开放域问答技术研究进展

 问答技术是信息检索和自然语言理解中的研究热点.本文对开放域问答技术在近年来的研究进展进行了总结.针对不同类型的问题,比较了问答系统中问题分析、文档和句段检索、答案抽取各个部分不同实现方法的特点,讨论了这些技术中存在的不足.最后,对问答技术未来的研究趋势进行了分析与展望.     

中文问答系统中问题理解的研究

问答系统中包括三个主要的部分:问题理解,信息检索和答案抽取.而问题理解是问答系统的第一部分也是非常关键的一部分.本文应用了基于规则的问题分类方法,并提出了一种基于维基百科关键词扩展和基于同义词的关键词扩展相结合,以及对问题中的关键词进行重要性标识的问题理解方法.该方法先对问题进行分类,然后提取出问题的关键信息.通过实验结果来看:该方法取了了较好的效果,达到了预期的目的.     

面向医疗行业的智能问答系统研究与实现

为了实现医疗信息化中的疾病与诊疗的智能问答系统,本文提出一种基于人机交互的用户自然语言问题理解方法.该方法在关键词库的基础上,采用正向最大匹配算法对用户咨询进行词法分析.进而对医疗领域的问题进行分类,系统根据问题类别进行数据检索;并利用模板匹配方法,完成答案的提取,以句子作为答案的基本单位.最后,在JSP和Java平台上实现了B/S模式的医疗领域智能问答系统,验证了该方法的有效性.     

中文问答系统中问题理解的算法研究

问答系统能够用准确、简洁的答案回答用户提出的问题。问题理解是其首要的分析工作。本文应用了基于规则的问题分类方法,提出了一种基于维基百科关键词扩展和基于同义词的关键词扩展相结合,并对问题中的关键词进行重要性标识的问题理解方法。实验结果证明,该方法对提高问题分类的准确率有较好的效果。     

智能问答的聊天机器人系统的设计与实现

本设计利用众包思想来构建聊天机器人的知识库,并研究了在自动问答系统中基于文档集和基于知识库的答案提取过程,实现了答案的抽取。基于众包的聊天机器人主要包含用户管理、问题匹配、利用众包思想的知识库构建用户交流等功能。用户提出问题后,系统会从对方提问中提取关键词,然后到数据库中检索匹配度高的答案,回答对方,并能够通过反馈,提取对话内容中有效的问答信息增加到自己的数据库中。     

基于生成对抗网络的无监督域适应分类模型

生成适应模型利用生成对抗网络实现模型结构,并在领域适应学习上取得了突破.但其部分网络结构缺少信息交互,且仅使用对抗学习不足以完全减小域间距离,从而使分类精度受到影响.为此,提出一种基于生成对抗网络的无监督域适应分类模型（Unsupervised Domain Adaptation classification model based on GAN,UDAG）.该模型通过联合使用生成对抗网络和多核最大均值差异度量准则优化域间差异,并充分利用无监督对抗训练及监督分类训练之间的信息传递以学习源域分布和目标域分布之间的共享特征.通过在四种域适应情况下的实验结果表明,UDAG模型学习到更优的共享特征嵌入并实现了域适应图像分类,且分类精度有明显提高.     

面向大规模社区问答数据的问题检索方法

随着问答社区网站的兴起,越来越多的用户生成数据积累了起来。这些用户生成数据不仅具有海量的、多样性的等特点,还有着极高的质量和重用价值。为了高效地管理和利用这些数据,近年来研究人员基于这些数据进行了大量的研究和实践,而社区问答中的问题检索就是一个被广泛研究的课题。主要研究了面向大规模社区问答数据的问题检索方法。收集来自Yahoo!Answers等社区网站的超过1．3亿问题和10亿答案的大规模数据,与之前的基于百万量级的数据的问答社区相关研究工作相比有着明显的不同和极高的实用价值。在此数据的基础上,通过查询自动分类方法来提高每次查询效率和效果。在问题检索过程中,提出了应用查询问句和问题的结构信息和语义信息,结合排序学习算法来融合多种不同类别的特征的方法,通过应用训练数据生成排序模型来提高问题检索的相关性和词语不匹配等问题。实验表明,本文应用RankingSVM方法来训练的排序模型在不同数据集上,其准确率等评价指标上都相比以往的方法有着显著的提高。     

基于深度学习的通用目标检测研究综述

目标检测是计算机视觉领域中最基础且最重要的任务之一,是行为识别与人机交互等高层视觉任务的基础.随着深度学习技术的发展,目标检测模型的准确率和效率得到了大幅提升.与传统的目标检测算法相比,深度学习利用强大的分层特征提取和学习能力使得目标检测算法性能取得了突破性进展.与此同时,大规模数据集的出现及显卡计算能力的极大提高也促成了这一领域的蓬勃发展.本文对基于深度学习的目标检测现有研究成果进行了详细综述.首先回顾传统目标检测算法及其存在的问题,其次总结深度学习下区域提案和单阶段基准检测模型.之后从特征图、上下文模型、边框优化、区域提案、类别不平衡处理、训练策略、弱监督学习和无监督学习这八个角度分类总结当前主流的目标检测模型,最后对目标检测算法中待解决的问题和未来研究方向做出展望.     

基于模糊谱聚类的电力客服智能问答系统

为解决电力公司用户服务需求较大与客服数量较少的问题，设计基于模糊谱聚类的电力客服智能问答系统。通过信息辅助模块将问题工单上传至知识库，将完成预处理的问题储存至知识库，通过深度学习分类模型完成问题属性分类后，模糊谱聚类匹配问题答案，计算差异关键词的相似度，查找系列相关问题，通过自助客服模块即时分析后，在系统界面向用户呈现问题答案。实例测试结果表明，该系统具有较低的最坏时间复杂度，可有效加快电力公司的信息传输效率。     

Tr-SLDA:一种面向交叉领域的迁移主题模型

当目标领域缺少足够多的标注数据时,迁移学习利用相关源领域的标注数据,辅助提升目标域的学习性能,但是目标域与源域的数据通常不满足独立同分布,容易导致"负迁移"问题.本文在有监督主题模型(Supervised LDA,SLDA)的基础上,融合迁移学习方法提出一种共享主题知识的迁移主题模型(Transfer SLDA,Tr-SLDA),提出Tr-SLDA-Gibbs主题采样新方法,在类别标签的约束下对不同领域文档中的词采取不同的采样策略,且无需指定主题个数.辅助源域与目标域共享潜在主题空间,Tr-SLDA通过发现潜在共享主题与不同领域类别之间的语义关联从源域迁移知识,可以有效解决"负迁移"问题.基于Tr-SLDA迁移主题模型提出Tr-SLDA-TC(Tr-SLDA Text Categorization)文本分类方法.对比实验表明,该方法可有效利用源域知识来提高目标领域的分类性能.     

基于差异性和重要性的问句特征组合

在问答系统问句分类研究中,对问句特征进行组合有助于构造高效的问句分类器.针对当前问句分类中的特征组合问题,提出一种基于差异性和重要性的特征组合 (Diversity and Importance based Feature Combination,DIFC)方法.通过计算待组合特征与当前特征组合的错分差异度和正分差异度,以及待组合特征本身的重要度,从候选特征集中动态获取优化的特征组合.在哈工大中文问句集上对词袋绑定特征进行组合的实验结果表明,与其他特征组合方法相比,DIFC方法灵活高效,准确率更高.     

Coupled Matrix Factorization for Question Similarity

Community question answering (CQA) has provided an increasingly popular service where users ask and answer questions and access historical question-answer pairs. As a fundamental task in CQA, question similarity measure is to compute the similarity between the queried question and the historical questions which have been solved by other users. We mine and use the most important semantic features as the semantic repre-sentation of questions, and try to incorporate the couplings of semantic features into vector space model. We propose Coupled question similarity (CQS) model, and compute the similarity in matrix factorization framework. Experi-ments conducted on real CQA data sets demonstrate that with the incorporation of such couplings, the performance of sentence similarity is improved compared to a variety of baseline methods significantly.     

Enhancing Performance with a Learnable Strategy for Multiple Question Answering Modules

A question answering (QA) system can be built using multiple QA modules that can individually serve as a QA system in and of themselves. This paper proposes a learnable, strategy‐driven QA model that aims at enhancing both efficiency and effectiveness. A strategy is learned using a learning‐based classification algorithm that determines the sequence of QA modules to be invoked and decides when to stop invoking additional modules. The learned strategy invokes the most suitable QA module for a given question and attempts to verify the answer by consulting other modules until the level of confidence reaches a threshold. In our experiments, our strategy learning approach obtained improvement over a simple routing approach by 10.5% in effectiveness and 27.2% in efficiency.     

基于.NET的答疑系统研究与实现

随着网络技术的发展，网络答疑系统在网络教学中起着十分重要的作用。在分析目前的一些答疑系统的基础上，提出了一个比较全面的答疑系统模型，并对系统进行了功能模块的划分和详细分析。该系统充分结合学科知识库和FAQ库，灵活采用多种形式进行答疑；最后在．NET下使用流行的网络开发技术，实现了系统的主要功能，同时对一些关键技术进行了介绍。     

Visual question answering model based on visual relationship detection

visual question answering (VQA) is a learning task involving two major fields of computer vision and natural language processing. The development of deep learning technology has contributed to the advancement of this research area. Although the research on the question answering model has made great progress, the low accuracy of the VQA model is mainly because the current question answering model structure is relatively simple, the attention mechanism of model is deviated from human attention and lacks a higher level of logical reasoning ability. In response to the above problems, we propose a VQA model based on multi-objective visual relationship detection. Firstly, the appearance feature is used to replace the image features from the original object, and the appearance model is extended by the principle of word vector similarity. The appearance features and relationship predicates are then fed into the word vector space and represented by a fixed length vector. Finally, through the concatenation of elements between the image feature and the question vector are fed into the classifier to generate an output answer. Our method is benchmarked on the DQAUAR data set, and evaluated by the Acc WUPS@0.0 and WUPS@0.9.     

VS3‐NET: Neural variational inference model for machine‐reading comprehension

We propose the VS³‐NET model to solve the task of question answering questions with machine‐reading comprehension that searches for an appropriate answer in a given context. VS³‐NET is a model that trains latent variables for each question using variational inferences based on a model of a simple recurrent unit‐based sentences and self‐matching networks. The types of questions vary, and the answers depend on the type of question. To perform efficient inference and learning, we introduce neural question‐type models to approximate the prior and posterior distributions of the latent variables, and we use these approximated distributions to optimize a reparameterized variational lower bound. The context given in machine‐reading comprehension usually comprises several sentences, leading to performance degradation caused by context length. Therefore, we model a hierarchical structure using sentence encoding, in which as the context becomes longer, the performance degrades. Experimental results show that the proposed VS³‐NET model has an exact‐match score of 76.8% and an F1 score of 84.5% on the SQuAD test set.     

Restricting Answer Candidates Based on Taxonomic Relatedness of Integrated Lexical Knowledge Base in Question Answering

This paper proposes an approach using taxonomic relatedness for answer‐type recognition and type coercion in a question‐answering system. We introduce a question analysis method for a lexical answer type (LAT) and semantic answer type (SAT) and describe the construction of a taxonomy linking them. We also analyze the effectiveness of type coercion based on the taxonomic relatedness of both ATs. Compared with the rule‐based approach of IBM's Watson, our LAT detector, which combines rule‐based and machine‐learning approaches, achieves an 11.04% recall improvement without a sharp decline in precision. Our SAT classifier with a relatedness‐based validation method achieves a precision of 73.55%. For type coercion using the taxonomic relatedness between both ATs and answer candidates, we construct an answer‐type taxonomy that has a semantic relationship between the two ATs. In this paper, we introduce how to link heterogeneous lexical knowledge bases. We propose three strategies for type coercion based on the relatedness between the two ATs and answer candidates in this taxonomy. Finally, we demonstrate that this combination of individual type coercion creates a synergistic effect.     

基于句法分析和答案分类的中文问答系统

本文根据疑问词和谓语的距离信息对问句进行细致的句型分析,然后对答句进行浅层句法分析,在此基础上,抽取出问题特征集、答句特征集和组合特征集作为分类特征,引入最大熵模型和支持向量机训练答案抽取分类器.基于不同特征组合训练得到的分类器在五类事实性问题上进行了测试,其F值分别达到70.87%和85.75%.