事件抽取是自然语言处理领域的重要研究方向.传统的事件类型抽取系统采用分类方式,无法解决跨句子的事件角色和事件类型匹配问题.为了解决该问题,提出一种基于序列标注的事件联合抽取模型,结合卷积神经网络(convolutional neural networks,CNN)与长短期记忆网络(long short-term memory,LSTM)提取全局特征和局部特征;并在浅层LSTM层共享参数实现联合抽取,以序列标注方式抽取事件论元并匹配事件类型.实验结果表明,模型能有效提取司法领域的文档事件信息.

在粒计算理论中,通过不同的粒计算机制可以生成不同的粒结构.在粗糙集中,对于同一个信息表而言,通过不同的属性添加顺序可以得到由不同的序贯层次结构,即粗糙粒结构.在粗糙粒结构中,不同的属性获取顺序导致了对不确定性问题求解的不同程度.因此,如何有效评价粗糙粒结构是一个值得研究的问题.本文将从知识距离的角度研究这个问题.首先,在前期工作所提出的知识距离框架上提出了一种粗糙近似空间距离,用于度量粗糙近似空间之间差异性.基于提出的知识距离,研究了粗糙粒结构的结构特征.在粗糙粒结构中,在对不确定性问题进行求解时,本文希望在约束条件下可以利用尽可能少的知识空间使不确定性降低达到最大化.基于这个思想并利用以上得出的结论,在属性代价约束条件下,引入了一个评价参数λ,并在此基础建立了一种粗糙粒结构的评价模型,该方法实现了在属性代价约束条件下选择粗糙粒结构的功能.最后,通过实例验证了本文提出的模型的有效性.

粗糙集的近似集用已有知识粒对不确定性目标概念进行近似描述,但在构建近似集时并没有考虑数据的代价信息这一实际因素.对此,首先分析在构建粗糙集的近似集时考虑代价信息的必要性;然后,从代价敏感角度构建误分类代价的粗糙集近似集模型,并分析该模型下求得的近似集的相关性质.为了在多粒度空间中寻找一个合适的粒度空间来对不确定性目标概念进行近似描述,使误分类代价与测试代价之和尽可能小,给出属性代价贡献率的定义,并提出一种代价敏感的粒度寻优算法.实验结果表明,所提出算法能适用于现有代价认知场景,并在给定代价场景下求出合理的层次粒度空间结构以及不确定性目标概念的近似集.

目前面向分类的差分隐私保护算法中,大部分都是基于决策树或者随机森林等树模型.若数据集中同时存在连续数据和离散数据时,算法往往会选择调用2次指数机制,并且进行隐私预算分配时往往选择平均分配.这都使得隐私预算过小、噪声过大、时间成本增加以及分类准确性降低.如何在保证数据隐私的同时尽可能地保证数据可用性,并提高算法性能,成为目前差分隐私保护技术研究的重点.提出了面向决策树和随机森林的2种差分隐私保护数据挖掘算法,使用Laplace机制来处理离散型特征,使用指数机制处理连续型特征,选择最佳分裂特征和分裂点,并采用最优特征选择策略和等差预算分配加噪策略.对金融数据集的测试结果表明,提出的基于树模型的差分隐私保护算法都能在保护数据隐私的同时,具有较高的分类准确性,并且能够充分利用隐私保护预算,节省了时间成本.

众所周知,经典粗糙集的不确定性来自于边界域,但是对于粗糙模糊集来说,其正域和负域中的元素存在不确定性,从而导致粗糙模糊集的不确定性不仅来自于边界域,还来自于正域和负域.另外,在粗糙模糊集中,一个模糊概念可以通过层次粒结构中不同的粗糙近似空间进行刻画,随着粒度的变化,模糊概念的不确定性的变化规律如何?对此,文中提出一种基于模糊度的不确定性度量公式,并基于均值模糊集分析了粗糙模糊集模型,得出粗糙模糊集不确定性度量的模型同样适合于度量概率粗糙集的不确定性的结论.其次,采用基于模糊度的不确定性度量方法,揭示了分层递阶的多粒度空间下粗糙模糊集不确定性的变化规律.然后,分析了3个域(正域、边界域和负域)的不确定性,并揭示了它们在分层递阶的多粒度空间下的变化规律.最后,通过实验验证了所提不确定性度量理论的有效性.

垃圾邮件过滤是信息时代的一个重要研究课题,一封重要邮件被错分会产生不可估量的代价.因此,如何提高过滤器的性能成为垃圾邮件过滤领域中的核心问题.目前,业界通常采用机器学习算法中的二分类模型以处理垃圾邮件过滤问题.然而,较之于三支决策模型,二分类模型会产生较大的错分代价.作为三支决策的一个重要分支,基于决策理论粗糙集的三支决策模型符合人类认知习惯,且能有效地降低错分代价,进而提高过滤器的性能.然而,在构造损失函数时,少有研究考虑由于等价类之间的差异性而对分类结果带来的影响.因此,在基于决策理论粗糙集的三支决策模型的基础上,提出了一种基于相似度量的自适应三支垃圾邮件分类模型.该模型根据集合方差计算了条件属性的权重,并基于相似度量建立了一种刻画差异信息的综合评价函数,进而根据贝叶斯决策规则构建了一种计算自适应阈值对的方法.实验结果表明所提模型在垃圾邮件过滤领域表现优异.

南海东部珠江口盆地疏松砂岩油藏埋藏浅,泥质含量高,易出砂.为优选适用于该油田水平井的防砂筛管,采用实尺寸筛管防砂物理模拟实验装置,结合该油田储层特征(d50=250 μmn,非均匀系数3.17,泥质12％)和实际工况,进行了不同类型防砂筛管的防砂物理模拟实验,通过测定不同防砂精度(150、200、250、300 μm)条件下的流量、压降、油中含砂量、产出砂粒径等参数,系统评价了改进型星孔复合筛管及其他3种不同结构的金属网布复合筛管防砂性能、抗堵塞性能和综合性能.结果 表明:金属网布复合筛管Ⅱ(保护套+斜纹网+双层密纹网+斜纹网+基管)和金属网布复合筛管Ⅰ(保护套+斜纹网+三层密纹网+斜纹网+基管)防砂性能、抗堵塞性能以及综合性能指标都较高,适合作为南海东部珠江口盆地疏松细～中砂岩油藏高含水期水平井防砂筛管,并推荐油田防砂精度为200～250 μm.该研究可为该油田进行防砂方案设计提供依据.

In this paper, a Boolean determinant is introduced, and some properties of the determinants are discussed in Boolean algebra. Next, applying with the properties of characteristic determinant of directed graph, a new Hamiltonian-cycles decision theorem and a new necessary and sufficient conditions for a Hamiltonian directed graph are studied, and a new Hamiltonian-cycles decision algorithm is established. Finally, an example is provided. The obtained results seem to be general in nature.

The rank of a dependent fuzzy set in a closed fuzzy matroid is discussed and several useful results for the rank of a dependent fuzzy set are given in this paper. Based on these theorems, an algorithm of obtaining the rank of a dependent fuzzy set is presented.

Pawlak教授提出的粗糙集理论是解决集合边界不确定的重要手段,他构建了边界不确定集合的两条精确边界,但没有给出用已有知识基来精确或近似地构建目标概念(集合)X的方法.在前期的研究中提出了寻找目标概念X的近似集方法,但并没有给出最优的近似集.首先,回顾了集合间的相似度概念和粗糙集的近似集Rλ(X)的构建方法,提出并证明了Rλ(X)所满足的运算性质.其次,找到了Rλ(X)比上近似集R(X)和下近似集R(X)更近似于目标概念X的λ成立的区间.最后,提出了R0.5(X)作为目标概念的最优近似集所满足的条件.

传统的模糊C均值聚类(FCM)算法具有简单、稳定和高效等特点,但在噪声点较多的情况下容易受噪声影响,使得算法效率降低.文章结合变精度粗糙集模型,提出一种改进的FCM算法,该算法利用变精度粗糙集模型刻画不确定集合上近似集和下近似集的原理,将经过聚类算法后的类簇边缘范围中的对象根据变精度粗糙集的阈值特性划分为正域、负域、边界域三个部分,使得聚类的准确率得到提升.仿真实验结果表明该算法使得聚类结果更加清晰,在边界域较模糊的情况下聚类准确率比传统FCM算法有一定的提高.

Generally, when talking about attribute reduction of a decision table, it usually keeps the positive region unchanged based on the Pawlak’s rough sets theory. However, the needs may be different for different precision of the reduction in real life as well as the actual cost to obtain attribute values and personal preferences. Based on the risk of personal preference for the subjective aspect, the accuracy of reduction, the actual cost of obtaining attribute value, and the risk of interval misjudgment for the objective aspects, a novel attribute reduction algorithm is proposed. Then, the relationship between the reduction cost and the reduction accuracy is discussed. Based on the genetic algorithm, a heuristic method for searching the local optimal reduction subset is proposed. Simulation experiments show that the algorithm is feasible, and more realistic to deal with practical decision-making problems.

Rough set describes an uncertain target set with upper and lower approximation sets,and approximation set of rough set uses 0. 5-approximation set as an approximation set of the uncertain target set.In this paper,we firstly find that the theory of attribute reduction algorithm based on similarity between target set and its 0. 5-approximation set is still incom-plete,and this similarity is not sensitive to changing granularities.In order to overcome above shortcomings,the change rule of similarity with changing granularities in a multi-granularity space is analyzed,fuzzy degree of approximation set is de-fined,and the change rules of this fuzziness with changing granularities are analyzed in detail in a hierarchical space.Finally, a new attribute reduction algorithm is proposed.From a new perspective,a kind of differentiation measure between an uncer-tain target set and its approximation set is presented.

The basic idea of granular computing is to solve complicated problems in different granularity levels. To a great extent, the method indicates human intelligence in the solving process. Based on the human multi-granularity mechanism for solving complicated problems and the principle of probability statistics, an efficient multi-granularity search model based on statistical expectation is proposed from the viewpoint of granular computing. Then, the variety rule of the expectation in quotient spaces with different granularities is analyzed in detail. The experimental results demonstrate that with the granules divided into many sub-granules, the efficiency of the proposed method is gradually reduced and tends to be stable for searching a certain target. In addition, the complexity for solving the problem can be greatly reduced in different probability models.

The human cognitive process is very complicated,and the bidirectional cognitive computing is one of the key problems for studying intelligent cognition.In this paper,the p-order normal cloud model and its properties of the bidirectional cognitive computing are studied in detail.On the basis of normal cloud proposed by professor Li De-Yi,the recursive definition of p-order normal cloud model is proposed firstly,and the properties of the certainty-degree of a2-order normal cloud and the reflected cognitive characteristics are analyzed.Secondly,the joint probability density function of cloud drops and their certainty-degree which was presented by Li De-Yi et al.is corrected.Finally,the properties of the certainty-degree of a p-order normal cloud and cognitive characteristics are also analyzed.It will contribute to the development and application of normal cloud model in bidirectional cognitive computing and uncertain information processing.

以粗糙集理论为基础分析彩色图像分割概念,研究已有的彩色图像分割方法,将粗糙集的上、下近似集理论与云模型理论相结合,提出基于粗糙集和云模型的彩色图像分割方法.该方法在HSV颜色空间对彩色图像进行非均匀量化,并寻找量化后图像的基本直方图和Histon直方图,根据粗糙集理论中粗糙度概念得到图像的粗糙直方图,最后通过云模型的“3En规则”对图像进行前景/背景分割.通过三组实验验证了该方法的正确性,并与K均值算法、A-IFS HRI算法进行比较,实验结果表明了该方法对彩色图像分割的有效性.

In order to emphasize the importance of practical teaching,we have compared the contents of discrete mathematics between several countries.We find out practical teaching of discrete mathematics plays an important role in improving students’logical thinking ability,operating ability,innovation ability and studying interest and so on.Hence,we put forward some strategies in practical teaching of discrete mathematics.

该文对现有Vague集(值)相似度量方法进行研究,发现目前Vague值(集)相似度量方法存在一些缺陷,给出了Vague集(值)相似度的一种定义,提出了两种Vague值(集)相似度量的改进方法,最后通过数据分析验证了改进方法的有效性和优越性.