基于深度学习的文本分类研究进展

文本分类技术是自然语言处理领域的研究热点,其主要应用于舆情检测、新闻文本分类等领域。近年来,人工神经网络技术在自然语言处理的许多任务中有着很好的表现,将神经网络技术应用于文本分类取得了许多成果。在基于深度学习的文本分类领域,文本分类的数值化表示技术和基于深度学习的文本分类技术是两个重要的研究方向。对目前文本表示的有关词向量的重要技术和应用于文本分类的深度学习方法的实现原理和研究现状进行了系统的分析和总结,并针对当前的技术发展,分析了文本分类方法的不足和发展趋势。     

A <word,part of speech> embedding model for text classification

Existing word embeddings learning algorithms only employ the contexts of words, but different text documents use words and their relevant parts of speech very differently. Based on the preceding assumption, in order to obtain appropriate word embeddings and further improve the effect of text classification, this paper studies in depth a representation of words combined with their parts of speech. First, using the parts of speech and context of words, a more expressive word embeddings can be obtained. Further, to improve the efficiency of look‐up tables, we construct a two‐dimensional table that is in the <word, part of speech> format to represent words in text documents. Finally, the two‐dimensional table and a Bayesian theorem are used for text classification. Experimental results show that our model has achieved more desirable results on standard data sets. And it has more preferable versatility and portability than alternative models.     

CNN-ELM混合短文本分类模型

针对目前自然语言处理研究中,使用卷积神经网络（CNN）进行短文本分类任务时可以结合不同神经网络结构与分类算法以提高分类性能的问题,提出了一种结合卷积神经网络与极速学习机的CNN-ELM混合短文本分类模型。使用词向量训练构成文本矩阵作为输入数据,然后使用卷积神经网络提取特征并使用Highway网络进行特征优化,最后使用误差最小化极速学习机(EM-ELM)作为分类器完成短文本分类任务。与其他模型相比,该混合模型能够提取更具代表性的特征并能快速准确地输出分类结果。在多种英文数据集上的实验结果表明提出的CNN-ELM混合短文本分类模型比传统机器学习模型与深度学习模型更适合完成短文本分类任务。     

Accuracy-based learning classifier systems: models, analysis and applications to classification tasks

Recently, Learning Classifier Systems (LCS) and particularly XCS have arisen as promising methods for classification tasks and data mining. This paper investigates two models of accuracy-based learning classifier systems on different types of classification problems. Departing from XCS, we analyze the evolution of a complete action map as a knowledge representation. We propose an alternative, UCS, which evolves a best action map more efficiently. We also investigate how the fitness pressure guides the search towards accurate classifiers. While XCS bases fitness on a reinforcement learning scheme, UCS defines fitness from a supervised learning scheme. We find significant differences in how the fitness pressure leads towards accuracy, and suggest the use of a supervised approach specially for multi-class problems and problems with unbalanced classes. We also investigate the complexity factors which arise in each type of accuracy-based LCS. We provide a model on the learning complexity of LCS which is based on the representative examples given to the system. The results and observations are also extended to a set of real world classification problems, where accuracy-based LCS are shown to perform competitively with respect to other learning algorithms. The work presents an extended analysis of accuracy-based LCS, gives insight into the understanding of the LCS dynamics, and suggests open issues for further improvement of LCS on classification tasks.     

A thousand words in a scene

This paper presents a novel approach for visual scene modeling and classification, investigating the combined use of text modeling methods and local invariant features. Our work attempts to elucidate (1) whether a textlike bag-of-visterms (BOV) representation (histogram of quantized local visual features) is suitable for scene (rather than object) classification, (2) whether some analogies between discrete scene representations and text documents exist, and 3) whether unsupervised, latent space models can be used both as feature extractors for the classification task and to discover patterns of visual co-occurrence. Using several data sets, we validate our approach, presenting and discussing experiments on each of these issues. We first show, with extensive experiments on binary and multiclass scene classification tasks using a 9,500-image data set, that the BOV representation consistently outperforms classical scene classification approaches. In other data sets, we show that our approach competes with or outperforms other recent more complex methods. We also show that probabilistic latent semantic analysis (PLSA) generates a compact scene representation, is discriminative for accurate classification, and is more robust than the BOV representation when less labeled training data is available. Finally, through aspect-based image ranking experiments, we show the ability of PLSA to automatically extract visually meaningful scene patterns, making such representation useful for browsing image collections.     

Text classification method based on self-training and LDA topic models

Supervised text classification methods are efficient when they can learn with reasonably sized labeled sets. On the other hand, when only a small set of labeled documents is available, semi-supervised methods become more appropriate. These methods are based on comparing distributions between labeled and unlabeled instances, therefore it is important to focus on the representation and its discrimination abilities. In this paper we present the ST LDA method for text classification in a semi-supervised manner with representations based on topic models. The proposed method comprises a semi-supervised text classification algorithm based on self-training and a model, which determines parameter settings for any new document collection. Self-training is used to enlarge the small initial labeled set with the help of information from unlabeled data. We investigate how topic-based representation affects prediction accuracy by performing NBMN and SVM classification algorithms on an enlarged labeled set and then compare the results with the same method on a typical TF-IDF representation. We also compare ST LDA with supervised classification methods and other well-known semi-supervised methods. Experiments were conducted on 11 very small initial labeled sets sampled from six publicly available document collections. The results show that our ST LDA method, when used in combination with NBMN, performed significantly better in terms of classification accuracy than other comparable methods and variations. In this manner, the ST LDA method proved to be a competitive classification method for different text collections when only a small set of labeled instances is available. As such, the proposed ST LDA method may well help to improve text classification tasks, which are essential in many advanced expert and intelligent systems, especially in the case of a scarcity of labeled texts.     

基于语义理解和机器学习的混合的中文文本情感分类算法框架

快速、准确和全面地从大量互联网文本信息中定位情感倾向是当前大数据技术领域面临的一大挑战.文本情感分类方法大致分为基于语义理解和基于有监督的机器学习两类.语义理解处理情感分类的优势在于其对不同领域的文本都可以进行情感分类,但容易受到中文存在的不同句式及搭配的影响,分类精度不高.有监督的机器学习虽然能够达到比较高的情感分类精度,但在一个领域方面得到较高分类能力的分类器不适应新领域的情感分类.在使用信息增益对高维文本做特征降维的基础上,将优化的语义理解和机器学习相结合,设计了一种新的混合语义理解的机器学习中文情感分类算法框架.基于该框架的多组对比实验验证了文本信息在不同领域中高且稳定的分类精度.     

稀疏数据下结合词向量的短文本分类模型研究

针对短文本缺乏足够共现信息所产生的词与词之间弱连接,且难以获取主题词的情况,导致面向短文本分类工作需要人工标注大量的训练样本,以及产生特征稀疏和维度爆炸的问题,提出了一种基于注意力机制和标签图的单词共生短文本分类模型(WGA-BERT)。首先利用预先训练好的BERT模型计算上下文感知的文本表示,并使用WNTM对每个单词的潜在单词组分布进行建模,以获取主题扩展特征向量;其次提出了一种标签图构造方法捕获主题词的结构和相关性;最后,提出了一种注意力机制建立主题词之间,以及主题词和文本之间的联系,解决了数据稀疏性和主题文本异构性的问题。实验结果表明,WGA-BERT模型对于新闻评论类的短文本分类,比传统的机器学习模型在分类精度上平均提高了3%。     

Online multi-label dependency topic models for text classification

Multi-label text classification is an increasingly important field as large amounts of text data are available and extracting relevant information is important in many application contexts. Probabilistic generative models are the basis of a number of popular text mining methods such as Naive Bayes or Latent Dirichlet Allocation. However, Bayesian models for multi-label text classification often are overly complicated to account for label dependencies and skewed label frequencies while at the same time preventing overfitting. To solve this problem we employ the same technique that contributed to the success of deep learning in recent years: greedy layer-wise training. Applying this technique in the supervised setting prevents overfitting and leads to better classification accuracy. The intuition behind this approach is to learn the labels first and subsequently add a more abstract layer to represent dependencies among the labels. This allows using a relatively simple hierarchical topic model which can easily be adapted to the online setting. We show that our method successfully models dependencies online for large-scale multi-label datasets with many labels and improves over the baseline method not modeling dependencies. The same strategy, layer-wise greedy training, also makes the batch variant competitive with existing more complex multi-label topic models.     

基于MAC-LSTM的问题分类研究

针对问句文本通常较短、语义信息与词语共现信息不足等问题,提出一种多层级注意力卷积长短时记忆模型（multi-level attention convolution LSTM neural network,MAC-LSTM）的问题分类方法。相比基于词嵌入的深度学习模型,该方法使用疑问词注意力机制对问句中的疑问词特征重点关注。同时,使用注意力机制结合卷积神经网络与长短时记忆模型各自文本建模的优势,既能够并行方式提取词汇级特征,又能够学习更高级别的长距离依赖特征。实验表明,该方法较传统的机器学习方法和普通的卷积神经网络、长短时记忆模型有明显的效果提升。     

Combining reconstruction and discrimination with class-specific sparse coding

Sparse coding is an important approach for the unsupervised learning of sensory features. In this contribution, we present two new methods that extend the traditional sparse coding approach with supervised components. Our goal is to increase the suitability of the learned features for classification tasks while keeping most of their general representation capability. We analyze the effect of the new methods using visualization on artificial data and discuss the results on two object test sets with regard to the properties of the found feature representation.     

无监督混阶栈式稀疏自编码器的图像分类学习

目前多数图像分类的方法是采用监督学习或者半监督学习对图像进行降维,然而监督学习与半监督学习需要图像携带标签信息。针对无标签图像的降维及分类问题,提出采用混阶栈式稀疏自编码器对图像进行无监督降维来实现图像的分类学习。首先,构建一个具有三个隐藏层的串行栈式自编码器网络,对栈式自编码器的每一个隐藏层单独训练,将前一个隐藏层的输出作为后一个隐藏层的输入,对图像数据进行特征提取并实现对数据的降维。其次,将训练好的栈式自编码器的第一个隐藏层和第二个隐藏层的特征进行拼接融合,形成一个包含混阶特征的矩阵。最后,使用支持向量机对降维后的图像特征进行分类,并进行精度评价。在公开的四个图像数据集上将所提方法与七个对比算法进行对比实验,实验结果表明,所提方法能够对无标签图像进行特征提取,实现图像分类学习,减少分类时间,提高图像的分类精度。     

Rich document representation and classification: An analysis

There are three factors involved in text classification. These are classification model, similarity measure and document representation model. In this paper, we will focus on document representation and demonstrate that the choice of document representation has a profound impact on the quality of the classifier. In our experiments, we have used the centroid-based text classifier, which is a simple and robust text classification scheme. We will compare four different types of document representations: N-grams, Single terms, phrases and RDR which is a logic-based document representation. The N-gram representation is a string-based representation with no linguistic processing. The Single term approach is based on words with minimum linguistic processing. The phrase approach is based on linguistically formed phrases and single words. The RDR is based on linguistic processing and representing documents as a set of logical predicates. We have experimented with many text collections and we have obtained similar results. Here, we base our arguments on experiments conducted on Reuters-21578. We show that RDR, the more complex representation, produces more effective classifier on Reuters-21578, followed by the phrase approach.     

基于TNG特征扩展的MLFM-MN短文本分类算法

在海量短文本中由于特征稀疏、数据维度高这一问题,传统的文本分类方法在分类速度和准确率上达不到理想的效果。针对这一问题提出了一种基于Topic N-Gram(TNG)特征扩展的多级模糊最小-最大神经网络(MLFM-MN)短文本分类算法。首先通过使用改进的TNG模型构建一个特征扩展库并对特征进行扩展,该扩展库不仅可以推断单词分布,还可以推断每个主题文本的短语分布;然后根据短文本中的原始特征,计算这些文本的主题倾向,根据主题倾向,从特征扩展库中选择适当的候选词和短语,并将这些候选词和短语放入原始文本中;最后运用MLFM-MN算法对这些扩展的原始文本对象进行分类,并使用精确率、召回率和F1分数来评估分类效果。实验结果表明,本文提出的新型分类算法能够显著提高文本的分类性能。     

基于DSR和BGRU模型的聊天文本证据分类方法

即时通信等社交软件产生的聊天文本内容证据数据量大且聊天内容含有“黑话”等复杂语义,数字取证时无法快速识别和提取与犯罪事件有关的聊天文本证据。为此,基于DSR(dynamic semantic representation)模型和BGRU(bidirectional gated recurrent unit)模型提出一个聊天文本证据分类模型（DSR-BGRU）。通过预处理手段处理聊天文本数据,使其保存犯罪领域特征。设计并实现了基于DSR模型的聊天文本证据语义特征表示方法,从语义层面对聊天文本进行特征表示,通过聚类算法筛选出语义词,并通过单词属性与语义词的加权组合对非语义词词向量进行特征表示,且将语义词用于对新单词进行稀疏表示。利用Keras框架构建了包含DSR模型输入层、BGRU模型隐藏层和softmax分类层的多层聊天文本特征提取与分类模型,该模型使用DSR模型进行词的向量表示组成的文本矩阵作为输入向量,从语义层面对聊天文本进行特征表示,基于BGRU模型的多层隐藏层对使用这些词向量组成的文本提取上下文特征,从而能够更好地准确理解聊天文本的语义信息,并利用softmax分类层实现聊天文本...     

基于深度置信网络的维吾尔文垃圾短信分类技术研究

针对传统分类算法对维吾尔文文本分类准确率不高的问题,提出了一种基于深度置信网络的维吾尔文短信文本分类模型。深度学习模拟人脑的多层次结构,对数据从低层到高层逐渐地进行特征提取,深层挖掘数据集的分布规律,从而提高分类准确性。通过逐层无监督的方法完成深度置信网络的初始化,并结合softmax回归分类器实现文本的分类。最后在收集的维吾尔文短信数据集上进行实验论证。实验结果表明,相比KNN、SVM和决策树算法,深度置信网络具有更好的分类效果,准确率更高。     

针对低频词进行改进的中文短文本分类方法

中文短文本分类中存在大量低频词,利用好低频词中的信息能有效提高文本分类效果,针对基于词向量的文本分类研究中低频词不能被有效利用的问题,提出一种针对低频词进行数据增强的方法。首先,利用受限文本生成模型产生的数据来微调低频词的词向量,再利用一种词向量的构造算法将高频词的更新信息迁移到低频词中,使低频词获取更准确且符合训练集分布的词向量表示;其次,引入相似词和实体概念等先验知识来补充上下文信息;最后,利用改进的卡方统计去除明显的噪声词,以及设计词注意力层对每个词进行加权,减少无关噪声对分类的影响。在多个基础分类模型上进行实验,结果表明各基础模型经改进后都有明显提升,体现了提出方法的有效性,同时也说明了短文本分类任务中低频词能改善分类的效果。     

结合ALBERT和双向门控循环单元的专利文本分类

随着专利申请数量的快速增长,对专利文本实现自动分类的需求与日俱增。现有的专利文本分类算法大都采用Word2vec和全局词向量（GloVe）等方式获取文本的词向量表示,舍弃了大量词语的位置信息且不能表示出文本的完整语义。针对上述问题,提出了一种结合ALBERT和双向门控循环单元（BiGRU）的多层级专利文本分类模型ALBERT-BiGRU。该模型使用ALBERT预训练的动态词向量代替传统Word2vec等方式训练的静态词向量,提升了词向量的表征能力;并使用BiGRU神经网络模型进行训练,最大限度保留了专利文本中长距离词之间的语义关联。在国家信息中心公布的专利数据集上进行有效性验证,与Word2vec-BiGRU和GloVe-BiGRU相比,ALBERT-BiGRU的准确率在专利文本的部级别分别提高了9.1个百分点和10.9个百分点,在大类级别分别提高了9.5个百分点和11.2个百分点。实验结果表明,ALBERT-BiGRU能有效提升不同层级专利文本的分类效果。