基于对话文本智能挖掘的电力设备供应商评价

目前电网企业的电力设备供应商满意度评价主要依赖于人工统计和指标计算,其准确性受评价人员和评价内容的影响较大.以电力业务平台的对话文本为研究对象,在扩充已有电力本体词典的词条和属性的基础上,建立了基于文本挖掘技术的电力设备供应商评价模型.首先提出了基于Transformer的双向编码器下句预测与余弦相似度加权的单轮对话文...     

基于情感词典与LDA模型的股市文本情感分析

建立了一种基于股票情感词典与LDA分析股票文本情感倾向的模型。针对股票文本情感分析中情感词典不全面与句子分析片面的问题,构建较为全面的股票情感词典,同时以句子的倾向性、程度性与相关性三方面分析股票文本情感。引入针对股票的词语、程度性词语与转折性词语构建较为全面的情感词典;抽取预处理之后的股票文本句子的情感词;利用句子算法计算句子倾向、程度向量,并对句子向量利用支持向量机(SVM)和K均值算法分类;利用LDA(latent dirichlet allocation)对情感词计算文档 主题、文档 词语概率分布,以此概率分布获取句子的相关性;综合句子的倾向性、程度性、相关性计算句子情感;最后,通过句子情感获取股票文本的情感倾向比例。通过对百度新闻经济板块收集的股票文本进行实验并与其他算法比较,该模型对句子与文本分类准确率提高到82.78%与84.14%。     

Senti‐CS: Building a lexical resource for sentiment analysis using subjective feature selection and normalized Chi‐Square‐based feature weight generation

Sentiment analysis involves the detection of sentiment content of text using natural language processing. Natural language processing is a very challenging task due to syntactic ambiguities, named entity recognition, use of slangs, jargons, sarcasm, abbreviations and contextual sensitivity. Sentiment analysis can be performed using supervised as well as unsupervised approaches. As the amount of data grows, unsupervised approaches become vital as they cut down on the learning time and the requirements for availability of a labelled dataset. Sentiment lexicons provide an easy application of unsupervised algorithms for text classification. SentiWordNet is a lexical resource widely employed by many researchers for sentiment analysis and polarity classification. However, the reported performance levels need improvement. The proposed research is focused on raising the performance of SentiWordNet3.0 by using it as a labelled corpus to build another sentiment lexicon, named Senti‐CS. The part of speech information, usage based ranks and sentiment scores are used to calculate Chi‐Square‐based feature weight for each unique subjective term/part‐of‐speech pair extracted from SentiWordNet3.0. This weight is then normalized in a range of ?1 to +1 using min–max normalization. Senti‐CS based sentiment analysis framework is presented and applied on a large dataset of 50000 movie reviews. These results are then compared with baseline SentiWordNet, Mutual Information and Information Gain techniques. State of the art comparison is performed for the Cornell movie review dataset. The analyses of results indicate that the proposed approach outperforms state‐of‐the‐art classifiers.     

英语情态句的情感倾向性分析

该文研究了英语情态句的情感倾向性分析问题。情态句是英语中的常用句型,在用户评论文本中占有很大的比例。由于其独有的语言学特点,情态句中的情感倾向很难被已有的方法有效地分析。在该文中,我们借助词性标签进行了情态句的识别,并提出了一种情态特征用于帮助情态句情感倾向性的分析。为了进一步提高分析效果,我们还给出了通过合并同义情态特征来缓解情态特征稀疏性问题的方法。实验结果表明,在二元及三元情感倾向性分类问题上,该文提出的方法在F值上较经典分类方法分别有4%及7%的提高。     

Modeling and Verification of a Sentiment Analysis System Using Aspect-Oriented Petri Nets

An increasing number of social media and networking platforms have been widely used. People usually post the online comments to share their own opinions on the networking platforms with social media. Business companies are increasingly seeking effective ways to mine what people think and feel regarding their products and services. How to correctly understand the online customers’ reviews becomes an important issue. This study aims to propose a method with the aspect-oriented Petri nets (AOPN) to improve the examination correctness without changing any process and program. We collect those comments from the online reviews with Scrapy tools, perform sentiment analysis using SnowNLP, and examine the analysis results to improve the correctness. In this paper, we apply our method for a case of the online movie comments. The experimental results have shown that AOPN is helpful for the sentiment analysis and verifying its correctness.     

基于ATT-DGRU的文本方面级别情感分析

方面级别情感分类是针对给定文本、分析其在给定方面所表达出的情感极性.现有的主流解决方案中,基于注意力机制的循环神经网络模型忽略了关键词邻近上下文信息的重要性,而结合卷积神经网络(Convolutional Neural Network,CNN)的多层模型不擅长捕捉句子级别的长距离依赖信息.因此,提出了一种基于截断循环神...     

Deep Learning Based Sentiment Analysis of COVID-19 Tweets via Resampling and Label Analysis

Twitter has emerged as a platform that produces new data every day through its users which can be utilized for various purposes. People express their unique ideas and views on multiple topics thus providing vast knowledge. Sentiment analysis is critical from the corporate and political perspectives as it can impact decision-making. Since the proliferation of COVID-19, it has become an important challenge to detect the sentiment of COVID-19-related tweets so that people’s opinions can be tracked. The purpose of this research is to detect the sentiment of people regarding this problem with limited data as it can be challenging considering the various textual characteristics that must be analyzed. Hence, this research presents a deep learning-based model that utilizes the positives of random minority oversampling combined with class label analysis to achieve the best results for sentiment analysis. This research specifically focuses on utilizing class label analysis to deal with the multiclass problem by combining the class labels with a similar overall sentiment. This can be particularly helpful when dealing with smaller datasets. Furthermore, our proposed model integrates various preprocessing steps with random minority oversampling and various deep learning algorithms including standard deep learning and bi-directional deep learning algorithms. This research explores several algorithms and their impact on sentiment analysis tasks and concludes that bidirectional neural networks do not provide any advantage over standard neural networks as standard Neural Networks provide slightly better results than their bidirectional counterparts. The experimental results validate that our model offers excellent results with a validation accuracy of 92.5% and an F1 measure of 0.92.     

Drug Usage Safety from Drug Reviews with Hybrid Machine Learning Approach

With the increasing usage of drugs to remedy different diseases, drug safety has become crucial over the past few years. Often medicine from several companies is offered for a single disease that involves the same/similar substances with slightly different formulae. Such diversification is both helpful and dangerous as such medicine proves to be more effective or shows side effects to different patients. Despite clinical trials, side effects are reported when the medicine is used by the mass public, of which several such experiences are shared on social media platforms. A system capable of analyzing such reviews could be very helpful to assist healthcare professionals and companies for evaluating the safety of drugs after it has been marketed. Sentiment analysis of drug reviews has a large potential for providing valuable insights into these cases. Therefore, this study proposes an approach to perform analysis on the drug safety reviews using lexicon-based and deep learning techniques. A dataset acquired from the ‘Drugs.Com’ containing reviews of drug-related side effects and reactions, is used for experiments. A lexicon-based approach, Textblob is used to extract the positive, negative or neutral sentiment from the review text. Review classification is achieved using a novel hybrid deep learning model of convolutional neural networks and long short-term memory (CNN-LSTM) network. The CNN is used at the first level to extract the appropriate features while LSTM is used at the second level. Several well-known machine learning models including logistic regression, random forest, decision tree, and AdaBoost are evaluated using term frequency-inverse document frequency (TF-IDF), a bag of words (BoW), feature union of (TF-IDF + BoW), and lexicon-based methods. Performance analysis with machine learning models, long short term memory and convolutional neural network models, and state-of-the-art approaches indicate that the proposed CNN-LSTM model shows superior performance with an 0.96 accuracy. We also performed a statistical significance T-test to show the significance of the proposed CNN-LSTM model in comparison with other approaches.     

基于汉语特征的中文对抗样本生成方法

深度神经网络容易受到来自对抗样本的攻击,例如在文本分类任务中修改原始文本中的少量字、词、标点符号即可改变模型分类结果.目前NLP领域对中文对抗样本的研究较少且未充分结合汉语的语言特征.从中文情感分类场景入手,结合了汉语象形、表音等语言特征,提出一种字词级别的高质量的对抗样本生成方法CWordCheater,涵盖字音、字形、标点符号等多个角度.针对形近字的替换方式,引入ConvAE网络完成汉字视觉向量的嵌入,进而生成形近字替换候选池.同时提出一种基于USE编码距离的语义约束方法避免对抗样本的语义偏移问题.构建一套多维度的对抗样本评估方法,从攻击效果和攻击代价两方面评估对抗样本的质量.实验结果表明,CWordAttacker在多个分类模型和多个数据集上能使分类准确率至少下降27.9%,同时拥有更小的基于视觉和语义的扰动代价.     

基于pu-learning的同行评议文本情感分析

最近几年逐渐出现了对同行评议文本情感分析的研究,包括通过同行评议文本预测审稿人的推荐状态的任务。现有模型融入了论文本身或摘要信息,采用神经网络学习论文或摘要的高层表示,结合同行评议文本预测审稿人的推荐状态,这使得模型变得非常复杂的同时结果却没有实质性的提高。为此,提出了OSA机制来提高情感分析模型中对观点句的关注度。具体来说,采用pu-learning从同行评议文本的前N个句子中学习观点句的特征,使每一个句子都得到一个观点句权重,将其应用于情感分析模型的倒数第二层,由此得到最终的预测结果。在ICLR2017—2018数据集上使用不同的情感分析模型对OSA进行了评估,实验结果验证了OSA的高效性,并在两个数据集上取得了优异的性能。