为农业经济学非相关文献知识发现而创建的,包括与农业经济有关的专业词汇96391个,设计并实现了用于文献集合自动分词的自适应分词算法。自适应分词算法在未登录词识别的召回率方面优于国内的其它算法,很适合非相关文献知识发现对专业领域的新词和概念重视程度较高的需求。接着,本文的研究进入农业经济学非相关文献知识发现实践阶段。通过自行编写的工具软件,实现了复合关联模式下基于向量空间模型的非相关文献知识发现算法,并以从维普期刊库中下载的题录为知识发现的文献集合,利用自行建立的词表体系,运行和实施了基于国内核心期刊农业经济学期刊论文的非相关文献知识发现过程,发现了世界贸易组织与原教旨主义、马尔萨斯与生态学、区域农业保险费率与地理信息系统三对关联概念。最后,遵循农业经济学的理论和逻辑验证并确认了这三对关联概念为农业经济领域的新知识。这些知识在此前从未引起国内农业经济领域研究人员的联想,在国内核心期刊公开发表的农业经济学文献中都没有报道,但这些知识隐藏在国内农业经济学期刊论文之间。本文的创新之处主要体现在以下三个方面：建立了非相关文献知识发现的复合关联模式。复合关联的概念与思想,在此前的国内外文献均未有类似提法。为了导出并描述非相关文献知识发现的复合关联模式,本文还分别提出了单级关联、多级关联、单一关联、多重关联的概念。单级关联、单一关联和多级关联的思想之前已经是研究者提出的成熟思想,但概念与提法为本文首次提出。本文将其纳入了复合关联的框架下并用新概念分别命名了对应的思想和方法。多重关联的概念与思想则是本文首次创立,已知文献中从未涉及。基于向量空间模型设计了复合关联模式下的非相关文献知识算法模型。本文基于向量空间模型设计了复合关联模式下的非相关文献知识发现的算法模型,并对算法效率进行了分析。该算法模型在Swanson及其它学者的论文中均没有涉及,文献调研中也没有发现相似或相近的算法。该算法模型效率较高,与文献集合做一次全局分析的时间代价相当。随后的程序实现和具体的农业经济学非相关文献知识发现实践也验证了该算法模型的可性行与合理性。该算法模型还充分考虑了汉语文献与农业经济领域的特殊性,解决了汉语文献自动分词以及农业经济领域专业词表与同义词的构建等问题。发现了三对隐含在文献中的关联概念,并证明这三对关联概念可作为国内农业经济学领域的新知识。以1989年至2008年的二十年间国内所有核心期刊上发表的与农业经济领域相关的期刊论文的篇名和摘要为对象,以自行开发的复合关联模式知识发现软件为工具,通过开放式和闭合式两阶段非相关文献知识发现过程的运行,本文发现了世界贸易组织与原教旨主义、马尔萨斯与生态学、区域农业保险费率与地理信息系统等概念之间的关联,并以农业经济学的理论与逻辑验证了这三对概念之间的关联的正确性,成为农业经济领域的新知识。检索国内文献,在目前能检索到的所有国内核心期刊上登载的农业经济学论文中对以上三对概念之间的联系从未涉及,相关文献为零。通过农业经济学非相关文献知识发现研究在方法论和知识发现实践两方面的工作,本文可以得出结论：在类似农业经济这样的人文社会科学领域,以国内的期刊文献为数据源,实现非相关文献的知识发现,是可行的,也是可以得出有价值的成果的,并能为农业经济学的研究提供新的方法与新的思路。此外,新知识发现是一个无止境的过程。本文在现有数据集的基础上发现了三条新知识,但这并不代表现有的数据集只隐含了这三条新知识。后续的研究者在现有数据集的基础上,经过细致的工作,一定可以挖掘、发现更多的农业经济领域的新知识。更多还原

【Abstract】 The Knowledge of science is the output of human’s scientific activity, and the literature is the carrier to record, store and transmit science knowledge. Generally speaking, science knowledge is counted by the number of the literatures of science. D. Price points out the exponential increasing law of science knowledge and the literature of science in his masterpiece "Science since Babylon". Since the Second World War, the phenomenon which called information explosion or knowledge explosion that describes the rapidly increasing amount of science knowledge and literature of science. Taking the domestic journals as an example, the number of the published Agricultural Economics literatures in 2008 is approximately seventeen times than the number in 1989. In fact, in the era of knowledge explosion, human beings are inundated by their own created knowledge. As a result, researchers have to select some literature to read because of the contradiction between the gross of science knowledge and human’s learning capacity. In addition, none of agricultural economics researcher can read through all of the agricultural economics literatures in the current year, which inevitably leads to the omitting of the knowledge and connection among the knowledge. Even read literatures completely and entirely, it’s still hard to discover the internal and covert relations among the knowledge.There are different kinds of relations among different science knowledge in this world. We can’t realize them only by routine searching or reading. In 1986, Don R. Swanson, who is the professor of mathematical science at the University of Chicago, first demonstrates the correctness and feasibility of non-interactive literature-based knowledge discovery based on sufficient literature proof. Swanson’s idea has a great influence on this academic world and non-interactive literature-based knowledge discovery includes not only KDD technique but also concrete domain fields. For over twenty years, researches from computer science、information science and other knowledge discovery fields such as Medicine、Biology、Engineering or Management all have concentrated on non-interactive literature-based knowledge discovery and circulated some related papers in academic journal. After that, researches of each subject as well as Swanson himself keep on studying the theories, models and applications of this idea. More and more new knowledge and relations are discovered, and the knowledge is extended from Medicine and Biology to the other areas like Aviation, Engineering and even Humanities and Social Sciences.Agricultural Economics is a mature science with her own research object、methods、 purpose and long history. Like other subjects, it also faces the problems with segmenting of subject and various directions of research. As a result, Agricultural Economics has no way to avoid the trend of knowledge fragmentation and abruption as Don R. Swanson points out. In all, there is a huge gap between the massive knowledge and the researchers’learning capacity. When the inner professional fields of Agricultural Economics are segmented, the knowledge communication among these specialties becomes harder to achieve. However, information from one professional field may be valuable to another one, and there maybe existing a potential relation which has not been discovered among field. This study is based on Don R. Swanson’s non-interactive literature-based knowledge discovery; supported by agricultural economics literature resource from domestic Core Journal; together with the information processing technology of Chinese and studying the non-interactive literature-based knowledge theory、algorithm and application cases, in order to find out the concept relations behind these literatures, furthermore, proving the correctness of new knowledge with this theory and method.This paper tries to make breakthrough in two aspects. First, in the methodology aspect, this paper is based on analyzing, summarizing and comparing the existing research achievement and algorithm models home and abroad, trying to design a logical and simple knowledge discovery method based on characteristics of National Journal Papers and Agricultural Economics, to solve the problems of literature aggregate, vocabulary scheme, participial scheme and algorithm model. Secondly, in the aspect of knowledge discovery practice, this paper is also on the basis of Don R. Swanson’s non-interactive literature-based knowledge discovery, using these published Agricultural Economics literature to find out several correlative conceptual-pair of Agriculture Economics, at last, discovering new knowledge.Firstly, the paper is on the basis of summarizing, concluding and analyzing non-interactive literatures at home and abroad, and then detailed discusses the model of linkage among conception in the non-interactive literature-based knowledge discovery thought pattern and then put forward single-stage relation, multi-stage relation, single relation, multiple relation according to Don R. Swanson’s A-B-C model and Gordon’s mediated document. Further more, multiple relation and compound relation mode are established. Finally, using Vector Space Model (VSM) which has been widely accepted in the natural language processing field, we designed an open and a closed non-interactive literature-based knowledge discovery system and algorithmic model.Then, this paper built a Chinese agricultural economics non-interactive literature-based knowledge discovery corpus. First through software download and conversion, it set up Chinese agriculture economics literature title database, which included all titles of 87506 pieces of research papers published in domestic agriculture economics core journals from 1989 to 2008. Then this paper created a thesaurus system including stop word list, general word list, synonyms list, professional word list which covered 96391 words related to agriculture economics. Also the author designed and implemented self-adapting segmentation algorithm for segmenting literature, which was better than other algorithms in recognizing and recalling unknown words and suitable for knowledge discovery from non-interactive literature.Finally this paper entered the practice period. Through the self- developed software, the non-interactive literature knowledge discovery algorithm based on vector space model in compound relation mode was put into practice. Taking the literature titles from vip Chinese journal database as the object and using the self-built thesaurus system, the author implemented the non-interactive literature knowledge discovery and discovered three pairs of relative concepts. According to agricultural economics theories, WTO and fundamentalism, Malthus and ecology, regional agricultural insurance premium rates and geographic information system (GIS) were logically proved new knowledge in this field. The new knowledge did not appear in researchers’minds and in domestic agricultural economics core journal papers, but hidden in these papers.There are three innovations in this paper as follows:It built compound relation mode for non-interactive literature-based knowledge discovery.The concept and idea of compound relation had not been similarly brought forward before in the literature at home and abroad. For deriving and describing compound relation mode, this paper put forward single-stage relation, multi-stage relation, single relation, multiple relations. Although the ideas of single-stage relation, single relation, multi-stage relation were mature, their concepts were first presented in this paper. This paper interfiled them in the framework of compound relation and explained their corresponding ideas and methods when elaborating concepts. The author built the concept and idea of compound relation for the first time, which was not found in the existing literature.It designed the non-interactive literature-based knowledge discovery algorithm model based on vector space model in compound relation mode.The author designed the non-interactive literature-based knowledge discovery algorithm model based on vector space model in compound relation mode and analyzed its algorithm efficiency. Dr. Swanson and other researchers had not put forward it, and similar or related algorithms were not found in the literature. This algorithm model was efficient and took almost the same time as the overall analysis for literature collection. The author validated its feasibility and rationality in the follow practice. Considering the particularity of Chinese literature and agricultural economics, this algorithm model successfully realized Chinese automatic word segmentation and established agriculture economics professional word list and synonyms list and so on.This paper discovered three pairs of relative concepts hidden in the domestic literature and proved them to be new knowledge in the Chinese agricultural economics field.In this paper, the author designed a new algorithm of complex connection model of non-interactive literature-based KDD based on vector space model (VSM).Applied non-interactive literature-based knowledge discovery methods in Chinese social sciences literature. Using the literature sources of agricultural economics in the twoL decades from 1989 to 2008 between the domestic core journals, did the computing process of non-interactive literature-based knowledge discovery by the help of the software we developed, and then selected three pairs of the related concept--’WTO’ and’fundamentalism’,’Malthus’and’ecology’,’regional agricultural insurance premium rates’and’GIS’. The results of verification showed that the relationship between these concepts was correct, which could be used as the new knowledge of agricultural economics. Through retrieving domestic agricultural economics core journals, it was found that nobody had mentioned the relations between three pairs of concepts in domestic literature so far.By studying on methodology and practice of non-interactive literature-based knowledge discovery in Chinese Agricultural Economics, we could make conclusions that it is feasible to realize non-interactive literature-based knowledge discovery on field of Humanities and Social Sciences, just like Chinese Agricultural Economics based on the literature collection of domestic journal papers. And also the discovered knowledge is valuable for the development of this field. Non-interactive literature-based knowledge discovery would can provide new methods and new ideas for agricultural economics research.Moreover, new knowledge discovery is a eternal process. This paper discovers three pairs of relative concepts, but that does not mean there are only three pairs of relative concepts based on the existing literature collections. In the future, researchers could discover more new knowledge by microMeSH work based on the existing literature collections in the agricultural economics.更多还原