【作者】 刘刚；

【导师】 黄少滨；

【作者基本信息】 哈尔滨工程大学 ， 计算机应用技术， 2008， 博士

【摘要】 在当前国家建设和谐社会的背景下，控制国民经济命脉的各领域信息系统越来越受到重视。需求分析是信息系统开发的起点，对整个开发的成败有着重要的影响。当系统需求存在缺陷时，无论系统设计和实现过程是否完善，实现的系统都必然存在缺陷。如何验证信息系统需求与国家法律政策、行业法纪法规的一致性，以及验证信息系统需求内部的一致性，已成为需求工程研究领域的一个重要研究课题。本文的研究工作主要围绕需求一致性验证技术展开，重点研究基于自然语言需求文本的需求一致性验证技术。当需求以自然语言形式存在时，直接应用各种自动化方法进行需求检查是不可行的，必须先对需求进行形式化。本文以自然语言需求子句为树结点构建了一种需求树模型。模型基于给出的汉语语言需求词汇切分算法，将需求子句划分成若干需求关键词，并提出了一种需求关键词七元组表示方法。通过构建领域规范词表，确定了七元组表示的规范词和外语义相似度。关键词七元组表示的提出使需求树结点从需求子句转化成了需求关键词集，所构成的细化需求树模型实现了对自然语言需求文本的预处理操作。提出了一种需求依存分析树模型，以解决细化需求树不能表征需求子句句法结构和需求关键词词间关系问题。基于给出的一种基于分析树的依存关系标注算法，论述了一种从细化需求树结点到需求依存分析树模型的转换方法，并提出了一种需求依存分析树的建树与更新算法，从实验结果看，算法优于同类算法。模型转换过程中通过词性回收进行了需求关键词去词操作，并讨论了转换过程中需求二义性的发现和消除方法。需求依存分析树模型可精确的描述需求子句句法结构和词间依存关系。提出了一种概念从属森林模型，以解决需求依存分析树模型不善于表达隶属于不同需求子句的词间关系问题。并提出了一种概念从属森林构建与更新算法，并给出了在更新过程中识别需求遗漏性的方法。经实验分析，算法具有接近线性的时间复杂度。归纳了森林操作，给出了一种森林操作提取算法，算法对需求依存分析树集进行遍历和二阶动词子树提取，所提取的森林操作函数可用于表征隶属不同需求子句的词间关系。提出了需求关键度的概念和计算方法，以提高需求检查结果的确认率。构建了外语义相似度向量，基于依存分析树模型给出了一种局部重要性比例标度填充算法，并依据填充内容构建了判断矩阵，提出了一种需求关键词权重向量的计算方法。基于概念从属树模型给出了一种全局重要性比例标度填充算法，使需求关键词七元组数据满秩，并基于填充内容给出了需求关键词重要性标度向量的计算方法。基于以上三种向量的构建，提出了一种需求关键度计算方法，并提出了一种基于需求关键度的需求树与需求依存分析树剪枝算法。算法不但能提高需求检查结果的确认率，而且缩小了算法状态空间，提高了各种需求检查算法的执行效率。提出了需求相似度和相关度的概念和计算方法，以解决概念从属森林模型不能直接表征不同需求子句句间关系的问题。基于概念从属树模型，给出了对从属结点概念从属强度和非从属结点内语义相似度的计算方法，并进一步给出了一种用于判定需求子句概念结构相似程度的需求向量相似度计算方法。基于需求依存分析树的动词骨架，提出了一种用于判定需求子句句法结构相似程度的需求依存相似度计算方法。基于概念从属森林操作模型，提出了一种需求相关度计算方法。基于相似度和相关度计算，提出了一种需求一致性检查算法，算法通过对相似属性的属性值、属性值域和相关属性的属性等式结果进行比较，可确认不同需求子句的句间关系，并且能检测出需求冗余性和矛盾性等需求不一致性的存在。更多还原

【Abstract】 At the background of the national construction of a harmonious society,thelifeline of information systems which control the national economy is receivingincreasing attention.Requirement analysis is the starting point of an informationsystem’s development,on which the success or failure of the entire developmenthas a significant impact.When requirements have their own shortcomings,whether the system design and implementation process is perfect or not,thesystem shortage will be there.It has become an important research field inrequirement engineering area that how to validate the consistency between systemrequirements and national laws,and the internal consistency within systemrequirements.This thesis’s research work mainly focuses on the requirementconsistency validation,especially on the validation to a natural languagerequirement text.This thesis proposes a requirement tree model,which is a branch-levelstructure with the nodes of Chinese language requirement clause.Whenrequirements exist in form of natural language requirement text,it must beformalized first.Based on a given Chinese language requirement vocabularysegmentation algorithm,the thesis divides requirement clauses into a series ofrequirement keywords.Further more,a method of requirement keywords withseven tuples denotation come up.The thesis proposes to fill the regular words andouter semantic comparability of the seven tuples by constructing the domainregular vocabulary table.The keyword seven tuples denotation transforms therequirement tree nodes from the requirement clauses to requirement-keyword sets,and constitutes a refinement requirement tree model,which completes thepreprocessing operation to natural language requirement text.As the refinement requirement tree cannot characterize the acceptationrelation between the syntax structure and the requirement keywords of arequirement clause,this thesis proposes a requirement-interdependence-parse treemodel(RIPT).Based on a given labeling algorithm based on interdependence parsing tree model,the thesis elaborates a model transformation method which isfrom a refinement requirement tree node to a requirement interdependence parsingtree,and proposes an establishing and updating algorithm of the RIPT.In themodel transforming process,it carries on a word-off operation of the requirementkeywords through the lexical category recycling,and it also discusses therequirement ambiguous discovery and elimination method in the modeltransforming process.The RIPT can accurately describe the dependence relationamong syntax structure and the keywords.Since the RIPT is not a good exPression on words relationship whichsubordinates in different requirement clauses,this thesis proposes a conceptsubordinate forest model(CSFM).The thesis proposes a model establishing andupdating algorithm,and shows a method to distinguish requirement omission inthe updating process.It builds a forest operation extraction algorithm by inducingthe forest operation,which carries on the traversal and the second-order verbsubtree extraction to the requirement interdependence parsing tree set,and theextracted forest operation function may be used in attributing the relationshipbetween different subordination requirement clauses.In order to enhance the confirmation rate of requirement validation result,this thesis put forward the concept of requirement crucial degree.The extractionof the fifth item of keyword seven tuples is performed to construct outer semanticsimilarity vector.First of all,the thesis proposes a local important proportion scalefilling algorithm which was based on the RIPT,and constructs judgment matrixbased on these operations,then shows a method to calculating the weight vectorof requirement keywords.Then,the author statements a global importantproportion scale filling algorithm,which is based on the CSFM,and causes therequirement keywords seven tuples data non-singular,shows a calculating methodof the requirement keywords important scale vector based on these operations.Based on the above three kind of vectors,the thesis proposes a requirementcrucial degree calculating method,and requirement tree,and a requirementinterdependence parsing tree cut-off algorithm based on the requirement crucialdegree.Not only the algorithm can enhance the confirmation rate of the validation result,moreover as a result of state space reduction,also will enhance eachefficiency validation algorithm which have been carried out.As the CSFM cannot characterize the relationship among differentrequirement clauses directly,this thesis proposes the concept of requirementsimilarity degree and requirement pertinence degree.Based on the CSFM,theauthor proposes the calculating method to concept subordinate intensity ofsubordinated nodes and the inner semantic similarity degree of non-subordinatenodes.And based on the calculating of inner semantics similarity and therequirement keywords weight vector,the thesis shows how to calculate therequirement vector similarity degree,which can judge the concept structuresimilarity degree between two requirement clauses.Based on the RIPT’s verbskeleton,it shows a calculating method of requirement interdependence similaritydegree,which is used in determining the similarity between two requirementclauses’ syntax structure.Based on the forest operation of CSFM,the thesisproposes a calculating method of requirement pertinence degree.Based on thesimilarity degree and the pertinence degree,it proposes a requirement consistencyvalidation algorithm.Through comparing the attribute value,the attribute valuerange,attribute equation among similar attributes,the algorithm not only confirmsthe relationship among different requirement clauses,but also can inspect therequirement inconsistency existences such as the redundancy and thecontradiction.更多还原