Twitter Content Research Articles

Together we can do it! A roadmap to effectively tackle propaganda-related tasks

PurposeIn this paper, we address the need to study automatic propaganda detection to establish a course of action when faced with such a complex task. Although many isolated tasks have been proposed, a roadmap on how to best approach a new task from the perspective of text formality or the leverage of existing resources has not been explored yet.Design/methodology/approachWe present a comprehensive study using several datasets on textual propaganda and different techniques to tackle it. We explore diverse collections with varied characteristics and analyze methodologies, from classic machine learning algorithms, to multi-task learning to utilize the available data in such models.FindingsOur results show that transformer-based approaches are the best option with high-quality collections, and emotionally enriched inputs improve the results for Twitter content. Additionally, MTL achieves the best results in two of the five scenarios we analyzed. Notably, in one of the scenarios, the model achieves an F1 score of 0.78, significantly surpassing the transformer baseline model’s F1 score of 0.68.Research limitations/implicationsAfter finding a positive impact when leveraging propaganda’s emotional content, we propose further research into exploiting other complex dimensions, such as moral issues or logical reasoning.Originality/valueBased on our findings, we provide a roadmap for tackling propaganda-related tasks, depending on the types of training data available and the task to solve. This includes the application of MTL, which has yet to be fully exploited in propaganda detection.

Mapping the discursive landscape of data activism: Articulations and actors in an emerging movement

Growing awareness of the societal consequences of datafication in recent years has given rise to a new form of civil society engagement called data activism. This article examines the discourse surrounding data activism on the social media platform Twitter. Through a mixed-methods approach combining computational analysis of Twitter content and close readings of Twitter profiles, we explore how new forms of civil society action related to data justice are articulated and linked to other forms of activism, conflicts and problems, and the actors involved in these articulations. Our analysis reveals a distinction between two articulatory patterns in the data activism discourse. The first involves grassroots actors, such as community organisations and individual citizens, who challenge existing power structures and advocate for social change. The second, on the other hand, is associated with academics, capitalists and policymakers who already hold positions of power and influence. This asymmetry is consistent with previous findings in data activism research. We encourage future research to extend these patterns, using additional methods and case studies, to further refine and contextualise the understanding of data activism within the civil society realm.

Understanding the current backlash against LGBTIQ+ rights through the lens of heteroactivism: A case study of the International Organization for the Family’s transnational norm diffusion on Twitter

Abstract A new generation of transnational anti-gender actors are framing themselves as human rights champions and protectors of the rights of the “natural family”. To better understand these actors’ norm diffusion, including their re-styling of anti-gender narratives in contemporary iterations of heteroactivism and its potential threat to LGBTIQ+ rights, in this article we analyse the social media tactics of a key transnational anti-gender actor: the International Organization for the Family (IOF). The analysis is focused on the organisation’s Twitter (now X) account and we draw on theories of network media logic, connective action, and connective emotions. Two periods of activism – the first a low-intensity period in 2021 and the second a high-intensity period in 2022 – were purposefully selected for the analysis. In this article, we identify differences between the two periods, noting that the period covering the flagship event, the World Congress of Families (WCF) in 2022, was markedly less LGBTIQ-hostile and adversarial than the first period. IOF thus appears to be inspired by heteroactivist frames during the WCF and abides by logic that should trigger user interaction and content spreading. The lack of interactions and engagement with IOF Twitter content in either period indicates the need for more research on which logics apply to anti-gender audiences. We discuss the findings and what they may imply in a context like Sweden, a country whose self-image as a global champion for women’s and LGBTIQ+ rights may make it ill-equipped to counter the onslaught of transnational actors’ norm entrepreneurial activities.

Social Media, Sentiments and Political Discourse ? An Exploratory Study of the 2021 Canadian Federal Election

Social media are widely used for online political discourse. Opinions shared on social media have different sentiments associated with them. Given the very high adoption rates of Twitter (now X) among adults, those who share their opinions on Twitter (X) not only represent a sizable segment of the society, but also influence (through emotion contagion) an even larger segment who are passive (non-contributing) users of the platform. Further, the discourse that is initiated on Twitter (X) typically spreads to other more traditional media. As a result, Twitter (X) is influential, which makes it useful to understand the factors related to the sentiments expressed in tweets. Such understanding can help policymakers to take actions that align with public needs and priorities. This research focuses on identifying the drivers (keywords) of sentiments associated with political discourse on Twitter (X). We also explore virality, i.e., how much a message (the tweet) spreads, and the relationship between sentiments and virality. Finally, we explore whether the clustering of tweets among sentiment and virality groups can improve the potential of social media content for predicting election results. Sentiment Analysis of 764,000 tweets related to the 2021 Canadian Federal election was followed by text clustering to identify sentiment-driving topics. We found some keywords predominantly present within a positive or negative sentiment that are suggestive of entities or ideas to invest in or mitigate by political decision makers. We were also able to find partial evidence for “negativity bias” by detecting a negative relationship between sentiment (positivity) and virality (number of retweets). Finally, we demonstrated that high positivity on the political discourse does not reflect election outcomes and examining Twitter (X) content in more neutral groups can improve predictive power. Our findings have implications for political decision makers and social media analytics researchers.

Boosting Institutional Identity on X Using NLP and Sentiment Analysis: King Faisal University as a Case Study

Universities increasingly leverage social media platforms, especially Twitter, for news dissemination, audience engagement, and feedback collection. King Faisal University (KFU) is dedicated to enhancing its institutional identity (ID), grounded in environmental sustainability and food security, encompassing nine critical areas. This study aims to assess the impact of KFU’s Twitter interactions on public awareness of its institutional identity using systematic analysis and machine learning (ML) methods. The objectives are to: (1) Determine the influence of KFU’s Twitter presence on ID awareness; (2) create a dedicated dataset for real-time public interaction analysis with KFU’s Twitter content; (3) investigate Twitter’s role in promoting KFU’s institutional identity across 9-ID domains and its changing impact over time; (4) utilize k-means clustering and sentiment analysis (TFIDF and Word2vec) to classify data and assess similarities among the identity domains; and (5) apply the categorization method to process and categorize tweets, facilitating the assessment of word meanings and similarities of the 9-ID domains. The study also employs four ML models, including Logistic Regression (LR) and Support Vector Machine (SVM), with the Random Forest (RF) model combined with Word2vec achieving the highest accuracy of 100%. The findings underscore the value of KFU’s Twitter data analysis in deepening the understanding of its ID and guiding the development of effective communication strategies.

A computational approach to cryptocurrency marketing on social media

This study aims to explore social media content associated with cryptocurrency marketing. We employ unsupervised Latent Dirichlet allocation topic modeling and sentiment analysis techniques to 98,716 tweets to examine Twitter (now known as X) content for subjects and sentiments related to cryptocurrency. Our findings reveal that cryptocurrency tweets fell into four categories, with ‘cryptocurrency trading,’ ‘NFT airdrop,’ ‘cryptocurrency affiliate program,’ and ‘Dogecoin on social media’ being the most popular. Furthermore, most of these topics exhibited positive sentiments. This study contributes theoretically by integrating cryptocurrency marketing into the diffusion of innovation paradigm. In addition, it offers strategic insights for digital marketers in identifying prevalent topics and sentiments related to cryptocurrency, enabling the tailoring of affiliate marketing communication strategies on social media.

Exploring Causal Relationships in Mental Health Literacy Through Twitter Content: A Machine Learning Approach

IntroductionThe concept of Mental Health Literacy (MHL) is inherently multidimensional. However, the interrelationships among its various dimensions remain insufficiently elucidated. In recent years, the textual analysis of social media posts has emerged as a promising methodological approach for longitudinal research in this domain.ObjectivesThis study aimed to investigate whether temporal causal associations exist between recognition of mental illness (R), mental illness stigma (S), help-seeking efficacy (HE), maintenance of positive mental health (M), and help-seeking attitude (HA).MethodsTweets were collocted at three distinct time points: T1, T2, and T3, spanning the period from November 1, 2021, to December 31, 2022. We employed a machine-learning approach to categorize the posts into five MHL facets. Using these facets, we trained a machine learning model, specifically Bidirectional Encoder Representations from Transformers (BERT), to determine the MHL scores. To be eligible, an account must have an R facet score at T1, and M, S, HE facet scores at T2, as well as an HA facet score at T3. In total, we retrieved 4,471,951 MHL-related tweets from 941 users. We further employed structural equation modeling to validate the causal relationships within the MHL framework.ResultsIn the evaluation, BERT achieved average accuracy scores exceeding 89% across the five MHL facets in the validation set, along with F1-scores ranging between 0.75 and 0.89. Among the five MHL facets—maintenance of positive mental health, recognition of mental illness, help-seeking efficacy, and help-seeking attitudes—each demonstrated a statistically significant positive correlation with the others. Conversely, mental illness stigma exhibited a statistically significant negative correlation with the remaining four facets. In the analysis using single-mediation models, each of the individual mediator variables—namely, mental illness stigma, help-seeking efficacy, and maintenance of positive mental health—exhibited significant indirect effects. In the multiple-mediation model, two mediator variables—help-seeking efficacy and maintenance of positive mental health—demonstrated significant indirect effects. These findings suggested that the recognition of mental illness exerted an influence on help-seeking attitudes through one or more of these mediators.ConclusionsBy leveraging machine learning techniques for the textual analysis of social media and employing a longitudinal research design with panel data, this study elucidates the potential mechanisms through which the MHL framework influences attitudes toward seeking mental health services. These insights hold significant implications for the design of future interventions and the development of targeted policies aimed at promoting help-seeking behaviors.Disclosure of InterestNone Declared

Emotions on Twitter as crisis imprint in high-trust societies: Do ambient affiliations affect emotional expression during the pandemic?

During the Covid-19 crisis, citizens turned to Twitter for information seeking, emotional outlet and sense-making of the crisis, creating ad hoc social communities using crisis-specific hashtags. The theory of ambient affiliation posits that the use of hashtags upscales the call to affiliate with the values expressed in the tweet. Given the deep functional tie between values and emotions, hashtag use might further amplify certain emotions. While emotions in crises-hashtagged communities have been previously investigated, the hypothesis of amplification of emotions through hashtag use has not yet been tested. We investigate such effect during the Covid-19 crisis in a scenario of high-trust Nordic societies, focusing on non-hashtagged, crisis hashtagged (e.g., '#Covid-19') and threat hashtagged (e.g., '#misinformation') tweets. To do so we apply XLM-RoBERTa to estimate Anger, Fear, Sadness, Disgust, Joy and Optimism. Our results revealed that crisis-hashtagged (#Covid-19) tweets expressed more negative emotions (Anger, Fear, Disgust and Sadness) and less positive emotions (Optimism and Joy) than non-hashtagged Covid-19 tweets for all countries except Finland. Threat tweets (#misinformation) expressed even more negative emotions (Anger, Fear, Disgust) and less positive emotions (Optimism and Joy) than #Covid-19 tweets, with a particularly large effect for Anger. Our findings provide useful context for previous research on collective emotions during crises, as most Twitter content is not hashtagged, and given the faster spread of emotionally charged content, further support the special focus on specific ad hoc communities for crisis and threat management and monitoring.

Semantic Networks of Election Fraud: Comparing the Twitter Discourses of the U.S. and Korean Presidential Elections

Traditional news outlets, such as newspapers and television, are no longer major sources of news. These media channels have been replaced by social platforms, which have increased in value as information distributors. This change in communication is an underlying reason for the election fraud controversies that occurred in the United States and South Korea, which hold high standards of democracy, during similar periods. This study investigates a model for sharing political disputes over social networks, especially Twitter, and illustrates the influence of political polarization. This study examines Twitter content around the presidential elections in the United States and South Korea in 2020 and 2022, respectively. It applies semantic network analysis and structural topic modeling to describe and compare the dynamics of online discourse on the issue of election fraud. The results show that online spaces such as Twitter serve as public spheres for discussion among active political participants. Social networks are key settings for forming and spreading election fraud controversies in the United States and South Korea, with differences in content. In addition, the study applies large-volume text data and new analytical methods such as the structural topic model to examine the in-depth relationships among political issues in cyberspace.

Diurnal patterns in Twitter sentiment in Italy and United Kingdom are correlated.

Diurnal variations in indicators of emotion have been reliably observed in Twitter content, but confirmation of their circadian nature has not been possible due to the many confounding factors present in the data. We report on correlations between those indicators in Twitter content obtained from 9 cities of Italy and 54 cities in the United Kingdom, sampled hourly at the time of the 2020 national lockdowns. This experimental setting aims at minimizing synchronization effects related to television, eating habits, or other cultural factors. This correlation supports a circadian origin for these diurnal variations, although it does not exclude the possibility that similar zeitgebers exist in both countries including during lockdowns.

Enhancing Stock Price Prediction with Deep Cross-Modal Information Fusion Network

Stock price prediction is considered a classic and challenging task, with the potential to aid traders in making more profitable trading decisions. Significant improvements in stock price prediction methods based on deep learning have been observed in recent years. However, most existing methods are reliant solely on historical stock price data for predictions, resulting in the inability to capture market dynamics beyond price indicators, thus limiting their performance to some extent. Therefore, combining social media text with historical stock price information has proposed a novel stock price prediction method, known as the Deep Cross-Modal Information Fusion Network (DCIFNet). The process is initiated by DCIFNet, which employs temporal convolution processes to encode stock prices and Twitter content. This ensures that each element has sufficient information about its surrounding components. Following this, the outcomes are inputted into a cross-modal fusion structure based on transformers to enhance the integration of crucial information from stock prices and Twitter content. Lastly, a multi-graph convolution attention network is introduced to depict the relationships between different stocks from diverse perspectives. This facilitates the more effective capturing of industry affiliations, Wikipedia references, and associated relationships among linked stocks, ultimately leading to an enhancement in stock price prediction accuracy. Trend prediction and simulated trading experiments are conducted on high-frequency trading datasets spanning nine different industries. Comparative assessments with the Multi-Attention Network for Stock Prediction (MANGSF) method, as well as ablation experiments, confirm the effectiveness of the DCIFNet approach, resulting in an accuracy rate of 0.6309, a marked improvement compared to representative methods in the field.

Twitter content strategies to maximize engagement: The case of Thai Banks

Purpose– This study aims to examine how the marketer-generated content (MGC) strategies of banks affect their followers’ engagement on Twitter. Design/methodology/approach – A sample of MGC tweets from official Twitter accounts of both public and private banks in Thailand was collected using a Twitter analytic tool. Message features, content, intent, and post engagement were manually and automatically extracted from these tweets. Proposed relationships were analyzed using descriptive statistics and hierarchical regression analyses. Findings – The results indicate the negative influences of two control variables, media type (text), timing (evening), and some sub-types of information-sharing content and emotion-evoking content, as well as positive impacts of public bank type, some message features i.e., hashtags and mentions, media type, some content regarding emotion-evoking and action-inducing, and intent (request). Practical implications – Useful guidance for bank marketers to maximize engagement is offered. The suggestion for revising bank content strategies on Twitter are presented. Originality/value – This study examines the content strategy from a different perspective, which has rarely been employed in the literature. Additionally, this study reveals the Twitter content strategies of the Thai financial sector, which is a crucial sector supporting others. Understanding consumer behavioral engagement could help them to publish the right MGCs.

Forecasting inflation with twitter

We use Twitter content to generate an indicator of attention allocated to inflation. The analysis corresponds to Argentina for the period 2012-2019. The attention index provides valuable information regarding future levels of inflation. A one standard deviation increment in the index is followed by an increment of approximately 0.4% in inflation in the consecutive month. Out-of-sample exercises confirm that social media content allows for gains in forecast accuracy. Beyond point forecasts, the index provides valuable information regarding inflation uncertainty. The proposed indicator compares favorably with other indicators such as media content, media tweets, google search intensity and consumer surveys.

Public perception of the resumption of HPV vaccine recommendation in Japan: Twitter content analysis.

On 12 November 2021, after a stagnation of 8.5 years, Japan decided to resume proactive recommendations for HPV vaccines. However, it is not known how the public reacted to the resumption of proactive recommendations for HPV vaccines, which are key elements in health communication that promote HPV vaccination. This study aimed to capture discussions on HPV vaccination on Twitter and to identify user characteristics, sentiments, discussed themes and their potential reach before and after 12 November 2021, when the Ministry of Health, Labour, and Welfare's decision to resume proactive recommendation for HPV vaccines was first reported by the media in Japan. This study conducted a content analysis of tweets related to HPV vaccines posted on 11 November and 12 November 2021. Tweets were coded in terms of user characteristics, sentiments (i.e. positive, negative and neutral) and themes. Reach was measured by the number of retweets, likes and followers. A total of 3623 tweets were identified. The results showed that approximately 50% of health professionals and researchers tweeted neutral content about HPV vaccines. The most frequently discussed theme was the safety and side effects of HPV vaccines, which was accompanied by mainly negative sentiments. Although health professionals and researchers are influential on Twitter, half of them tweeted neutrally about HPV vaccines. Influential professionals, such as health professionals and researchers, are expected to disseminate accurate information to correct misinformation and recommend HPV vaccination on Twitter to overcome the HPV vaccination crisis, which is characterized by the low vaccination rates.

Review Paper: Social Media Sentiment Analysis Using Twitter Dataset

Review Paper: Social Media Sentiment Analysis Using Twitter Dataset

Nutrition Information Post COVID-19: A twitter content analysis

Realizing social media's importance, many doctors, nutritionists, health coaches and general users have registered on social media and actively share health information. Users may easily access and exchange health information. It benefits both users and practitioners. Qualitative data analysis is employed to study Twitter communication content to understand better the relationship between users’ interest in healthy eating. The research examined Twitter nutrition health information using hashtags. The frequency of hashtags was ranked. The content analysis quantifies social media healthy diet hashtags. Theme modification and word and phrase recurrence analysis to identify two primary themes and significant sentiments relating to Covid-19 and nutrition. Python and NLP are used to analyze and interpret the data to help acquire in-depth information.

The Influence of Social Media on the Transparency and Accountability of the South African Public Sector

There is a growing debate on the potential role of social media in improving government transparency and accountability and, ultimately, policy-making. This article analyses the contribution of social media, specifically Twitter, in increasing government transparency and accountability in South Africa by transferring information between the government and the public. To achieve this aim, we conducted an in-depth analysis of Twitter content, focusing on a curated selection of randomly selected top hashtags that became prominent in the South African context. These hashtags represent notable trends such as #FeesMustFall, #ZumaMustFall, #NotInMyName, #AmINext, and the #ThaboBester saga. By examining the discourses surrounding these hashtags, we aimed to uncover insights into how social media interactions intersected with government actions and reactions. Our results indicate that these hashtags served as potent catalysts for public engagement, driving discussions that critiqued government decisions and compelled timely reactions from the government.

Hustle Culture in Social Media: Exploring the Imagined Success in the Modern Era

This study examines how the notion of success among modern Indonesian youth, particularly the millennial generation, is constructed through the hustle culture trend on social media. With social media playing a significant role in shaping the aspirations and lifestyles of young individuals, the study centers on the hustle culture trend due to its strong promotion of values such as hard work, ambition, and productivity as integral to their way of life. Data was collected from various platforms including Instagram, Twitter, and TikTok, comprising textual, visual, and multimedia content related to hustle culture and the modern urban youth lifestyle in Indonesia. Additionally, interviews were conducted with six millennial respondents. The collected data is analyzed using the Ethnography Content Analysis (ECA) approach while reflecting on various theoretical concepts regarding youth, urban, social media, and hustle culture. The findings demonstrate that the hustle culture trend constructs an idealized image and lifestyle of modern urban youth, portraying them as modern, productive, and inspirational, thereby equating the social media image of success with actual success. This research provides a fresh perspective on understanding success among Indonesian youth in the modern era, specifically by examining the influence of social media practices in their daily l­­ives

Comparison of Turkish and English-Speaking ISIS Sympathizers’ Twitter Content between 2015 and 2016

Abstract The development of social media technologies has had a significant impact on individuals, organizations and societies. However, social media has not only affected people and communities, but also terrorist organizations have started to use social media platforms effectively. The Islamic State of Iraq and Syria (ISIS) is one such group that actively utilizes social media. Social media has played a critical role in formulating and disseminating ISIS’s strategies. Twitter is one of the most effective social media platforms used by ISIS, and it actively uses Twitter in predominantly Muslim countries such as Turkey. In this article, I use a dictionary-based method to compare and analyze 29,419 tweets from English-speaking supporters of ISIS and 40,526 tweets from Turkish supporters of ISIS between 2015 and 2016.

Neither Words nor Images: Health Departments' and Hospitals' Use of Twitter During COVID-19 Had Limited Racial Representation and Content on Disparities Decreased Over Time.

To determine the degree to which hospitals and state health departments used written content or visual representation on social media to draw attention to racial disparities during the COVID-19 pandemic. A retrospective analysis, using Twitter content (words and images) between May-June 2020 and May-June 2021 from organizations in the 5 states with the largest documented racial disparities in COVID-19-related mortality. All tweets and retweets (n = 6790) were coded along several lines. For May-June 2020 and May-June 2021, posts were coded as pandemic related (yes/no) and disparities related (yes/no). Open-coding methods categorized pandemic-related content into content areas, including COVID-19 education, hospital or public policy, and addressing misinformation. After self-identifying their own race/ethnicity, survey respondents (n = 100) coded pandemic-related Twitter images (n = 198) as including individuals of a similar race/ethnicity (yes/no). In May-June 2020, health departments posted more pandemic-related content than hospitals (μ = 204 and 71 tweets, respectively; P = .03), including more about health disparities (μ = 14.3% and 2.11% of tweets, respectively; P = .03). Between May-June 2020 and May-June 2021, content addressing health disparities decreased for both groups (47% decrease for health departments and 69% decrease for hospitals). Black respondents were more likely to feel represented in images from health departments than in those from hospitals (44.3% and 23.7% of images, respectively; P = .05). Both hospitals and health departments were more likely to use images where White respondents felt represented (hospitals = 76.1% of images; health departments = 59.7%) than images where respondents from racial/ethnic minorities felt represented (hospitals = 19.3% of images; health departments = 21.4%) ( P ≤ .001 for hospitals; P = .004 for health departments). Health education ideally comes in a variety of ways. Hospitals used social media for this purpose less than health departments, and neither group increased such content during the COVID-19 pandemic even as evidence of racial disparities grew.