介绍 (Introduction)

The growth of the internet due to social networks such as Facebook, Twitter, Linkedin, Instagram etc. has led to significant users interaction and has empowered users to express their opinions about products, services, events, their preferences among others. It has also provided opportunities to the users to share their wisdom and experiences with each other. The faster development of social networks is causing explosive growth of digital content. It has turned online opinions, blogs, tweets, and posts into a very valuable asset for the corporates to get insights from the data and plan their strategy. Business organizations need to process and study these sentiments to investigate data and to gain business insights(Yadav & Vishwakarma, 2020). Traditional approach to manually extract complex features, identify which feature is relevant, and derive the patterns from this huge information is very time consuming and require significant human efforts. However, Deep Learning can exhibit excellent performance via Natural Language Processing (NLP) techniques to perform sentiment analysis on this massive information. The core idea of Deep Learning techniques is to identify complex features extracted from this vast amount of data without much external intervention using deep neural networks. These algorithms automatically learn new complex features. Both automatic feature extraction and availability of resources are very important when comparing the traditional machine learning approach and deep learning techniques(Araque et al., 2017). Here the goal is to classify the opinions and sentiments expressed by users. In this article, we will discuss about various sentiment analysis techniques and several ensemble models to aggregate the information from multiple features.

由于诸如Facebook，Twitter，Linkedin，Instagram等社交网络的互联网发展，导致了重要的用户互动，并使用户能够表达其对产品，服务，事件，偏好的看法。 它还为用户提供了彼此分享他们的智慧和经验的机会。 社交网络的快速发展正在引起数字内容的爆炸性增长。 它已将在线意见，博客，推文和帖子变成了非常有价值的资产，使企业可以从数据中获得洞察力并计划其战略。 商业组织需要处理和研究这些情绪，以调查数据并获得商业见解(Yadav和Vishwakarma，2020年)。 手动提取复杂特征，识别哪个特征相关并从大量信息中导出模式的传统方法非常耗时，并且需要大量人力。 但是，深度学习可以通过自然语言处理(NLP)技术表现出出色的性能，从而对大量信息进行情感分析。 深度学习技术的核心思想是使用深层神经网络无需大量外部干预即可识别从大量数据中提取的复杂特征。 这些算法会自动学习新的复杂功能。 在比较传统的机器学习方法和深度学习技术时，自动特征提取和资源可用性都非常重要(Araque et al。，2017)。 这里的目标是对用户表达的观点和情感进行分类。 在本文中，我们将讨论各种情感分析技术和几种集成模型，以汇总来自多个功能的信息。

抽象 (Abstract)

Deep Learning uses powerful neural network algorithms to mimics the way human brain process data for translating languages, recognizing speech, detecting objects and making decisions. Deep Learning algorithms are able to identify and learn the patterns from both unstructured and unlabeled data without human intervention. Deep Learning techniques learn through multiple layers of representation and generate state of the art predictive results. In the past years, Deep Learning techniques have been very successful in performing the sentiment analysis. It provides automatic feature extraction, rich representation capabilities and better performance than traditional feature based techniques. These long-established approaches can yield strong baselines, and their predictive capabilities can be used in conjunction with the arising deep learning methods(Preethi et al., 2017). Two techniques of neural networks are very common — Convolutional Neural Networks(CNN) for image processing and Recurrent Neural Networks (RNN) — for natural language processing (NLP) tasks(Goularas & Kamis, 2019). Deep Learning is used to optimize the recommendations depending on the sentiment analysis performed on the different reviews, which are taken from different social networking sites. The Experiments performed indicate that the RNN based Deep-learning Sentiment Analysis (RDSA) improvises the behavior by increasing the accuracy of the sentiment analysis, which in turn yields better recommendations to the user and thus helps to identify a particular position as per the requirement of the user need(Preethi et al., 2017). In this article, we will discuss popular deep learning models which are increasingly applied in the sentiment analysis including CNN, RNN, various ensemble techniques. This article provides insights on various techniques for sentiment analysis.

深度学习使用强大的神经网络算法来模仿人脑处理数据以翻译语言，识别语音，检测物体并做出决策的方式。 深度学习算法无需人工干预即可从非结构化和未标记的数据中识别和学习模式。 深度学习技术通过多层表示进行学习，并生成最新的预测结果。 在过去的几年中，深度学习技术在执行情感分析方面非常成功。 与传统的基于特征的技术相比，它提供了自动特征提取，丰富的表示功能以及更好的性能。 这些历史悠久的方法可以产生强大的基线，其预测能力可以与新兴的深度学习方法结合使用(Preethi等人，2017)。 神经网络的两种技术非常普遍-用于图像处理的卷积神经网络(CNN)和用于自然语言处理(NLP)任务的递归神经网络(RNN)(Goularas＆Kamis，2019)。 深度学习用于根据对不同评论的情感分析来优化建议，这些评论来自不同的社交网站。 进行的实验表明，基于RNN的深度学习情感分析(RDSA)通过提高情感分析的准确性来改善行为，从而反过来为用户提供更好的建议，从而有助于根据用户的需求确定特定职位用户需求(Preethi等人，2017)。 在本文中，我们将讨论流行的深度学习模型，这些模型越来越广泛地用于情感分析中，包括CNN，RNN和各种集成技术。 本文提供了各种情感分析技术的见解。

情绪分析 (Sentiment analysis)

It is a set of techniques / algorithms used to detect the sentiment (positive, negative, or neutral) of a given text. It is a very powerful application of natural language processing (NLP) and finds usage in a large number of industries. It refers to the use of NLP, text analysis, computational linguistics, and biometrics to systematically identify, extract, quantify, and study different states and subjective information. The sentiment analysis sometimes goes beyond the categorization of texts to find opinions and categorizes them as positive or negative, desirable or undesirable. Below figure describes the architecture of sentiment classification on texts. In this, we modify the provided reviews by applying specific filters, and we use the prepared datasets by applying the parameters and implement our proposed model for evaluation(Ghorbani et al., 2020).

它是用于检测给定文本的情绪(正，负或中性)的一组技术/算法。 它是自然语言处理 (NLP)的非常强大的应用程序，并在许多行业中得到了使用。 它指的是使用NLP， 文本分析 ， 计算语言学和生物识别技术来系统地识别，提取，量化和研究不同的状态和主观信息。 情感分析有时会超出文本的分类范围，以找到观点并将其归类为正面或负面，理想或不良。 下图描述了文本中情感分类的体系结构。 在这种情况下，我们通过应用特定的过滤器来修改提供的评论，并通过应用参数来使用准备好的数据集，并实施我们提出的评估模型(Ghorbani等，2020)。

There are three approaches to perform sentiment analysis –

进行情感分析的方法有以下三种：

1. Lexicon based techniques — It can be classified in two types -

1.基于词汇的技术 -可以分为两种类型-

a. Dictionary based — In this approach, classification is done by using dictionary of terms, which can be found in WordNet or SentiWordNet.

一个。 基于字典-在这种方法中，分类是通过使用术语词典完成的，可以在WordNet或SentiWordNet中找到它们。

b. Corpus based — In this approach, classification is done based on the statistical analysis of the content of group of documents using techniques such as hidden Markov models (HMM) , conditional random field (CRF), k-nearest neighbors (k-NN) among others.

b。 基于语料库-在这种方法中，分类是基于对文档组内容的统计分析，使用的技术包括隐马尔可夫模型(HMM)，条件随机场(CRF)，k个近邻(k-NN)其他。

2. Machine learning based techniques — It can be classified in two groups –

2.基于机器学习的技术 -可以分为两类-

a. Traditional Models — It refers to classical techniques of machine learning such as support vector machines , maximum entropy classifier, naive Bayes classifier. The inputs of these models includes sentiment lexicon based features, lexical features, parts of speech, adverbs and adjectives.

一个。 传统模型-指机器学习的经典技术，例如支持向量机，最大熵分类器，朴素贝叶斯分类器。 这些模型的输入包括基于情感词典的功能，词汇功能，词性，副词和形容词。

b. Deep Learning Models — It provides more accurate results than traditional models. It includes models such as CNN, RNN, and DNN. These models address classification problems at document level, sentence level or aspect level.

b。 深度学习模型-比传统模型提供更准确的结果。 它包括诸如CNN，RNN和DNN之类的模型。 这些模型解决了文档级别，句子级别或方面级别的分类问题。

3. Hybrid Approach — It combine machine learning and lexicon based approaches. Sentiment lexicons plays a significant role within most of these approaches. Below figure illustrates taxonomy of various methods including deep-learning for sentiment analysis techniques. Sentiment analysis, whether performed by means of deep learning or traditional machine learning, requires that text training data be cleaned before being used to induce the classification(Dang et al., 2020).

3.混合方法 -它结合了机器学习和基于词典的方法。 情感词典在大多数这些方法中起着重要作用。 下图说明了各种方法的分类法，包括用于情感分析技术的深度学习。 无论是通过深度学习还是传统机器学习进行的情感分析，都要求在将文本训练数据用于归类之前进行清理(Dang等人，2020)。

深度学习 (Deep Learning)

Deep Learning leverages multilayer approach to the hidden layers of neural networks. Traditionally, in machine learning models, features are identified and extracted either manually or using feature selection methods. However, in the case of Deep Learning, features are learned, extracted automatically resulting in higher accuracy and performance. Below figure shows the differences in sentiment polarity classification between the two approaches: traditional machine learning (Support Vector Machine (SVM), Bayesian networks, or decision trees) and deep learning. Artificial neural networks and deep learning currently provide the best solutions to many problems in the fields of image and speech recognition, as well as in natural language processing(Ghorbani et al., 2020). Below figure illustrates differences in sentiment polarity classification between the two approaches: traditional machine learning (Support Vector Machine (SVM), Bayesian networks, or decision trees) and deep learning techniques.

深度学习将多层方法应用于神经网络的隐藏层。 传统上，在机器学习模型中，特征是通过手动或使用特征选择方法来识别和提取的。 但是，在深度学习的情况下，将学习特征并自动提取特征，从而获得更高的准确性和性能。 下图显示了两种方法在情感极性分类上的差异：传统机器学习(支持向量机(SVM)，贝叶斯网络或决策树)和深度学习。 目前，人工神经网络和深度学习为图像和语音识别领域以及自然语言处理领域的许多问题提供了最佳解决方案(Ghorbani等，2020)。 下图说明了两种方法之间的情感极性分类的差异：传统机器学习(支持向量机(SVM)，贝叶斯网络或决策树)和深度学习技术。

Key Deep Learning techniques, which can be used, are listed below –

下面列出了可以使用的关键深度学习技术–

Convolution Neural Networks (CNN) — It is a class of deep neural networks, most commonly used to analyze visual imagery. They are also known as space invariant or shift invariant artificial neural networks, due to shared-weights architecture and translation in-variance characteristics. CNN consists of an input and an output layer, as well as multiple hidden layers. The hidden layers of a CNN typically consist of a series of convolution layers that convolve with a multiplication or other dot product. The activation function is commonly a RELU layer, and is subsequently followed by additional convolutions such as pooling layers, fully connected layers and normalization layers, referred to as hidden layers because their inputs and outputs are masked by the activation function and final convolution. Below is the deep architecture using a 10-layer convolution neural network. Starting from the inputs, this model consists of three conv-pool stages with a convolution and max-pooling each, one flatten layer, two fully-connected layers, and one softmax layer for outputs(Wang & Fey, 2018).

卷积神经网络(CNN) -这是一类深层神经网络，最常用于分析视觉图像。 由于共享权重架构和翻译不变性特征，它们也被称为空间不变或移位不变人工神经网络。 CNN由一个输入层和一个输出层以及多个隐藏层组成 。 CNN的隐藏层通常由一系列与乘法或其他点积卷积的卷积层组成。 激活函数通常是RELU层 ，随后是其他卷积，例如池化层，完全连接的层和归一化层，称为隐藏层，因为它们的输入和输出被激活函数和最终卷积掩盖了。 下面是使用10层卷积神经网络的深度架构。 从输入开始，此模型包括三个卷积级，每个卷积级和最大池级，一个平坦层，两个完全连接层和一个用于输出的softmax层(Wang＆Fey，2018)。

Deep Neural Networks (DNN) — It is an artificial neural network (ANN) with multiple layers between the input and output layers. It finds the correct mathematical manipulation to turn the input into the output, whether it be a linear relationship or a non-linear relationship.

深度神经网络(DNN) -这是一种人工神经网络(ANN)，在输入层和输出层之间具有多层。 无论是线性关系还是非线性关系，它都能找到正确的数学运算以将输入转换为输出。

Recurrent Neural Networks — A recurrent neural network (RNN) is a class of artificial neural networks where connections between nodes form a directed graph along a temporal sequence. This allows it to exhibit temporal dynamic behavior. The main function of RNN is the processing of sequential information on the basis of the internal memory captured by the directed cycles. Unlike traditional neural networks, RNN can remember the previous computation of information and can reuse it by applying it to the next element in the sequence of inputs. A special type of RNN is long short-term memory (LSTM), which is capable of using long memory as the input of activation functions in the hidden layer.

递归神经网络 —递归神经网络(RNN)是一类人工神经网络，其中节点之间的连接沿时间序列形成有向图。 这使其具有时间动态行为。 RNN的主要功能是基于有向循环捕获的内部存储器来处理顺序信息。 与传统的神经网络不同，RNN可以记住先前的信息计算，并且可以通过将其应用于输入序列中的下一个元素来重用它。 RNN的一种特殊类型是长短期记忆(LSTM)，它能够将长记忆用作隐藏层中激活函数的输入。

Above figure illustrates the architecture of LSTM architecture. In this figure, input data is preprocessed to reshape the data for the embedding matrix, next layer is the LSTM and the final layer is fully connected layer for text classification(Dang et al., 2020).

上图说明了LSTM体系结构的体系结构。 在该图中，对输入数据进行了预处理以对嵌入矩阵的数据进行整形，下一层是LSTM，最后一层是用于文本分类的完全连接层(Dang等，2020)。

结论 (Conclusion)

In this article, we discussed the core of deep learning models and the techniques that can be applied to sentiment analysis for social network data. We discussed about various approaches for sentiment analysis including machine learning based, lexicon based and hybrid model. The architectures of CNN, DNN and LSTM are discussed. It is better to combine deep learning techniques with word embedding when performing a sentiment analysis. Also, the effectiveness of the algorithms is largely dependent on the characteristics of the datasets, hence the convenience of testing deep learning methods with more datasets is important in order to cover a greater diversity of characteristics.

在本文中，我们讨论了深度学习模型的核心以及可用于社交网络数据情感分析的技术。 我们讨论了各种情感分析方法，包括基于机器学习，基于词典和混合模型的方法。 讨论了CNN，DNN和LSTM的体系结构。 进行情感分析时，最好将深度学习技术与单词嵌入结合起来。 而且，算法的有效性在很大程度上取决于数据集的特征，因此，为了覆盖更大的特征多样性，测试具有更多数据集的深度学习方法的便利性很重要。

参考书目 (Bibliography)

Araque, O., Corcuera-Platas, I., Sánchez-Rada, J. F., & Iglesias, C. A. (2017). Enhancing deep learning sentiment analysis with ensemble techniques in social applications. Expert Systems with Applications, 77, 236–246. https://doi.org/https://doi.org/10.1016/j.eswa.2017.02.002

Ora的Araque，I。的Corcuera-Platas，JF的Sánchez-Rada和CA的Iglesias(2017)。 在社交应用程序中使用集成技术增强深度学习情感分析。 专家系统与应用 ，77，236-246。 https://doi.org/https://doi.org/10.1016/j.eswa.2017.02.002

Dang, N. C., Moreno-García, M. N., & De la Prieta, F. (2020). Sentiment Analysis Based on Deep Learning: A Comparative Study. Electronics, 9(3), 483. https://doi.org/10.3390/electronics9030483

北卡罗来纳州的当当，明尼苏达州的莫雷诺-加西亚，与德拉普莱塔(F.La 2020)。 基于深度学习的情感分析：一项比较研究。 电子学 ， 9 (3)，483. https://doi.org/10.3390/electronics9030483

Ghorbani, M., Bahaghighat, M., Xin, Q., & Özen, F. (2020). ConvLSTMConv network: a deep learning approach for sentiment analysis in cloud computing. Journal of Cloud Computing, 9(1), 16. https://doi.org/10.1186/s13677-020-00162-1

Ghorbani，M.，Bahaghighat，M.，Xin，Q.，＆Özen，F.(2020年)。 ConvLSTMConv网络：一种用于云计算中情感分析的深度学习方法。 云计算学报 ， 9 (1)，16. https://doi.org/10.1186/s13677-020-00162-1

Goularas, D., & Kamis, S. (2019). Evaluation of Deep Learning Techniques in Sentiment Analysis from Twitter Data. 2019 International Conference on Deep Learning and Machine Learning in Emerging Applications (Deep-ML), 12–17. https://doi.org/10.1109/Deep-ML.2019.00011

Goularas，D.和Kamis，S.(2019年)。 从Twitter数据进行情感分析中的深度学习技术评估。 2019年新兴应用程序深度学习和机器学习国际会议(Deep-ML) ，12-17。 https://doi.org/10.1109/Deep-ML.2019.00011

Preethi, G., Krishna, P. V, Obaidat, M. S., Saritha, V., & Yenduri, S. (2017). Application of Deep Learning to Sentiment Analysis for recommender system on cloud. 2017 International Conference on Computer, Information and Telecommunication Systems (CITS), 93–97. https://doi.org/10.1109/CITS.2017.8035341

Preethi，G.，Krishna，P.V，Obaidat，MS，Saritha，V.，＆Yenduri，S.(2017年)。 深度学习在云推荐系统情感分析中的应用 2017年计算机，信息和电信系统国际会议(CITS) ，93-97。 https://doi.org/10.1109/CITS.2017.8035341

Wang, Z., & Fey, A. M. (2018). Deep learning with convolutional neural network for objective skill evaluation in robot-assisted surgery. International Journal of Computer Assisted Radiology and Surgery, 13, 1959–1970.

Wang，Z.，＆Fey，AM(2018)。 使用卷积神经网络进行深度学习，以评估机器人辅助手术中的客观技能。 国际计算机的辅助放射学和外科学 ，13，1959-1970。

Yadav, A., & Vishwakarma, D. K. (2020). Sentiment analysis using deep learning architectures: a review. Artificial Intelligence Review, 53(6), 4335–4385. https://doi.org/10.1007/s10462-019-09794-5

Yadav，A.＆Vishwakarma，DK(2020)。 使用深度学习架构的情感分析：回顾。 人工智能评论 ， 53 (6)，4335-4385。 https://doi.org/10.1007/s10462-019-09794-5