Automatic Sentiment Analysis Research Articles

Voting intentions on social media and political opinion polls

Opinion polls play an important role in modern democratic processes: they are known to not only affect the outcomes of elections, but also have a significant influence on government policy after elections. Recent years have seen large discrepancies between polls and outcomes at several major elections and referendums, stemming from decreased participation in polls and an increasingly volatile electorate. This calls for new ways to measure public support for political parties. In this paper, we propose a method for measuring the popularity of election candidates on social media using Machine Learning-based Natural Language Processing techniques. The method is based on detecting voting intentions in the data. This is a considerable advance upon earlier work using automatic sentiment analysis. We evaluate the method both intrinsically on a set of hand-labelled social media posts, and extrinsically – by forecasting daily election polls. In the extrinsic evaluation, we analyze data from the 2016 US presidential election, and find that voting intentions measured from social media provide significant additional predictive value for forecasting daily polls. Thus, we demonstrate that the proposed method can be used to interpolate polls both spatially and temporally, thus providing reliable, continuous and fine-grained information about public opinion on current political issues.

Sentiment Analysis Towards Kartu Prakerja Using Text Mining with Support Vector Machine and Radial Basis Function Kernel

Background: The introduction of Kartu Prakerja (Pre-employment Card) Programme, henceforth KPP, which was claimed to have launched in order to improve the quality of workforce, spurred controversy among members of the public. The discussion covered the amount of budget, the training materials and the operations brought out various reactions. Opinions could be largely divided into groups: the positive and the negative sentiments.Objective: This research aims to propose an automated sentiment analysis that focuses on KPP. The findings are expected to be useful in evaluating the services and facilities provided.Methods: In the sentiment analysis, Support Vector Machine (SVM) in text mining was used with Radial Basis Function (RBF) kernel. The data consisted of 500 tweets from July to October 2020, which were divided into two sets: 80% data for training and 20% data for testing with five-fold cross validation.Results: The results of descriptive analysis show that from the total 500 tweets, 60% were negative sentiments and 40% were positive sentiments. The classification in the testing data show that the average accuracy, sensitivity, specificity, negative sentiment prediction and positive sentiment prediction values were 85.20%; 91.68%; 75.75%; 85.03%; and 86.04%, respectively.Conclusion: The classification results show that SVM with RBF kernel performs well in the opinion classification. This method can be used to understand similar sentiment analysis in the future. In KPP case, the findings can inform the stakeholders to improve the programmes in the future. Keywords: Kartu Prakerja, Sentiment Analysis, Support Vector Machine, Text Mining, Radial Basis Function

Evaluation of Ensemble Algorithms and Deep Learning Transformers in Medical Sentiment Prediction

Social media continuously produces digital information that can be used to improve service quality. In this aspect sentiment prediction, automated analysis of written user reviews, is an important research area from service quality point of view. Online sentiment prediction is a rich research area from e-business perspective. However, identification of sentiment from medical service user reviews is particularly researched less frequently. From Turkish language point of view, the medical informatics literature needs more research to design automated medical sentiment systems. Automated sentiment analysis systems particularly make use of Machine Learning (ML) algorithm in tandem with Natural Language Processing (NLP) methods to address written user reviews. In this work, ensemble learning approaches are compared with newly developed deep learning variations, Bidirectional Encoder Representations from Transformers (BERT), to investigate medical sentiments. As the obtained results are evaluated, it is observed that newly proposed transformer models are perfectly successful to identify sentiment of Turkish medical reviews.

When do service employees smile? Response‐dependent emotion regulation in emotional labor

SummaryWe advance the theoretical and practical understanding of affect in service interactions by conceptualizing employees and customers as concurrent participants in the same interaction. We analyzed employees' emotional labor requirements, which comprise both the well‐recognized requirement to display positive affect (i.e., acting is response independent) and the less‐recognized requirement to attend to customers' affective displays (i.e., acting is response dependent). We found support for our hypotheses across two studies, which compare the affective displays of employees and customers within the same interactions. In Study 1, we examined field data comprising 1 320 392 customer and employee messages from 164 899 real‐life chat‐based service interactions and used automated sentiment analysis to identify displays of positive and negative affect. In Study 2, we used simulated service interactions to examine discrete emotions. Using different methodologies, both Studies 1 and 2 found that employees and customers differ in their response‐dependent affective behaviors. Study 2 also demonstrated that employee response‐dependent affective behavior improves customer outcomes. Our analyses enrich the understanding of emotional labor, contribute to the theory of affect in social interactions by considering concurrent yet differing affective behaviors of partners in the same interaction, and suggest new and exciting methods for Organizational Behavior research.

Weighted holoentropy-based features with optimised deep belief network for automatic sentiment analysis: reviewing product tweets

ABSTRACT In this paper, a novel sentiment analysis model is implemented, which consists of six stages: (i) Pre-processing, (ii) Keyword extraction and its sentiment categorisation, (iii) Semantic word extraction, (iv) Semantic similarity checking, (v) Feature extraction, and (vi) Classification. Initially, the Mongodb documented tweets are subjected to pre-processing that includes steps such as stop word removal, stemming, and blank space removal. Accordingly, from the pre-processed tweets, the keywords are extracted. Based on the extracted keywords, the prevailing semantic words are extracted after classifying the sentimental keywords. Further, the evaluation of the semantic similarity score with the keywords takes place. Also, it exploits joint holoentropy and cross holoentropy. Here, the extraction of weighted holoentropy features is the main contribution, where a weight function is multiplied by the holoentropy features. To improve the performance of classification, a constant term is used for calculating weight function. It is tuned or optimised in such a way that the accuracy of the proposed method is better. The optimisation strategy uses the hybrid model that merges Particle Swarm Optimisation (PSO) into Whale Optimisation Algorithm (WOA). Hence, the proposed algorithm is named as Swarm Velocity-based WOA (SV-WOA). Finally, the analysis is done to prove the efficiency of the proposed model.

Analysis of sentiment in tweets addressed to a single domain-specific Twitter account: Comparison of model performance and explainability of predictions

Many institutions and companies find it valuable to know how people feel about their ventures; hence, scientific research in sentiment analysis has been intensely developed over time. Automated sentiment analysis can be considered as a machine learning (ML) prediction task, with classes representing human affective states. Due to the rapid development of ML and deep learning (DL), improvements in automatic sentiment analysis performance are achieved almost every year. Since 2013, Semantic Evaluation (SemEval) has hosted a worldwide community-acknowledged competition that allows for comparisons of recent innovations. The sentiment analysis tasks focus on assessing sentiment in Twitter posts authored by various publishers and addressing multiple subjects. Our study aimed to compare selected popular and recent natural language processing methods using a new data set of Twitter posts sent to a single Twitter account. For improved comparability of our experiments with SemEval, we adopted their metrics and also deployed our models on data published for SemEval-2017. In addition, we investigated if an unsupervised ML technique applied for the detection of topics in tweets can be leveraged to improve the predictive performance of a selected transformer model. We also demonstrated how a recent explainable artificial intelligence technique can be used in Twitter sentiment analysis to gain a deeper understanding of the models’ predictions. Our results show that the most recent DL language modeling approach provides the highest quality; however, this quality comes at reduced model transparency.

Exploring the impact of chatbots on consumer sentiment and expectations in retail

Retailers are increasingly using conversational AI (chatbots) for customer service due to the perceived benefits and reduced operational costs of this emerging technology. Yet our understanding of how consumers perceive interactions with chatbots, and how these interactions may influence other consumer service programs, remains limited. This paper investigates the differences in consumers' sentiments towards chatbots across retail sectors, and the influence chatbots have on consumers’ sentiments and expectations towards other service interactions with online human agents. Using a hybrid automated sentiment analysis approach, we identify that (1) overall sentiment towards bots are less negative than sentiment towards online human agents; (2) these sentiments differ across fashion and telecommunications sectors, and finally (3) sentiments towards online human agents in both sectors become more negative after a retailer implements a chatbot.

The Automatic Analysis of Emotion in Political Speech Based on Transcripts

ABSTRACT Automatic sentiment analysis is used extensively in political science. The digitization of legislative transcripts has increased the potential application of established tools for the automated analyses of emotion in text. Unlike in writing, however, expressing emotion in speech involves intonation, facial expressions, and body language. Drawing on a new dataset of annotated texts and videos from the Canadian House of Commons, this paper does three things. First, we examine whether transcripts capture the emotional content of speeches. We find that transcripts capture sentiment, but not emotional arousal. Second, we compare strategies for the automated analysis of sentiment in text. We find that leading approaches performed reasonably well, but sentiment dictionaries generated using word embeddings surpassed these other approaches. Finally, we test the robustness of the approach based on word embeddings. Although the methodology is reasonably robust to alternative specifications, we find that dictionaries created using word embeddings are sensitive to the choice of seed words and to training corpus size. We conclude by discussing the implications for analyses of political speech.

E-learning Text Sentiment Classification Using Hierarchical Attention Network (HAN)

Massive Open Online Courses (MOOCs) have recently become a very motivating research field in education. Analyzing MOOCs discussion forums presents important issues since it can create challenges for understanding and appropriately identifying student sentiment behaviours. Using the high effectiveness of deep learning, this study aims to classify forum posts based on their sentiment polarity using two experiments. The first use the three known sentiment labels (positive/negative/neutral) and the second one employs sevens labels. The classification method implemented the Hierarchical Attention Network (HAN) algorithm; it combines the attention mechanism with a hierarchical network that simulates the same hierarchical structure of the document. The analysis of 29604 discussion posts from Stanford University affirms the effectiveness of our model. HAN achieved a classification accuracy of 70.3%, which surpassed the other prediction results using usual text classification models. These results are promising and have implications on the future development of automated sentiment analysis tool on e-learning discussion forum.

Analysis of the sentiment of the text in the Ukrainian language

The article is devoted to the study of the methods of automatic sentiment analysis and the identification of the most effective methods of analysis of the sentiment of the text in the Ukrainian language. In the course of the work, the problem of generating a dataset in Ukrainian was solved based on user reviews of mobile applications. The resulting dataset was used to conduct an experiment to identify the optimal binary classification algorithm for texts in the Ukrainian language, and a binary classifier model was built based on the results of the experiment. The problem of binary classification of Ukrainian text was solved using a pretrained multilingual BERT model using a generated dataset.

Methadone and suboxone® mentions on twitter: thematic and sentiment analysis

Background According to the latest medical evidence, Methadone and buprenorphine-naloxone (Suboxone®) are effective treatments for opioid use disorder (OUD). While the evidence basis for the use of these medications is favorable, less is known about the perceptions of the general public about them. Objective This study aimed to use Twitter to assess the public perceptions about methadone and buprenorphine-naloxone, and to compare their discussion contents based on themes/topics, subthemes, and sentiment. Methods We conducted a descriptive analysis of a small and automatic analysis of a large volume of microposts (“tweets”) that mentioned “methadone” or “suboxone”. In the manual analysis, we categorized the tweets into themes and subthemes, as well as by sentiment and personal experience, and compared the information posted about these two medications. We performed automatic topic modeling and sentiment analysis over large volumes of posts and compared the outputs to those from the manual analyses. Results We manually analyzed 900 tweets, most of which related to access (15.3% for methadone; 14.3% for buprenorphine-naloxone), stigma (17.0%; 15.5%), and OUD treatment (12.8%; 15.6%). Only a small proportion of tweets (16.4% for Suboxone® and 9.3% for methadone) expressed positive sentiments about the medications, with few tweets describing personal experiences. Tweets mentioning both medications primarily discussed MOUD broadly, rather than comparing the two medications directly. Automatic topic modeling revealed topics from the larger dataset that corresponded closely to the manually identified themes, but sentiment analysis did not reveal any notable differences in chatter regarding the two medications. Conclusions Twitter content about methadone and Suboxone® is similar, with the same major themes and similar sub-themes. Despite the proven effectiveness of these medications, there was little dialogue related to their benefits or efficacy in the treatment of OUD. Perceptions of these medications may contribute to their underutilization in combatting OUDs.

Twitter users’ coping behaviors during the COVID-19 lockdown: an analysis of tweets using mixed methods

PurposeUsing data from Twitter, the purpose of this paper is to assess the coping behaviour and reactions of social media users in response to the initial days of the COVID-19-related lockdown in different parts of the world.Design/methodology/approachThis study follows the quasi-inductive approach which allows the development of pre-categories from other theories before the sampling and coding processes begin, for use in those processes. Data was extracted using relevant keywords from Twitter, and a sample was drawn from the Twitter data set to ensure the data is more manageable from a qualitative research standpoint and that meaningful interpretations can be drawn from the data analysis results. The data analysis is discussed in two parts: extraction and classification of data from Twitter using automated sentiment analysis; and qualitative data analysis of a smaller Twitter data sample.FindingsThis study found that during the lockdown the majority of users on Twitter shared positive opinions towards the lockdown. The results also found that people are keeping themselves engaged and entertained. Governments around the world have also gained support from Twitter users. This is despite the hardships being faced by citizens. The authors also found a number of users expressing negative sentiments. The results also found that several users on Twitter were fence-sitters and their opinions and emotions could swing either way depending on how the pandemic progresses and what action is taken by governments around the world.Research limitations/implicationsThe authors add to the body of literature that has examined Twitter discussions around H1N1 using in-depth qualitative methods and conspiracy theories around COVID-19. In the long run, the government can help citizens develop routines that help the community adapt to a new dangerous environment – this has very effectively been shown in the context of wildfires in the context of disaster management. In the context of this research, the dominance of the positive themes within tweets is promising for policymakers and governments around the world. However, sentiments may wish to be monitored going forward as large-spikes in negative sentiment may highlight lockdown-fatigue.Social implicationsThe psychology of humans during a pandemic can have a profound impact on how COVID-19 shapes up, and this shall also include how people behave with other people and with the larger environment. Lockdowns are the opposite of what societies strive to achieve, i.e. socializing.Originality/valueThis study is based on original Twitter data collected during the initial days of the COVID-19-induced lockdown. The topic of “lockdowns” and the “COVID-19” pandemic have not been studied together thus far. This study is highly topical.

VADER Sentiment Analysis without and with English Punctuation Marks

Sentiment analysis is the foremost task in Natural Language Processing to understand the user’s attitude (positive, neutral, or negative) by capturing their thoughts, opinions, and feeling about a particular product. This helps companies to fulfill customer satisfaction and make better future decisions about the product. Various techniques have been used in the literature forsentiment analysis, such as polarity scores, classifications, and automated sentiment analysis. In this paper, Valence Aware Dictionary and sEntiment Reasoner (VADER) sentiment analysis tool has been employed on a Twitter dataset (downloaded from https://www.kaggle.com). The study aims to measure the performance of VADER sentiment while concatenating fourteen English language punctuations marks, including Exclamation (!), Comma (,), Full Stop (.), Question Mark (?), Round Brackets (), Curly Brackets {}, Square Brackets [], Colon (:), Apostrophe (‘), Dash (-), Hyphen (--), Semi-Colon (;), Slash (/), Quotation Mark (“ ”) and to observe whether the polarity (positive, neutral and negative) of a sentence changes or remains the same. After the analysis, the study found that Exclamation (!) maximizes the average positive polarity and average negative polarity and lowers the average neutralpolarity. The Hyphen (--) and Comma (,) increase the average positive and neutral polarity and decrease the aver-age negative polarity. For Round Brackets (), Curly Brackets {}, Square Brackets [], Colon (:), Apostrophe (‘), Dash (-), Semi-Colon (;), Slash (/) and Full Stop (.) the average positive and average neutral polarity decreases and average negative polarity increases.

Monitoring drug trends in the digital environment-New methods, challenges and the opportunities provided by automated approaches.

Developments in information technology have impacted on all areas of modern life and in particular facilitated the growth of globalisation in commerce and communication. Within the drugs area this means that both drugs discourse and drug markets have become increasingly digitally enabled. In response to this, new methods are being developed that attempt to research and monitor the digital environment. In this commentary we present three case studies of innovative approaches and related challenges to software-automated data mining of the digital environment: (i) an e-shop finder to detect e-shops offering new psychoactive substances, (ii) scraping of forum data from online discussion boards, (iii) automated sentiment analysis of discussions in online discussion boards. We conclude that the work presented brings opportunities in terms of leveraging data for developing a more timely and granular understanding of the various aspects of drug-use phenomena in the digital environment. In particular, combining the number of e-shops, discussion posts, and sentiments regarding particular substances could be used for ad hoc risk assessments as well as longitudinal drug monitoring and indicate "online popularity". The main challenges of digital data mining involve data representativity and ethical considerations.

A mixed-method analysis of the #SugarTax debate on Twitter.

To explore the #SugarTax debate on Twitter to assess actors involved, their connections and the topics being discussed during the implementation and first anniversary of the UK Soft Drink Industry Levy. The structure of the #SugarTax debate on Twitter was assessed using social network analysis. The actors involved, their connections and the topics of discussion taking place were also explored using content, sentiment and thematic analyses. Twitter between 2017 and 2019. Twitter users engaging in discussions relating to the hashtag 'SugarTax'. Tweets (n 5366) posted between 5 August 2017 and 7 May 2019 containing #SugarTax were downloaded from Twitter using Network Overview for Discovery and Exploration in Excel. The network included 1883 users, with 686 unique edges and 4679 edges with duplicates. The majority of tweets were negative in sentiment, when assessed by both automatic (64 %, n 141) and manual sentiment analysis (52 %, n 115) methods. Nine key themes were identified and grouped into two groups according to 'support for a sugar or SSB tax' or 'opposition for a sugar or SSB tax'. Twitter was used as a platform for debating the benefits and limitations of sugar-sweetened beverage taxes. The findings indicate that numerous actors are involved in the debates on Twitter, with advocates and lobbyists using the platform to raise support for their campaigns and reshape public perceptions. The findings and the methods used may be of interest to policymakers as well as to academics and members of the public looking to explore and engage in policy debates.

Air-travelers' concerns emerging from online comments during the COVID-19 outbreak.

This research aims to assess air travelers' concerns affected by the Coronavirus pandemic, expressed in the comments they wrote online. A sample of 639 comments written on the Italian National Consumer Union website and related to the airline industry was assessed through an automated sentiment analysis in this study. The achieved results showed that travelers' concerns were directed mainly towards compensations, cancellations, and COVID-19 and at the same time, they had mixed and unpredictable feelings. This element suggests that consumers may have understood that airline companies are facing unsustainable cash-flow and revenue situations. Moreover, all our hypotheses, grounded on existing literature, were refuted. Accordingly, we argue that the actual context prevents assessments based on previous assumptions, and studies related to the impact of COVID-19 need to be conducted anew.

What Makes Content Engaging? How Emotional Dynamics Shape Success

Abstract Some cultural products (e.g., books and movies) catch on and become popular, while others fail. Why? While some have argued that success is unpredictable, we suggest that period-to-period shifts in sentiment—what we term sentiment volatility—enhance engagement. Automated sentiment analysis of over 4,000 movies demonstrates that more volatile movies are evaluated more positively. Consistent with the notion that sentiment volatility makes experiences more stimulating, the effect is stronger in genres where evaluations are more likely to be driven stimulation (i.e., thrillers rather than romance). Further, analysis of over 30,000 online articles demonstrate that people are more likely to continue reading more volatile articles. By manipulating sentiment volatility in follow-up experiments, we underscore its causal impact on evaluations, and provide evidence for the role of stimulation in these effects. Taken together, the results shed light on what drives engagement, the time dynamics of sentiment, and cultural analytics or why some cultural items are more successful.

Deep Learning for Sentiment Analysis of Tunisian Dialect

Automatic sentiment analysis has become one of the fastest growing research areas in the Natural Language Processing (NLP) field. Despite its importance, this is the first work towards sentiment analysis at both aspect and sentence levels for the Tunisian Dialect in the field of Tunisian supermarkets. Therefore, we experimentally evaluate, in this paper, three deep learning methods, namely convolution neural networks (CNN), long short-term memory (LSTM), and bi-directional long-short-term-memory (Bi-LSTM). Both LSTM and Bi-LSTM constitute two major types of Recurrent Neural Networks (RNN). Towards this end, we gathered a corpus containing comments posted on the official Facebook pages of Tunisian supermarkets. To conduct our experiments, this corpus was annotated on the basis of five criteria (very positive/ positive/ neutral/ negative/ very negative) and other twenty categories of aspects. In this evaluation, we show that the gathered features can lead to very encouraging performances through the use of CNN and Bi-LSTM neural networks.

Neural Networks with Emotion Associations, Topic Modeling and Supervised Term Weighting for Sentiment Analysis.

Automated sentiment analysis is becoming increasingly recognized due to the growing importance of social media and e-commerce platform review websites. Deep neural networks outperform traditional lexicon-based and machine learning methods by effectively exploiting contextual word embeddings to generate dense document representation. However, this representation model is not fully adequate to capture topical semantics and the sentiment polarity of words. To overcome these problems, a novel sentiment analysis model is proposed that utilizes richer document representations of word-emotion associations and topic models, which is the main computational novelty of this study. The sentiment analysis model integrates word embeddings with lexicon-based sentiment and emotion indicators, including negations and emoticons, and to further improve its performance, a topic modeling component is utilized together with a bag-of-words model based on a supervised term weighting scheme. The effectiveness of the proposed model is evaluated using large datasets of Amazon product reviews and hotel reviews. Experimental results prove that the proposed document representation is valid for the sentiment analysis of product and hotel reviews, irrespective of their class imbalance. The results also show that the proposed model improves on existing machine learning methods.

Discovering Sentimental Interaction via Graph Convolutional Network for Visual Sentiment Prediction

With the popularity of online opinion expressing, automatic sentiment analysis of images has gained considerable attention. Most methods focus on effectively extracting the sentimental features of images, such as enhancing local features through saliency detection or instance segmentation tools. However, as a high-level abstraction, the sentiment is difficult to accurately capture with the visual element because of the “affective gap”. Previous works have overlooked the contribution of the interaction among objects to the image sentiment. We aim to utilize interactive characteristics of objects in the sentimental space, inspired by human sentimental principles that each object contributes to the sentiment. To achieve this goal, we propose a framework to leverage the sentimental interaction characteristic based on a Graph Convolutional Network (GCN). We first utilize an off-the-shelf tool to recognize objects and build a graph over them. Visual features represent nodes, and the emotional distances between objects act as edges. Then, we employ GCNs to obtain the interaction features among objects, which are fused with the CNN output of the whole image to predict the final results. Experimental results show that our method exceeds the state-of-the-art algorithm. Demonstrating that the rational use of interaction features can improve performance for sentiment analysis.