COTS-based systems top 10 list

Presents a COTS-based system (CBS) software defect-reduction list as hypotheses, rather than results, that also serve as software challenges for enhancing our empirical understanding of CBSs. The hypotheses are: (1) more than 99% of all executing computer instructions come from COTS products (each instruction passed a market test for value); (2) more than half the features in large COTS software products go unused; (3) the average COTS software product undergoes a new release every 8-9 months, with active vendor support for only its latest three releases; (4) CBS development and post-deployment efforts can scale as high as the square of the number of independently developed COTS products targeted for integration; (5) CBS post-deployment costs exceed CBS development costs; (6) although glue-code development usually accounts for less than half the total CBS software development effort, the effort per line of glue code averages about three times the effort per line of developed applications code; (7) non-development costs, such as licensing fees, are significant, and projects must plan for and optimize them; (8) CBS assessment and tailoring efforts vary significantly by COTS product class (operating system, database management system, user interface, device driver, etc.); (9) personnel capability and experience remain the dominant factors influencing CBS development productivity; and (10) CBS is currently a high-risk activity, with effort and schedule overruns exceeding non-CBS software overruns, yet many systems have used COTS successfully for cost reduction and early delivery     

On one integral Volterra model of developing dynamical systems

We develop an approach for constructing continuous solutions of integral Volterra equations of the first kind that arise in modeling developing dynamical systems. We give existence theorems for the solutions and study their asymptotic behavior. Theoretical results are illustrated with numerical computations on test examples.     

On linear list recursion in parallel

We propose a classification for a set of linearly recursive functions, which can be expressed as instances of a skeleton for parallel linear recursion, and present new parallel implementations for them. This set includes well known higher-order functions, like Broadcast, Reduction and Scan, which we call basic components. Many compositions of these basic components are also linearly recursive functions; we present transformation rules from compositions of up to three basic components to instances of our skeleton. The advantage of this approach is that these instances have better parallel implementations than the compositions of the individual implementations of the corresponding basic components. Received: 27 May 1997 / Revised version: 17 March 1998     

On one class of model optimization problems with a continuum of solutions

We consider critical sets of an H-regular functional. We propose a condition under which the set of all critical points forms a critical set. We discuss several problems that lead to such sets and show a connection with the notion of Morse index. As examples we consider integral functionals for functions defined on a segment.     

On list update with locality of reference

We present a comprehensive study of the list update problem with locality of reference. More specifically, we present a combined theoretical and experimental study in which the theoretically proven and experimentally observed performance guarantees of algorithms match or nearly match. Firstly, we introduce a new model of locality of reference that closely captures the concept of runs. Using this model we develop refined theoretical analyses of popular list update algorithms. Secondly, we present an extensive experimental study in which we have tested the algorithms on traces from benchmark libraries. The theoretical and experimental bounds differ by just a few percent. Our new theoretical bounds are substantially lower than those provided by standard competitive analysis. It shows that the well-known Move-To-Front strategy exhibits the best performance. Its refined competitive ratio tends to 1 as the degree of locality increases. This confirms that Move-To-Front is the method of choice in practice.     

Multilingual news extraction via stopword language model scoring

Web news provides a quick and convenient means to create collections of large documents. The creation of a web news corpus has typically required the construction of a set of HTML parsing rules to identify content text. In general, these parsing rules are written manually and treat different web pages differently. We address this issue and propose a news content recognition algorithm that is language and layout independent. Our method first scans a given HTML document and roughly localizes a set of candidate news areas. Next, we apply a designed scoring function to rank the best content. To validate this approach, we evaluate the systems performance using 1092 items of multilingual web news data covering 17 global regions and 11 distinct languages. We compare these data with nine published content extraction systems using standard settings. The results of this empirical study show that our method outperforms the second-best approach (Boilerpipe) by 6.04 and 10.79 % with regard to the relative micro and macro F-measures, respectively. We also apply our system to monitor online RSS news distribution. It collected 0.4 million news articles from 200 RSS channels in 20 days. This sample quality test shows that our method achieved 93 % extraction accuracy for large news streams.     

基于特征聚合的假新闻内容检测模型

针对假新闻内容检测中分类算法模型的检测性能与泛化性能无法兼顾的问题,提出了一种基于特征聚合的假新闻检测模型CCNN。首先,通过双向长短时循环神经网络提取文本的全局时序特征,并采用卷积神经网络（CNN）提取窗口范围内的词语或词组特征;然后,在卷积神经网络池化层之后,采用基于双中心损失训练的特征聚合层;最后,将双向长短时记忆网络（Bi-LSTM）和CNN的特征数据按深度方向拼接成一个向量之后提供给全连接层,采用均匀损失函数uniform-sigmoid训练模型后输出最终的分类结果。实验结果表明,该模型的F1值为80.5%,在训练集和验证集上的差值为1.3个百分点;与传统的支持向量机（SVM）、朴素贝叶斯（NB）和随机森林（RF）模型相比,所提模型的F1值提升了9~14个百分点;与长短时记忆网络（LSTM）、快速文本分类（FastText）等神经网络模型相比,所提模型的泛化性能提升了1.3~2.5个百分点。由此可见,所提模型能够在提高分类性能的同时保证一定的泛化能力,提升整体性能。     

基于线性带权值的广义表的模型检验方法

提出了一个线性带权值的广义表（Linear Weighted Generalized List,LWGL）模型,同时给出了LWGL加法和乘法、析取和合取运算规则,实现了基于LWGL的高层次模型检验方法。实验结果表明LWGL模型对高层次模型检验是有效的。     

面向新闻推荐的用户兴趣模型构建与更新

针对新闻推荐系统中用户兴趣模型构建与用户兴趣漂移问题,提出了一种面向新闻推荐的用户兴趣模型构建与更新方法。首先采用向量空间模型与bisecting K-means聚类算法构建了原始用户兴趣模型;然后以艾宾浩斯遗忘曲线为基础构造了遗忘函数,并以此对用户兴趣模型进行时间加权,从而达到对用户兴趣模型更新的目的。实验以基于用户的协同过滤推荐、基于物品的协同过滤推荐为baseline,实验结果表明所构建的原始用户兴趣模型推荐性能更优,在◢F◣值上提升了4%,更新后的模型与原始模型相比◢F◣值提高了1.3%。     

On one variational Butkovskii problem

One of the unsolved problems of the theory of optimal control, the Butkovskii variational problem which is of importance for applications, was solved.     

On the longest increasing subsequence of a circular list

The longest increasing circular subsequence (LICS) of a list is considered. A Monte Carlo algorithm to compute it is given which has worst case execution time O(n3/2logn) and storage requirement O(n). It is proved that the expected length μ(n) of the LICS satisfies . Numerical experiments with the algorithm suggest that .     

A support vector machine-based model for detecting top management fraud

Detecting fraudulent financial statements (FFS) is critical in order to protect the global financial market. In recent years, FFS have begun to appear and continue to grow rapidly, which has shocked the confidence of investors and threatened the economics of entire countries. While auditors are the last line of defense to detect FFS, many auditors lack the experience and expertise to deal with the related risks. This study introduces a support vector machine-based fraud warning (SVMFW) model to reduce these risks. The model integrates sequential forward selection (SFS), support vector machine (SVM), and a classification and regression tree (CART). SFS is employed to overcome information overload problems, and the SVM technique is then used to assess the likelihood of FFS. To select the parameters of SVM models, particle swarm optimization (PSO) is applied. Finally, CART is employed to enable auditors to increase substantive testing during their audit procedures by adopting reliable, easy-to-grasp decision rules. The experiment results show that the SVMFW model can reduce unnecessary information, satisfactorily detect FFS, and provide directions for properly allocating audit resources in limited audits. The model is a promising alternative for detecting FFS caused by top management, and it can assist in both taxation and the banking system.     

A user model neural network for a personal news service

User modelling has been widely applied to pedantic situations, where we are attempting to infer the user's knowledge. In teaching it is important to know that the user has mastered the elementary concepts before proceeding with the advanced topics. However, the application of user modelling to information retrieval demands a quite different type of user model. Here we construct a user model for browsing, where the user is uncertain of exactly which information he desires. This requires a more inexact and robust user model, that can quickly give guidance to the system. We propose a user model based on neural networks that can be constructed incrementally. Performance of the model shows some promise for this approach. We discuss the advantages and limitations of the approach and its implications for user modelling.     

基于网络新闻评论的四度用户影响力分析模型

由于网络新闻评论的开放性和传播性,经常引发舆论事件,为能正确引导社会舆论,需要重点关注某些具有较高影响力的用户.针对已有方法未能全面考虑表征网络新闻评论用户影响力的因素,提出四度用户影响力分析模型——FDRank(four-degree influence rank),通过综合考虑用户的评论内容、评论情感值、自身质量以...     

基于Q-learning模型的智能化放顶煤控制策略

传统的综放工作面放顶煤控制存在顶煤采出率低、出煤含矸率高等问题,而现有智能决策方法又存在建模困难、学习样本难以获取等障碍。针对上述问题,在液压支架放煤口动作决策过程中引入强化学习思想,提出一种基于Q-learning模型的智能化放顶煤控制策略。以最大化放煤效益为主要目标,结合顶煤放出体实时状态特征及顶煤动态赋存状态,采用基于Q-learning的放顶煤动态决策算法,在线生成多放煤口实时动作策略,优化多放煤口群组协同放煤过程,合理平衡顶煤采出率、出煤含矸率的关系。仿真和对比分析结果表明,该控制策略的顶煤平均采出率为91.24%,比传统"见矸关窗"的放煤方法提高约15.8%;平均全局奖赏值为685,比传统放煤方法提高约11.2%。该控制策略可显著减少混矸、夹矸等现象对放煤过程的影响,提高顶煤放出效益,减少煤炭资源浪费。     

On detection of lines using search techniques and list processing

Two-line detection algorithms using best first and depth first search techniques, and another algorithm for line detection by joining the line segments based on the principle of A?, are proposed. The algorithms are suitable to detect lines from low-resolution low-contrast images. Shortcomings of conventional approaches, including tracking, have been discussed, and the power of proposed techniques over tracking is illustrated. Results of proposed algorithms are compared with one another. The advantage of these AI-based techniques is that lines of different confidence value (strength) and with at least a certain length can be detected, hence the effect of noise is greatly reduced.