基于向量投影的KNN文本分类算法

针对KNN算法分类时间过长的缺点,分析了提高分类效率的方法.在KNN算法基础上,结合向量投影理论以及iDistance索引结构,提出了一种改进的KNN算法--PKNN.该算法通过比较待分类样本和训练样本的一维投影距离,获得最有可能的临近样本点,减小了参与计算的训练样本数,因此可以减少每次分类的计算量.实验结果表明,PKNN算法可以明显提高KNN算法的效率,PKNN算法的原理决定其适合大容量高维文本分类.     

文本分类技术研究进展

介绍文本分类的研究背景、关键技术;总结经典文本分类方法;讨论目前新涌现的分本分类模型及面临的问题,并对文本分类趋势进行展望。     

面向短文本的命名实体识别

针对短文本命名实体识别这项紧缺任务,提出了一种面向短文本的快速有效的命名实体识别方法。该方法主要分成三步：第一步,针对短文本表达不规范特性对命名实体识别的干扰,采取去干扰字符,化繁为简等规范化操作。第二步,针对短文本语意不完整特性,提出用HMM（隐马尔可夫模型）以词性做观察值进行初步命名实体识别。第三步,据初步识别结果,构建拼音同指关系库来识别潜在实体。在由8464篇短文本构成的测试集上运行的实验表明,该方法能较好地进行短文本命名实体识别。     

Using visual and text features for direct marketing on multimedia messaging services domain

Traditionally, direct marketing companies have relied on pre-testing to select the best offers to send to their audience. Companies systematically dispatch the offers under consideration to a limited sample of potential buyers, rank them with respect to their performance and, based on this ranking, decide which offers to send to the wider population. Though this pre-testing process is simple and widely used, recently the industry has been under increased pressure to further optimize learning, in particular when facing severe time and learning space constraints. The main contribution of the present work is to demonstrate that direct marketing firms can exploit the information on visual content to optimize the learning phase. This paper proposes a two-phase learning strategy based on a cascade of regression methods that takes advantage of the visual and text features to improve and accelerate the learning process. Experiments in the domain of a commercial Multimedia Messaging Service (MMS) show the effectiveness of the proposed methods and a significant improvement over traditional learning techniques. The proposed approach can be used in any multimedia direct marketing domain in which offers comprise both a visual and text component.

基于BP网络的中文文本分类技术

本文在对文本分类的问题,关键技术及系统结构进行介绍的基础上,详细阐述了一种利用带动力项的BP神经网络作为分类器的中文文本自动分类方法.该法采用归一化TFIDF算法对特征向量进行权值计算,并使用期望交叉熵统计方法对特征向量集进行精简.此外,我们在TanCorp12数据集上测试了特征项数目和训练次数对于分类器的宏平均和微平均性能的影响.     

The design of the TEI encoding scheme

This paper discusses the basic design of the encoding scheme described by the Text Encoding Initiative'sGuidelines for Electronic Text Encoding and Interchange (TEI document number TEI P3, hereafter simplyP3 orthe Guidelines). It first reviews the basic design goals of the TEI project and their development during the course of the project. Next, it outlines some basic notions relevant for the design of any markup language and uses those notions to describe the basic structure of the TEI encoding scheme. It also describes briefly the core tag set defined in chapter 6 of P3, and the default text structure defined in chapter 7 of that work. The final section of the paper attempts an evaluation of P3 in the light of its original design goals, and outlines areas in which further work is still needed.C. M. Sperberg-McQueen is a Senior Research Programmer at the academic computer center of the University of Illinois at Chicago; his interests include medieval Germanic languages and literatures and the theory of electronic text markup. Since 1988 he has been editor in chief of the ACH/ACL/ALLC Text Encoding Initiative. Lou Burnard is Director of the Oxford Text Archive at Oxford University Computing Services, with interests in electronic text and database technology. He is European Editor of the Text Encoding Initiative's Guidelines.     

A rationale for the TEI recommendations for feature-structure markup

In this paper, we concentrate on justifying the decisions we made in developing the TEI recommendations for feature structure markup. The first four sections of this paper present the justification for the recommended treatment of feature structures, of features and their values, and of combinations of features or values and of alternations and negations of features and their values. Section 5 departs briefly from the linguistic focus to argue that the markup scheme developed for feature structures is in fact a general-purpose mechanism that can be used for a wide range of applications. Section 6 describes an auxiliary document called a feature system declaration that is used to document and validate a system of feature-structure markup. The seventh and final section illustrates the use of the recommended markup scheme with two examples, lexical tagging and interlinear text analysis.Terry Langendoen is Professor and Head of the Department of Linguistics at The University of Arizona. He was Chair of the TEI Committee on Analysis and Interpretation. He received his PhD in Linguistics from the Massachusetts Institute of Technology in 1964, and held teaching positions at The Ohio State University and the City University of New York (Brooklyn College and the Graduate Center) before moving to Arizona in 1988. He is author, co-author, or co-editor of six books in linguistics, and of numerous articles.Gary Simons is Director of the Academic Computing Department of the Summer Institute of Linguistics, Dallas, TX. He served on the TEI Committee on Analysis and Interpretation. He received his PhD in Linguistics (with minor emphasis in Computer Science) from Cornell University in 1979. Before taking up his current position in 1984, he spent five years in the Solomon Islands doing field work with SIL. He is author, co-author, or co-editor of eight books in the fields of linguistics and linguistic computing.The initial feature-structure recommendations were formulated by the Analysis and Interpretation Committee at a meeting in Tucson, Arizona in March 1990, following suggestions by Mitch Marcus and Beatrice Santorini. The authors received valuable help in the further revision and refinement of the recommendations from Steven Zepp.     

不完全匹配的语音和文本语句级对齐

语音文本自动对齐技术广泛应用于语音识别与合成、内容制作等领域,其主要目的是将语音和相应的参考文本在语句、单词、音素等级别的单元进行对齐,并获得语音与参考文本之间的时间对位信息.最新的先进对齐方法大多基于语音识别,一方面,准确率受限于语音识别效果,识别字错误率高时文语对齐精度明显下降,识别字错误率对对齐精度影响较大;另一方面,这种对齐方法不能有效处理不完全匹配的长篇幅语音和文本的对齐.该文提出一种基于锚点和韵律信息的文语对齐方法,通过基于边界锚点加权的片段标注将语料划分为对齐段和未对齐段,针对未对齐段使用双门限端点检测方法提取韵律信息,并检测语句边界,降低了基于语音识别的对齐方法对语音识别效果的依赖程度.实验结果表明,与目前先进的基于语音识别的文语对齐方法比较,即使在识别字错误率为0.52时,该文所提方法的对齐准确率仍能提升45%以上;在音频文本不匹配程度为0.5时,该文所提方法能提高3%.     

基于端到端的多任务商标分卡模型

目前商标分卡处理方法是先进行文本检测再进行区域分类, 最后对不同的区域进行拆分组合形成商标分卡. 这种分步式的处理耗时长, 并且因为误差的叠加会导致最终结果准确率下降. 针对这一问题, 本文提出了多任务的网络模型TextCls, 通过设计多任务学习模型来提升商标分卡的检测和分类模块的推理速度和精确率. 该模型包含一个特征提取网络, 以及文本检测和区域分类两个任务分支. 其中, 文本检测分支采用分割网络学习像素分类图, 然后使用像素聚合获得文本框, 像素分类图主要是学习文本像素和背景像素的信息; 区域分类分支对区域特征细分为中文、英文和图形, 着重学习不同类型区域的特征. 两个分支通过共享特征提取网络, 像素信息和区域特征相互促进学习, 最终两个任务的精确率得以提升. 为了弥补商标图像的文本检测数据集的缺失以及验证TextCls的有效性, 本文还收集并标注了一个由2000张商标图像构成的文本检测数据集trademark_text (https://github.com/kongbailongtian/trademark_text), 结果表明: 与最佳的文本检测算法相比, 本文的文本检测分支将精确率由94.44%提升至95.16%, 调和平均值F1 score达92.12%; 区域分类分支的F1 score也由97.09%提升至98.18%.     

社交媒体文本数据的抑郁症检测研究综述

近年来,机器学习被逐渐运用到基于社交媒体文本数据的抑郁症检测中并凸显重要应用价值。为梳理其应用现状和发展方向,对用于抑郁症检测的社交媒体文本数据集、数据预处理和机器学习方法进行整理分类。在数据特征表示方面,对比分析了基础特征表示、静态词嵌入和语境词嵌入。全面分析了利用不同基础特征和不同算法类型的传统机器学习以及深度学习进行抑郁症检测的性能和特点。总结并建议未来在中文数据集的创建、模型的可解释性、基于隐喻的检测和轻量级预训练模型方面做进一步的探索。