关联规则快速聚焦算法研究与实现

提出了一种实用的快速聚焦关联规则更新算法。在需要反复调整最小支持度的情况下,如何充分利用以往挖掘过程中的信息,避免多次扫描数据集;在如何有针对性的产生候选项集,从而减少候选集的规模;在如何提高候选项集的支持事务计数的效率等方面进行了研究,给出了算法的具体实现。通过分析,本算法是可行的。     

基于卷积树核的中文微博情感要素识别

情感要素识别是情感分析的关键子任务之一,其目的是识别出文本情感所作用的情感对象。文本情感要素识别属于最细粒度的情感分析,吸引了大量研究者的关注。中文微博由于其语言简短灵活、文本不规范、噪声较大等特点,给中文微博情感分析研究工作带来了新的挑战。目前大部分情感要素识别方法都是基于规则的方法或者基于扁平化特征的统计学习方法,区分噪声的能力不强,性能提升有限。针对中文微博的特点,提出一种基于卷积树核的情感要素识别算法,即首先对句子进行词性标注与依存关系分析,将句子中的名词作为候选情感要素;然后基于两种不同的修剪策略对依存树进行修剪,以获取每个候选情感要素的结构化信息;最后采用卷积树核计算依存树的相似度,并在此基础上识别句子中的情感要素。NLPCC2012和NLPCC2013中文微博情感分析评测任务中的实验验证了该方法的性能,其准确率相比于传统方法有显著提升。     

Empirical mode decomposition on skeletonization pruning

This paper presents a novel skeleton pruning approach based on a 2D empirical mode like decomposition (EMD-like). The EMD algorithm can decompose any nonlinear and non-stationary data into a number of intrinsic mode functions (IMFs). When the object contour is decomposed by empirical mode like decomposition (EMD-like), the IMFs of the object provide a workspace with very good properties for obtaining the object's skeleton. The theoretical properties and the performed experiments demonstrate that the obtained skeletons match to hand-labeled skeletons provided by human subjects. Even in the presence of significant noise and shape variations, cuts and tears, the resulted skeletons have the same topology as the original skeletons. In particular, the proposed approach produces no spurious branches as many existing skeleton pruning methods and moreover, does not displace the skeleton points, which are all centers of maximal disks.     

Resolving permutation ambiguity in correlation-based blind image separation

We address the problem of permutation ambiguity in blind separation of multiple mixtures of multiple images (resulting, for instance, from multiple reflections through a thick grass plate or through two overlapping glass plates) with unknown mixing coefficients. In this paper, first we devise a generalized multiple correlation measure between one gray image and a set of multiple gray images and derive a decorrelation-based blind image separation algorithm. However, many blind image separation methods, including this algorithm, suffer from a permutation ambiguity problem that the success of the separation depends upon the selection of permutations corresponding to the orders of the update operations. To solve the problem, we improve the first algorithm above by decorrelating the mixtures while searching for the appropriate update permutation using a pruning technique. We show its effectiveness through experiments with artificially mixed images and real images.     

Accelerating filtering techniques for numeric CSPs

Search algorithms for solving Numeric CSPs (Constraint Satisfaction Problems) make an extensive use of filtering techniques. In this paper¹ we show how those filtering techniques can be accelerated by discovering and exploiting some regularities during the filtering process. Two kinds of regularities are discussed, cyclic phenomena in the propagation queue and numeric regularities of the domains of the variables. We also present in this paper an attempt to unify numeric CSPs solving methods from two distinct communities, that of CSP in artificial intelligence, and that of interval analysis.     

Biomechanical assessment of new hand-powered pruning shears

The aim of the study was to assess the biomechanical strains on the hand-wrist system during grapevine pruning with newly designed pruning shears. Surface electromyography of the finger flexor muscle and wrist postures were analyzed in four vineyard workers during actual work with the new and reference hand-powered pruning shears. Due to the large inter-individual differences in biomechanical data observed with both pruning shears, no systematic effect of the new pruning shears on "extreme" flexion/extension (F/E) and ulnar/radial (U/R) wrist deviations was demonstrated. However, the use of the new pruning shears was associated with a higher frequency of "neutral and moderate" U/R deviations (U/R < 20%(max)) and, to a lesser extent, of "neutral and moderate" F/E deviations (F/E < 20%(max)). Due to the small sample of workers, this result must be confirmed in a larger population of vineyard workers.     

Pruning Algorithms for Rule Learning

Pre-pruning and Post-pruning are two standard techniques for handling noise in decision tree learning. Pre-pruning deals with noise during learning, while post-pruning addresses this problem after an overfitting theory has been learned. We first review several adaptations of pre- and post-pruning techniques for separate-and-conquer rule learning algorithms and discuss some fundamental problems. The primary goal of this paper is to show how to solve these problems with two new algorithms that combine and integrate pre- and post-pruning.