多目标情感分类中文数据集构建及分析研究

目标级情感分类任务是为了得到句子中特定评价目标的情感倾向。一个句子中往往存在多个目标,多个目标的情感可能一致,也可能不一致。但在已有针对目标级情感分类的评测数据集中: ①大多数是一个句子一个目标; ②在少数有多个目标的句子中,多个目标情感倾向分布并不均衡,多个目标情感一致的句子占较大比例。数据集本身的缺陷限制了模型针对多个目标进行情感分类的提升空间。针对以上问题,该文构建了一个针对多目标情感分类的中文数据集,人工标注了6 339个评价目标,共2 071条数据。该数据集具备以下特点: ①评价目标个数分布平衡; ②情感正负极性分布平衡; ③多目标情感倾向分布平衡。随后,该文利用多个目标情感分类的主流模型在该数据集上进行了实验与比较分析。结果表明,现有主流模型尚不能对存在多个目标且目标情感倾向性不一致实例中的目标进行很好的分类,尤其是目标的情感倾向为中性时。因此多目标情感分类任务具有一定的难度与挑战性。     

Improving sentiment analysis via sentence type classification using BiLSTM-CRF and CNN

Different types of sentences express sentiment in very different ways. Traditional sentence-level sentiment classification research focuses on one-technique-fits-all solution or only centers on one special type of sentences. In this paper, we propose a divide-and-conquer approach which first classifies sentences into different types, then performs sentiment analysis separately on sentences from each type. Specifically, we find that sentences tend to be more complex if they contain more sentiment targets. Thus, we propose to first apply a neural network based sequence model to classify opinionated sentences into three types according to the number of targets appeared in a sentence. Each group of sentences is then fed into a one-dimensional convolutional neural network separately for sentiment classification. Our approach has been evaluated on four sentiment classification datasets and compared with a wide range of baselines. Experimental results show that: (1) sentence type classification can improve the performance of sentence-level sentiment analysis; (2) the proposed approach achieves state-of-the-art results on several benchmarking datasets.     

基于CRT机制混合神经网络的特定目标情感分析

特定目标情感分析的目的是从不同目标词语的角度来预测文本的情感,关键是为给定的目标分配适当的情感词。当句子中出现多个情感词描述多个目标情感的情况时,可能会导致情感词和目标之间的不匹配。由此提出了一个CRT机制混合神经网络用于特定目标情感分析,模型使用CNN层从经过BiLSTM变换后的单词表示中提取特征,通过CRT组件生成单词的特定目标表示并保存来自BiLSTM层的原始上下文信息。在三种公开数据集上进行了实验,结果表明,该模型在特定目标情感分析任务中较之前的情感分析模型在准确率和稳定性上有着明显的提升,证明CRT机制能很好地整合CNN和LSTM的优势,这对于特定目标情感分析任务具有重要的意义。     

基于浅层句法特征的评价对象抽取研究

随着网络评论文本数量的快速增长,文本情感分析越来越受到研究者的广泛关注. 句子级文本情感分析就是对主观性文本进行细粒度的挖掘,有重要的研究价值. 评论句中的评价对象抽取是句子级情感分析要研究的关键问题之一. 为了提高评价对象抽取的性能,本文提出在系统模型的训练过程中引入浅层句法信息和启发式位置信息,同时在不增加领域词典的情况下, 有效提高系统的精确率.实验结果表明,将本文提出的特征引入到条件随机域模型和对比模型后,系统的各项指标均有所提高, 并且条件随机域模型的结果优于对比模型.同时,将条件随机域模型的结果与2008年国内中文评测的最大值比较,其F值超过最大值 5%.     

基于CRFs和领域本体的中文微博评价对象抽取研究

微博情感分析是对微博内容进行细粒度的挖掘,有着重要的研究价值。微博评价对象的抽取是微博情感分析研究的关键问题之一。为了提高中文微博评价对象抽取的准确率,该文在中文微博特征分析和微博评论本体构建研究的基础上,尝试从词、词性、情感词以及本体四个方面进行特征选择,采用CRFs模型对评价对象进行抽取。该文将提出的方法运用到COAE2014测评的Task5评价对象抽取任务中,宏平均准确率达到61.20%,在所有测评队伍中居第一。实验结果表明,将本体特征引入到CRFs模型中,能够有效地提高评价对象抽取的准确率。     

Extracting common emotions from blogs based on fine-grained sentiment clustering

Recently, blogs have emerged as the major platform for people to express their feelings and sentiments in the age of Web 2.0. The common emotions, which reflect people’s collective and overall sentiments, are becoming the major concern for governments, business companies and individual users. Different from previous literatures on sentiment classification and summarization, the major issue of common emotion extraction is to find out people’s collective sentiments and their corresponding distributions on the Web. Most existing blog clustering methods take into account keywords, stories or timelines but neglect the embedded sentiments, which are considered very important features of blogs. In this paper, a novel method based on Probabilistic Latent Semantic Analysis (PLSA) is presented to model the hidden sentiment factors and an emotion-oriented clustering approach is proposed to find common emotions according to the fine-grained sentiment similarity between blogs. Extensive experiments are conducted on real-world datasets consisting of different topics. The results show that our approach can partition blogs into sentiment coherent clusters and the extracted common emotion words afford good navigation guidelines for embedded sentiments in each cluster.     

多注意力层次神经网络文本情感分析

在文本情感分析研究中,一条评论分别包含了篇章级、句子级和词语级等不同粒度的语义信息,而不同的词和句子在情感分类中所起的作用也是不同的,直接使用整条评论进行建模的情感分析方法则过于粗糙,同时也忽略了表达情感的用户信息和被评价的产品信息。针对该问题,提出一种基于多注意力机制的层次神经网络模型。该模型分别从词语级别、句子级别和篇章级别获取语义信息,并分别在句子级和篇章级引入基于用户和商品的注意力机制来计算不同句子和词的重要性。最后通过三个公开数据集进行测试验证,实验结果表明,基于多注意力层次神经网络的文本情感分析方法较其他模型性能有显著的提升。     

Associating targets with SentiUnits: a step forward in sentiment analysis of Urdu text

This paper presents, a grammatically motivated, sentiment classification model, applied on a morphologically rich language: Urdu. The morphological complexity and flexibility in grammatical rules of this language require an improved or altogether different approach. We emphasize on the identification of the SentiUnits, rather than, the subjective words in the given text. SentiUnits are the sentiment carrier expressions, which reveal the inherent sentiments of the sentence for a specific target. The targets are the noun phrases for which an opinion is made. The system extracts SentiUnits and the target expressions through the shallow parsing based chunking. The dependency parsing algorithm creates associations between these extracted expressions. For our system, we develop sentiment-annotated lexicon of Urdu words. Each entry of the lexicon is marked with its orientation (positive or negative) and the intensity (force of orientation) score. For the evaluation of the system, two corpora of reviews, from the domains of movies and electronic appliances are collected. The results of the experimentation show that, we achieve the state of the art performance in the sentiment analysis of the Urdu text.     

基于多通道双向长短期记忆网络的情感分析

当前存在着大量的语言知识和情感资源,但在基于深度学习的情感分析研究中,这些特有的情感信息,没有在情感分析任务中得到充分利用。针对以上问题,该文提出了一种基于多通道双向长短期记忆网络的情感分析模型(multi-channels bidirectional long short term memory network,Multi-Bi-LSTM),该模型对情感分析任务中现有的语言知识和情感资源进行建模,生成不同的特征通道,让模型充分学习句子中的情感信息。与CNN相比,该模型使用的Bi-LSTM考虑了词序列之间依赖关系,能够捕捉句子的上下文语义信息,使模型获得更多的情感信息。最后在中文COAE2014数据集、英文MR数据集和SST数据集进行实验,取得了比普通Bi-LSTM、结合情感序列特征的卷积神经网络以及传统分类器更好的性能。     

一种基于情感句模的文本情感分类方法

考虑到同类型的情感句往往具有相同或者相似的句法和语义表达模式,该文提出了一种基于情感句模的文本情感自动分类方法。首先,将情感表达相关句模人工分为3大类105个二级分类;然后,设计了一种利用依存特征、句法特征和同义词特征的句模获取方法,从标注情感句中半自动地获取情感句模。最后,通过对输入句进行情感句模分类实现文本情感分类。在NLP&CC2013中文微博情绪分类评测语料及RenCECps博客语料的实验结果显示,该文提出的分类方法准确率显著高于基于词特征支持向量机分类器。     

A New Method for Sentiment Analysis Using Contextual Auto-Encoders

Sentiment analysis, a hot research topic, presents new challenges for understanding users’ opinions and judgments expressed online. They aim to classify the subjective texts by assigning them a polarity label. In this paper, we introduce a novel machine learning framework using auto-encoders network to predict the sentiment polarity label at the word level and the sentence level. Inspired by the dimensionality reduction and the feature extraction capabilities of the auto-encoders, we propose a new model for distributed word vector representation “PMI-SA” using as input pointwise-mutual-information “PMI” word vectors. The resulted continuous word vectors are combined to represent a sentence. An unsupervised sentence embedding method, called Contextual Recursive Auto-Encoders “CoRAE”, is also developed for learning sentence representation. Indeed, CoRAE follows the basic idea of the recursive auto-encoders to deeply compose the vectors of words constituting the sentence, but without relying on any syntactic parse tree. The CoRAE model consists in combining recursively each word with its context words (neighbors’ words: previous and next) by considering the word order. A support vector machine classifier with fine-tuning technique is also used to show that our deep compositional representation model CoRAE improves significantly the accuracy of sentiment analysis task. Experimental results demonstrate that CoRAE remarkably outperforms several competitive baseline methods on two databases, namely, Sanders twitter corpus and Facebook comments corpus. The CoRAE model achieves an efficiency of 83.28% with the Facebook dataset and 97.57% with the Sanders dataset.     

基于注意力的多层次混合融合的多任务多模态情感分析

针对多模态情感分析中的模态内部特征表示和模态间的特征融合问题,结合注意力机制和多任务学习,提出了一种基于注意力的多层次混合融合的多任务多模态情感分析模型MAM(multi-level attention and multi-task)。首先,利用卷积神经网络和双向门控循环单元来实现单模态内部特征的提取;其次,利用跨模态注意力机制实现模态间的两两特征融合;再次,在不同层次使用自注意力机制实现模态贡献度选择;最后,结合多任务学习获得情感和情绪的分类结果。在公开的CMU-MOSEI数据集上的实验结果表明,情感和情绪分类的准确率和F;值均有所提升。     

用户评论方面级情感分析研究

方面级情感分析是自然语言处理的热门研究方向之一,相比于传统的情感分析技术,基于方面的情感分析是细粒度的,能够判断句子中多个目标的情感倾向,能更加准确地挖掘用户对目标的情感极性。针对以往研究忽略目标单独建模的问题,提出了一种基于双向长短期记忆神经网络(BiLSTM)的交互注意力神经网络模型(Bi-IAN)。该模型通过BiLSTM对目标和上下文分别进行建模,获得目标和上下文的隐藏表示,提取其中的语义信息。接下来利用交互注意模块学习上下文和目标之间的注意力,分别生成目标和上下文的表示,捕捉目标和上下文之内和之间的相关性,并重构评价对象和上下文的表示,最终通过非线性层得到分类结果。在数据集SemEval 2014任务4和Chinese review datasets上的实验训练显示,在正确率和F1-score上,比现有的基准情感分析模型有更好的效果。     

Sen-BiGAT-Inter:情绪原因对抽取方法

情绪原因对抽取任务是将情绪子句与原因子句同时抽取。针对该任务,现有模型的编码层未考虑强化情感词语义表示,且仅使用单一图注意力网络,因此,该文提出了一个使用情感词典、图网络和多头注意力的情绪原因对抽取方法(Sen-BiGAT-Inter)。该方法首先利用情感词典与子句中的情感词汇匹配,并将匹配的情感词汇与该子句进行合并,再使用预训练模型BERT(Bidirectional Encoder Representation from Transformers)对句子进行表示。其次,建立两个图注意力网络,分别学习情绪子句和原因子句表示,进而获取候选情绪原因对的表示。在此基础上,应用多头注意力交互机制学习候选情绪原因对的全局信息,同时结合相对位置信息得到候选情绪原因对的表示,用于实现情绪原因对的抽取。在中文情绪原因对抽取数据集上的实验结果显示,相比目前最优的结果,该文所提出的模型在 F₁ 值上提升约1.95。     

Exploring sentiment parsing of microblogging texts for opinion polling on chinese public figures

Microblogging websites such as twitter and Sina Weibo have attracted many users to share their experiences and express their opinions on a variety of topics, making them ideal platforms on which to conduct electronic opinion polls on products, services and public figures. However, conventional sentiment analysis methods for microblogging messages may not meet the demands of opinion polls for public figures. Therefore, in this study, we focus mainly on the problem of sentiment analysis for opinion polling on Chinese public figures. We propose a sentiment parsing-based architecture, which represents and labels opinion targets and their corresponding sentiments jointly to avoid the mismatching of them, for opinion poll of public figures using microblogs. Furthermore, we formulate sentiment parsing of microblogging sentences as a sequence labeling problem and adapt different Recurrent Neural Network (RNN) models to train and infer the model. Our experimental results demonstrate that the proposed sentiment parsing-based methods achieve better performance than conventional sentiment score-based methods in opinion polling on public figures using microblogs.     

SENTIMENT ASSESSMENT OF TEXT BY ANALYZING LINGUISTIC FEATURES AND CONTEXTUAL VALENCE ASSIGNMENT

Text is not only an important medium to describe facts and events, but also to effectively communicate information about the writer's positive or negative sentiment underlying an opinion, or to express an affective or emotional state, such as happiness, fearfulness, surpriseness, and so on. We consider sentiment assessment and emotion sensing from text as two different problems, whereby sentiment assessment is the task that we want to solve first. Thus, this article presents an approach to sentiment assessment, i.e., the recognition of negative or positive valence of a sentence. For the purpose of sentiment recognition from text, we perform semantic dependency analysis on the semantic verb frames of each sentence, and then apply a set of rules to each dependency relation to calculate the contextual valence of the whole sentence. By employing a domain-independent, rule-based approach our system is able to automatically identify sentence-level sentiment. A linguistic tool called “SenseNet” has been developed to recognize sentiments in text, and to visualize the detected sentiments. We conducted several experiments with a variety of datasets containing data from different domains. The obtained results indicate significant performance gains over existing state-of-the-art approaches.     

基于神经主题模型的对话情感分析

对话情感分析旨在识别出一段对话中每个句子的情感倾向,其在电商客服数据分析中发挥着关键作用。不同于对单个句子的情感分析,对话中句子的情感倾向依赖于其在对话中的上下文。目前已有的方法主要采用循环神经网络和注意力机制建模句子之间的关系,但是忽略了对话作为一个整体所呈现的特点。建立在多任务学习的框架下,该文提出了一个新颖的方法,同时推测一段对话的主题分布和每个句子的情感倾向。对话的主题分布,作为一种全局信息,被嵌入到每个词以及句子的表示中。通过这种方法,每个词和句子被赋予了在特定对话主题下的含义。在电商客服对话数据上的实验结果表明,该文提出的模型能充分利用对话主题信息,与不考虑主题信息的基线模型相比,Macro-F1值均有明显提升。     

基于加权融合字词向量的中文在线评论情感分析

随着社交网络平台的广泛使用,涌现出大量蕴涵丰富情感信息的在线评论文本,分析评论中表达的情感对企业、平台等具有重要意义。为了解决目前针对在线评论短文本情感分析中存在特征提取能力弱以及忽略短文本本身情感信息的问题,提出一种基于文本情感值加权融合字词向量表示的模型——SVW-BERT模型。首先,基于字、词级别向量融合表示文本向量,最大程度获取语义表征,同时考虑副词、否定词、感叹句及疑问句对文本情感的影响,通过权值计算得到文本的情感值,构建情感值加权融合字词向量的中文短文本情感分析模型。通过网络平台在线评论数据集对模型的可行性和优越性进行验证。实验结果表明,字词向量融合特征提取语义的能力更强,同时情感值加权句向量考虑了文本本身蕴涵的情感信息,达到了提升情感分类能力的效果。     

层次化双注意力神经网络模型的情感分析研究

在篇章级的情感分类中由于篇章级文本较长,特征提取较普通句子级分析相对较难,大多方法使用层次化的模型进行篇章文本的情感分析,但目前的层次化模型多以循环神经网络和注意力机制为主,单一的循环神经网络结构提取的特征不够明显。本文针对篇章级的情感分类任务,提出一种层次化双注意力神经网络模型。首先对卷积神经网络进行改进,构建词注意力卷积神经网络。然后模型从两个层次依次提取篇章特征,第一层次使注意力卷积神经网络发现每个句子中的重要词汇,提取句子的词特征,构建句子特征向量;第二层次以循环神经网络获取整个篇章的语义表示,全局注意力机制发现篇章中每个句子的重要性,分配以不同的权重,最后构建篇章的整体语义表示。在IMDB、YELP 2013、YELP 2014数据集上的实验表明,模型较当前最好的模型更具优越性。     

注意力增强的双向LSTM情感分析

在基于深度学习的情感分析工作中,传统的注意力机制主要以串行的方式作为其他模型的下一层,用于学习其他神经网络模型输出的权重分布。该文在探究使用深度学习进行句子级情感分析任务的基础上,提出一种注意力增强的双向LSTM模型。模型使用注意力机制直接从词向量的基础上学习每个词对句子情感倾向的权重分布,从而学习到能增强分类效果的词语,使用双向LSTM学习文本的语义信息。最终,通过并行融合的方式提升分类效果。通过在NLPCC 2014情感分析语料上进行测试,该模型的结果优于其他句子级情感分类模型。