面向课程的中文FAQ自动问答系统模型

面向课程的中文FAQ自动问答系统作为网络教学平台中的子系统,可以实现课程自动答疑,具有较强的实用性,该类问答系统的FAQ库大都人工构建,费时费力,针对此种状况,本文提出一种新的问答系统模型,从"百度知道"自动获取不同课程FAQ库的问答对,并根据提出的融合语义词典和句法依存关系的句子相似度算法,实现学生问句与根据问句关键字向量得到的FAQ库检索子集中的问句之间的相似度计算,自动应答,其他情况采用人工解答,并更新FAQ库。实验结果表明,本系统模型,具备根据不同课程自动构建高质量FAQ库的能力,具备更高的应答准确率,具备不断增强答疑效果的能力,具有广阔的应用前景。     

Queueing Theory for Semiconductor Manufacturing Systems: A Survey and Open Problems

This paper surveys applications of queueing theory for semiconductor manufacturing systems (SMSs). Due to sophisticated tool specifications and process flows in semiconductor manufacturing, queueing models can be very complicated. Research efforts have been on the improvement of model assumptions and model input, mainly in the first moment (averages) and the second moment (variations). However, practices show that implementation of classical queueing theory in semiconductor industry has been unsatisfactory. In this paper, open problems on queueing modeling of SMS are discussed. A potential solution is also proposed by relaxing the independent assumptions in the classical queueing theory. Cycle time reduction has constantly been a key focus of semiconductor manufacturing. Compared with simulation, queueing theory-based analytical modeling is much faster in estimating manufacturing system performance and providing more insights for performance improvement. Therefore, queueing modeling attracts generous semiconductor research grants. Unfortunately, existing queueing models focus on simple extensions of the classical queueing theory and fail to question its applicability to the complicated SMS. Hence, related researches have not been employed widely in the semiconductor industry. In this paper, we conduct a survey on the important works and also present some open problems. We also propose a novel solution by relaxing a key assumption in the classical queueing theory. We are currently funded by Intel to explore this potential solution, and we hope it can foster an interesting research field for the years to come.     

答疑系统的双焦点问句分解研究

现实生活中双焦点问句是一种常用的句型,然而自动答疑系统中没有给出相对应的回答方式.针对这种状况,提出了基于双焦点问句的自动答疑系统构架.通过利用六大策略抽取双焦点问句的焦点字符串、采用不同句型产生主次焦点子问句,提高系统对双焦点问句的理解能力.实验结果表明,系统生成的主次焦点子问句的正确率分别为85.28％和87.50％,主焦点中心语的正确率为93.33％.     

The effect of graphic style on data model interpretation

Abstract. Graphic data models are commonly used as a tool for presentation of information structures in the design, implementation, use and maintenance of the databases that support information systems. The methods proposed for database design assume that the use of graphic data models will enhance understanding of system specifications by both the end-users and the implementers of the system. For this assumption to hold, the information presented in the graphic data model must be readily comprehensible so that the design, represented by the model, can be confirmed and implemented correctly. The lack of standard representations for graphic models has led to a variety of graphic styles. To date, there has been little focus on studying the effect graphic style has on model comprehension. We have studied the effect of three graphic styles proposed for data models on model legibility and interpretation. Our study shows a significant variation in model interpretation that can be attributed to the graphic syntax used. Graphic style appears to influence which model elements are included in the interpretation, as well as the way data models are read.     

中文问答系统中基于主题和焦点的问题理解

问答系统应该能够用准确、简洁的语言回答用户提出的问题.问题理解是问答系统的首要的分析工作,分析的结果的正确率直接影响后续处理.提出了一种基于用户问题的主题和焦点的问题理解的方法,该方法以问题的疑问词、句法分析、问题焦点、问题主题在知网中的首义原作为问题分类特征.实验结果证明,该方法对提高分类的准确率有较好的效果.     

Model management system for IRT-based test construction decision support system

Decision support system (DSS) has become widespread for some specific domains in recent years. However, DSS for IRT-based (item response theory) test construction has not yet been developed. This domain basically imposes a semi-structured or unstructured decision and, therefore, involves a very complex modeling process. This study develops a model management system (MMS) architecture to assist a non-expert user in manipulating test construction process efficiently and effectively. This architecture consists of four components: problem analysis, model type selection, model formulation and solver. The model type selection subsystem is further organized into three levels of hierarchy, i.e., environment, structure and parameter. A prototype is presented to demonstrate the feasibility of this architecture. The results indicate that this approach can be applied for providing an integrated, flexible and user-friendly DSS environment for producing better quality of results in less solution time.     

Discovering genres of online discussion threads via text mining

As course management systems (CMS) gain popularity in facilitating teaching. A forum is a key component to facilitate the interactions among students and teachers. Content analysis is the most popular way to study a discussion forum. But content analysis is a human labor intensity process; for example, the coding process relies heavily on manual interpretation; and it is time and energy consuming. In an asynchronous virtual learning environment, an instructor needs to keep monitoring the discussion forum from time to time in order to maintain the quality of a discussion forum. However, it is time consuming and difficult for instructors to fulfill this need especially for K12 teachers. This research proposes a genre classification system, called GCS, to facilitate the automatic coding process. We treat the coding process as a document classification task via modern data mining techniques. The genre of a posting can be perceived as an announcement, a question, clarification, interpretation, conflict, assertion, etc. This research examines the coding coherence between GCS and experts’ judgment in terms of recall and precision, and discusses how we adjust the parameters of the GCS to improve the coherence. Based on the empirical results, GCS adopts the cascade classification model to achieve the automatic coding process. The empirical evaluation of the classified genres from a repository of postings in an online course on earth science in a senior high school shows that GCS can effectively facilitate the coding process, and the proposed cascade model can deal with the imbalanced distribution nature of discussion postings. These results imply that GCS based on the cascade model can perform as an automatic posting coding system.     

基于概念层次网络理论的问句分析研究

为了提高问答系统对问句理解的准确率,以概念层次网络理论结合传统计算语言学为思路,提出了适用于限定领域中问句分析模型,并根据限定领域的知识特点,设计了新的问句分类方法.在此问句分类方法的基础上,改进了基于多元贝努里模型的贝叶斯分类算法.在以实际教学过程中所收集的真实问句为问题集和训练集的测试中,取得了较好的实践效果.     

基于子问题渐进式推理的3D视觉问答

3D视觉问答可以帮助人们理解空间信息,在幼儿教育等方面具有广阔的应用前景。3D场景信息复杂,现有方法大多直接进行回答,面对复杂问题时容易忽视上下文细节,从而导致性能下降。针对该问题,提出了一种基于子问题渐进式推理的3D视觉问答方法,通过文本分析为复杂的原始问题构建多个简单的子问题。模型在回答子问题的过程中学习上下文信息,帮助理解复杂问题的含义,最终利用积累的联合信息得出原始问题的答案。子问题与原始问题呈现渐近式推理关系,使得模型具有明确的错误解释性和可追溯性。在现有3D数据集ScanQA上进行的实验表明,所提方法在EM@10和CIDEr两个指标上分别达到了51.49%和61.68%,均超过了现有的其他3D视觉问答方法,证实了该方法的有效性。     

A typology of different development and testing options for symbolic regression modelling of measured and calculated datasets

Data-driven modelling is used to develop two alternative types of predictive environmental model: a simulator, a model of a real-world process developed from either a conceptual understanding of physical relations and/or using measured records, and an emulator, an imitator of some other model developed on predicted outputs calculated by that source model. A simple four-way typology called Emulation Simulation Typology (EST) is proposed that distinguishes between (i) model type and (ii) different uses of model development period and model test period datasets. To address the question of to what extent simulator and emulator solutions might be considered interchangeable i.e. provide similar levels of output accuracy when tested on data different from that used in their development, a pair of counterpart pan evaporation models was created using symbolic regression. Each model type delivered similar levels of predictive skill to that other of published solutions. Input–output sensitivity analysis of the two different model types likewise confirmed two very similar underlying response functions. This study demonstrates that the type and quality of data on which a model is tested, has a greater influence on model accuracy assessment, than the type and quality of data on which a model is developed, providing that the development record is sufficiently representative of the conceptual underpinnings of the system being examined. Thus, previously reported substantial disparities occurring in goodness-of-fit statistics for pan evaporation models are most likely explained by the use of either measured or calculated data to test particular models, where lower scores do not necessarily represent major deficiencies in the solution itself.     

基于多特征实体消歧的中文知识图谱问答

问答系统应用于人工智能、自然语言处理和信息检索领域获得了较好的效果,知识图谱问答（KBQA）作为其中的重要组成部分,是一项极具挑战性的自然语言处理任务。然而,目前常见的中文KBQA系统对于实体链接的实体消歧部分并没有给出很好的解决方法。提出一种基于多特征实体消歧的中文KBQA系统,通过结合实体自身的知名度特征、问句与实体关系的语义相似度特征、问句与实体的字符相似度特征和语义相似度特征,构建多特征实体消歧模型,提高实体链接准确率,为系统的问句分类和最优路径选取部分提供更准确的主题实体,从而提升系统性能。实验结果表明,该系统在CCKS2019-CKBQA评测数据的验证集上平均F1值为72.08%,其中采用多特征消歧模型的实体链接准确率达到90.84%,较使用知名度消歧模型和评测大赛第1名分别提升6.35和0.11个百分点。     

双匹配焦点融合的开放域答案选择模型

开放域答案选择模型通过对同一问题的不同候选答案打分,寻找与问题最匹配的答案,是问答（QA）系统的重要组成部分。现有开放域QA系统中的答案选择模型较少关注词级与句子级的融合,导致在匹配关系上缺乏上下文的语义联系,或损失个别单词在语法语义上的细节信息。基于相邻相似原理,提出一种融合双匹配焦点的答案选择模型。根据问答任务多语句关联的特点,设计一种可以将词语的问答承接关系和问答语义关系嵌入进词向量的词嵌入方式,并利用该词向量直接计算词对的余弦相似度,得到词级匹配焦点。通过引入注意力机制的Encoder-Decoder模型提取句子级词对匹配焦点,以问题为基准对齐两个焦点分布矩阵,并使用焦点间的相对距离融合词级与句子级匹配矩阵,获得问题与答案的相关性得分。在Wiki-QA、TREC-QA两个公开问答数据集上的实验结果表明,该模型与多跳注意力模型、层级排序模型相比,平均准确率均值分别提高0.080 1和0.057 1,平均倒数排名分别提高0.017 6和0.006 6。     

Attention-based encoder-decoder model for answer selection in question answering

One of the key challenges for question answering is to bridge the lexical gap between questions and answers because there may not be any matching word between them. Machine translation models have been shown to boost the performance of solving the lexical gap problem between question-answer pairs. In this paper, we introduce an attention-based deep learning model to address the answer selection task for question answering. The proposed model employs a bidirectional long short-term memory (LSTM) encoder-decoder, which has been demonstrated to be effective on machine translation tasks to bridge the lexical gap between questions and answers. Our model also uses a step attention mechanism which allows the question to focus on a certain part of the candidate answer. Finally, we evaluate our model using a benchmark dataset and the results show that our approach outperforms the existing approaches. Integrating our model significantly improves the performance of our question answering system in the TREC 2015 LiveQA task.     

Answering contextual questions based on ontologies and question templates

Contextual question answering (CQA), in which user information needs are satisfied through an interactive question answering (QA) dialog, has recently attracted more research attention. One challenge is to fuse contextual information into the understanding process of relevant questions. In this paper, a discourse structure is proposed to maintain semantic information, and approaches for recognition of relevancy type and fusion of contextual information according to relevancy type are proposed. The system is evaluated on real contextual QA data. The results show that better performance is achieved than a baseline system and almost the same performance as when these contextual phenomena are resolved manually. A detailed evaluation analysis is presented.     

Arc-elasticity and hierarchical exploration of the neighborhood of solutions in mechanical design

In most industrial design processes, the approaches used to obtain a design solution that best fits the specification requirements result in many iterations of the “trial-and-error” type, starting from an initial solution. In this paper, a method is proposed to formalize the decision process in order to automate it, and to provide optimal design solutions. Two types of knowledge are formalized. The first expresses the satisfaction of design objectives, relating to physical behaviors of candidate design solutions. This formalization uses three models, an observation one, an interpretation one and an aggregation one; every design solution is qualified through a single performance variable (a single objective function). The second model is related to modifications that may or may not be applicable to the pre-existing solution. The Designer is often able to define preferences concerning design variables. Some modifications related to this pre-existing solution, can be preferred to other ones. A hierarchy of design variables is proposed to formalize these preferences. The concept of arc-elasticity is introduced as a post-processing indicator to qualify candidate solutions through a trade-off between the performance improvement and their relative distances to the initial solution. The proposed method is used and applied to a riveted assembly, and a genetic algorithm is used to identify optimal solutions.     

移动复制数据库系统冲突检测及消解策略

复制技术是提高移动库系统性能的一项关键技术，该文提出了一种新的移动复制数据库系统模型－事务级吉果集传递（TLRSP）移动复制模型，重点分析了该模型中的冲突检测及消解策略，并给出具体的实现算法，TLRSP移动复制模型允许移动用户在系统断连时存取数据库的本地副本并提交事务，重新连接时进行冲突的检测及消解，同时进行事务结果集的合并，最后通过增量刷新的方式进行同步处理，使得系统最终收敛于一致性的状态。此外，通过引入简化的事务日志，数据牌本号以及权限控制等概念，TLRSP模型有效地降低了移动数据库系统的资源消耗，保证了数据库的一致性，从而为移动数据库系统复制提供了一个可行的解决方案。     

A model-driven approach to develop high performance web applications

The evolution of web technologies in the last few years has contributed to the improvement of web applications, and with the appearance of AJAX and Web 2.0 technology, a new breed of applications for the Internet has emerged: widgets, gadgets and mashups are some examples of this trend. However, as web applications become more and more complex, development costs are increasing in an exponential rate, and we believe that considering a software engineering methodology in the development process of this type of applications, contributes to the solution of this problem. In order to solve this question, this paper proposes a model-driven architecture to support web application development from the design to the implementation model. With this aim, the following tasks have been performed: first a new profile extends UML with new concepts extracted from the web domain, then a new framework supports web application development by composing heterogeneous web elements, and finally, a transformation model generates web applications from the UML extension proposed. The main contribution of this work is a cost and complexity reduction due to the incorporation of a model-driven architecture into the web application development process, but other advantages that can be mentioned are a high performance degree achieved by a prefetching cache mechanism, and a high reusability, since web elements can be reused in different web applications.     

Numerical analysis and testing of a stable and convergent finite element scheme for approximate deconvolution turbulence models

This report presents a stable and convergent finite element scheme for the approximate deconvolution turbulence models (ADM). The ADM is a popular turbulence model intensely studied lately but the computation of its numerical solution raises issues in terms of efficiency and accuracy. This report addresses this question. The proposed scheme presented herein is based on a new interpretation of the ADM model recently introduced by the author. Following this interpretation, the solution of the ADM is viewed as the average of a perturbed Navier–Stokes system. The scheme uses the Crank–Nicolson time discretization and the finite element spatial discretization and is proved to be stable and convergent provided a moderate choice of the time step is made. Numerical tests to verify the convergence rates and performance on a benchmark problem are also provided and they prove the correctness of this approach to numerically solve the ADM.     

A model-theoretic interpretation of environment-induced superselection

The question of what constitutes a ‘system’ is foundational to quantum measurement theory. Environment-induced superselection or ‘einselection’ has been proposed as an observer-independent mechanism by which apparently classical systems ‘emerge’ from physical interactions between degrees of freedom described completely quantum mechanically. It is shown here that einselection can only generate classical systems if the ‘environment’ is assumed a priori to be classical; einselection therefore does not provide an observer-independent mechanism by which classicality can emerge from quantum dynamics. Einselection is then reformulated in terms of positive operator-valued measures acting on a global quantum state. It is shown that this reformulation enables a natural interpretation of apparently classical systems as virtual machines that requires no assumptions beyond those of classical computer science.