Twitter Data Research Articles

Twitter Sentiment Analysis Using NLP

Abstract—Social media platforms, particularly Twitter, have become pivotal sources of public opinion and sentiment expression. The analysis of these sentiments has significant applications across various domains, including marketing, politics, and public health. This paper presents a comprehensive review and implementation of Natural Language Processing (NLP) techniques for sentiment analysis of Twitter data. We explore various preprocessing methods, feature extraction techniques, and machine learning algorithms specifically optimized for the unique characteristics of Twitter content. Our implementation demonstrates a pipeline that handles the challenges of Twitter data, including abbreviated language, emoticons, hashtags, and context-specific jargon. Using a large-scale dataset of 1.8 million tweets from Kaggle, our hybrid approach combining traditional machine learning methods with deep learning techniques achieves superior performance with an accuracy of 87.6%, an F1-score of 0.862, and significantly improved handling of negation and sarcasm compared to baseline methods. The analysis further reveals important insights into the temporal and contextual nature of sentiment expression on Twitter and suggests promising directions for future research in this domain.) Key Words:. sentiment analysis, natural language processing, Twitter, social media analytics, machine learning, deep learning, text classification

Smart Decisions with Opinion Mining

Abstract: The runaway growth of web technology has resulted in an unprecedented volume of data being produced and published on the web each day. Social networking sites such as Twitter and Facebook have turned into indispensable zones for individuals to share thoughts, experiences, and opinions around the world. Sentiment analysis which involves the extraction and analysis of opinion from text, is central to gauging public feeling, monitoring trends, business strategy, and customer satisfaction with regards to unstructured and heterogeneous nature of Twitter data, most research has been conducted on how to use sentiment analysis methods to classify opinion as positive, negative, or neutral. In this paper, sentiment analysis of social media data is investigated based on a Twitter dataset, utilizing machine learning methods such as Long Short-Term Memory (LSTM) networks for precise sentiment classification.

Geosocial media’s perspective on energy: a text classification approach using natural language processing

ABSTRACT This study examines public opinion on various energy sources through Twitter data, focusing on fossil fuels, nuclear energy, and renewable energy sources like solar and wind. Utilizing natural language processing techniques, specifically BERTweet and GPT models, the research analyses tweet categorization based on sentiment and stance related to these energy sources. Our findings reveal a positive shift towards nuclear, solar, and wind energy, contrasting with increasing negative sentiment towards fossil fuels. Notably, BERTweet demonstrates superior precision and recall in tweet categorization compared to GPT-3.5 and GPT-4, which show potential bias against fossil fuels, misclassifying many tweets as opposing them. This study highlights the importance of social media analytics in understanding public opinions and shaping energy policy, suggesting that future research should broaden the scope of data, enhance multilingual capabilities, and improve data visualization to more accurately reflect global public opinion. The results underscore the need for balanced AI training to mitigate bias and more accurately capture diverse perspectives on contentious energy topics. The datasets, code utilized, and interactive maps with word clouds are available at https://doi.org/10.5281/zenodo.15020578 and https://doi.org/10.5281/zenodo.15084294.

Crowdsourcing dust storms in the United States utilizing social media data

Dust storms and other dust events are natural phenomena characterized by strong winds carrying large amounts of fine particles which have significant environmental and human impacts. However, capturing the occurrence of such phenomena is a challenge. Previous studies have limitations due to available data, especially regarding short-lived, intense dust storms and events that are not captured by observing stations and satellite instruments. In recent years, the advent of social media platforms has provided a unique opportunity to access vast amounts of crowdsourced data. This paper explores the utilization of Flickr and X (Twitter) data to study dust event occurrences within the United States and their correlation with National Weather Service (NWS) advisories. The work ascertains the reliability of using crowdsourced data as a supplementary source for dust events monitoring. Our analysis of Flickr and X indicates that the Southwest region is most susceptible to dust events, with Arizona leading in the highest number of occurrences. On the other hand, the Great Plains show a scarcity of crowdsourced data related to dust events, which can be attributed to the sparsely populated nature of the region. Furthermore, seasonal analysis reveals that dust events are prevalent during the Summer months followed by Spring. These results are consistent with previous traditional studies that did not use social media of dust occurrences in the U.S., and Flickr-identified images of dust events show substantial co-occurrence with regions of NWS dust warnings. This paper highlights the potential of using crowdsourced data for the often overlooked field of dust monitoring that has substantial health and economic impacts.

ED-Filter: dynamic feature filtering for eating disorder classification

Eating disorders (ED) are critical psychiatric problems that have alarmed the mental health community. Mental health professionals are increasingly recognizing the utility of data derived from social media platforms such as Twitter. However, high dimensionality and extensive feature sets of Twitter data present remarkable challenges for ED classification. To overcome these hurdles, we introduce a novel method, an informed branch and bound search technique known as ED-Filter. This strategy significantly improves the drawbacks of conventional feature selection algorithms such as filters and wrappers. ED-Filter iteratively identifies an optimal set of promising features that maximize the eating disorder classification accuracy. In order to adapt to the dynamic nature of Twitter ED data, we enhance the ED-Filter with a hybrid greedy-based deep learning algorithm. This algorithm swiftly identifies sub-optimal features to accommodate the ever-evolving data landscape. Experimental results on Twitter eating disorder data affirm the effectiveness and efficiency of ED-Filter. The method demonstrates significant improvements in classification accuracy and proves its value in eating disorder detection on social media platforms.

Beyond the GenAI gold rush: Mapping entrepreneurial tensions in the age of generative artificial intelligence

AbstractAs generative artificial intelligence (GenAI) emerges as the latest technological gold rush, this preliminary study examines how entrepreneurs made sense of this transformative moment in the early months of ChatGPT. Through semi‐structured interviews with entrepreneurs adopting GenAI, observations of AI creators in Silicon Valley, and computational analysis of our qualitative data, as well as Twitter discourse, we reveal a complex reality beneath the media hype. While Twitter discourse initially showed entrepreneurs aligning with techno‐optimistic expectations of their imagined audience, our ethnographic data exposed deeper tensions around job displacement, bias, privacy, alignment, and the more significant ethical and societal implications. The contrast between publicly oriented Twitter data and self‐reflective ethnographic data exemplifies the complex ways that entrepreneurs wrestle with their identity as innovators—a tension between public displays of optimism and internal grappling with uncertainty around AI. Drawing on Madsen et al.’s concept of productive friction, we employ this framework in two complementary ways: as an analytical lens for understanding entrepreneurs’ cognitive dissonance and as a methodological principle that deliberately generates tensions through the use of traditional and computational approaches. We argue that these contradictions serve as constructive catalysts for more thoughtful innovation and adoption of Gen AI by forcing critical examination of assumptions and practices. Our mixed‐methods approach, therefore, demonstrates a way to capture, analyze, and understand emerging productive frictions surrounding implications for entrepreneurship while contributing to emerging techno‐anthropological discourse and approaches for studying technological transformation.

Weighted Trilateration Using Domain Knowledge for Event Localization in Twitter Data

In this paper, we develop a novel approach called Domain Adapted Weighted Trilateration (DAWT). We use the idea of there are reference coordinates at known locations that scrape the microblog (tweet) counts in time and space (circular regions around the reference coordinate). The change in counts of tweets would be indicative of an event pattern. We propose the use of DAWT to combine the information from multiple reference points to find the exact location of the event. We propse the use of domain knowledge in the form of counts of tweet or spatial relationship between reference points to calculate the weights for trilateration. We use microblogging data collected from Twitter to evaluate our model and compare it with other baseline methods.

Sentiment Analysis of COVID-19 Booster Vaccines on Twitter Using Multi-Class Support Vector Machine

As part of its response to combat COVID-19, the Indonesian government has implemented a booster vaccination program. This policy has sparked various public responses, particularly across the Twitter platform. This research examines public sentiment regarding booster vaccines by analysing Twitter data through the Support Vector Machine (SVM) algorithm. The research utilises sentiment analysis, a text mining and processing technique, to represent and interpret text-based data. By examining the sentiment expressed in tweets, the study seeks better to understand the public discourse on booster shots on Twitter. The study also conducts a multi-class parameter assessment of SVM, combining One-against-one and One-against-rest approaches with various kernels (Sigmoid, Polynomial, and RBF) to obtain optimal results. The highest accuracy rate of 96% is achieved using the One Against One method combined with the RBF kernel. This is closely followed by implementations using the Polynomial kernel at 95.2% and the Sigmoid kernel at 93.7%. When employing the One Against Rest method, the RBF kernel demonstrates superior performance with 95.5% accuracy compared to both Polynomial and Sigmoid kernels. Based on these evaluation results, it is evident that integrating the One Against One approach with the RBF kernel delivers the most optimal accuracy among all tested combinations. The sentiment class distribution in the optimal model classifies 49 tweets as positive, 927 as neutral, and 24 as negative.

Ensemble Machine Learning Models Based on Predictions for Sentimental Analysis on Twitter Data

In current days, web content comes from social media, multiple companies, different types of events, online products and personal data. This sentiment analysis predicts findings with the help of different methodologies. We used machine learning models for this research. In this process, the input is so simple, but deriving this information is too difficult. Internet data usage is increasing throughout the world, using this data is used for feedback purposes. Such a type of data classification and organize was most difficult for sentiments. This feedback is most important for improving the business, gaining more profit and understanding the customer’s interest. Finally, from our research, Logistic regression accuracy is 92%, XGBoost accuracy is 90%, Decision trees predict 90% accuracy, and Random forests predict 95.5% accuracy. Compared to the ensemble learning model, the Random Forest Tree model achieves a higher accuracy rate than the ensemble models.

Access to credit and fintech: A lexicon-based sentiment analysis application on Twitter data

Access to credit and fintech: A lexicon-based sentiment analysis application on Twitter data

Twitter Data and Machine Learning?Based Prediction of Depression Levels

Millions of individuals around the world, suffer from depression, however, due to the stigma surrounding mental health and ineffective early detection mechanisms, depression often goes unnoticed until it escalates into catastrophic proportions. Especially Twitter, as one of the largest social media platforms, Twitter posts are an excellent source of usergenerated content that reflects the emotional/psychological state of the user from which it derives. Abstract: We obtain Twitter data and then conduct machine learning prediction on the status of the user according to his tweets. An overview of the methodology is presented, which involves data collection, data preprocessing, sentiment analysis, and feature extraction, while textual patterns, linguistic features, and sentiment polarity are considered as feature extraction methods. For classification, machine learning models (Support Vector Machines (SVM), Random Forest) or deep learning-based approaches (Long ShortTerm Memory (LSTM) networks or Bidirectional Encoder Representations from Transformers (BERT)) are used. The suggested model analyses tweets thinking about whether someone who tweets has a tendency towards depression and somewhat extra accurately classifies delivery tweets as mild, proxy, or excessive functionality for melancholy. Various Feature implementations (NLP approaches to be specific) to boost up the model performance. AbstractExperimental Results hold that machine learning methods even though deep learning methods particularly BERT show very high accuracy in depressive tendency detection. The study shows that social media data can be a useful source for monitoring mental health, providing important information about early warning of depression. These findings provide an avenue for exploring automated depression detection on Twitter as a tool to support healthcare professionals in intervening promptly to treat people experiencing mental health decline and the disease which bewitches it. Integration of multimodal data sources and real-time monitoring to enhance accuracy and reliability are other directions to explore for future work. The findings highlight the promise of AI-based solutions for mental health diagnosis, laying the foundation for new and potentially more effective digital mental health aids

Exploring Gender Disparities in Experiences of Being Hacked Using Twitter Data: A Focus on the Third-Level Digital Divide

Despite millions of hacked accounts fueling cybercrime, research on the hacking experience, particularly sociodemographic aspects, remains sparse. This study examines the experience of being hacked with a focus on gender disparities from the perspective of the third-level digital divide—socially constructed gaps of digital use outcomes even among users with similar digital access and skills. Analyzing 13,731 Twitter mentions of accounts being hacked, using topic modeling and classifying the gender of 12,586 users, we showed that women reported more experiences of being hacked across all types of online services except gaming. Women were more likely to experience negative consequences of being hacked, including reputational harm, money loss, and having personalized content modified. Gender differences were also found in coping strategies. Men were more likely to use active strategies like warning others, rebuilding accounts, and deducing hackers’ origins, while women were more likely to seek help from others to recover or report experiencing hacked accounts. The findings of this study imply the need for further research into the gendered experiences of being hacked from the third-level digital divide perspective, alongside the development of interventions to mitigate harm and empower users with diverse needs to cope with being hacked.

Hybrid model of machine and deep learning to analyze Twitter data and prediction of online public opinion: revisiting agenda-setting implications

Hybrid model of machine and deep learning to analyze Twitter data and prediction of online public opinion: revisiting agenda-setting implications

Echoes of Privacy: Deciphering Consumer Voices in the Data Deluge

ABSTRACTAs technology evolves, balancing product and service personalization with privacy protection remains a critical challenge for businesses and consumers. So, researchers have raised concerns about the potential negative effects of data‐driven marketing. This study embarks on a journey to unravel the intricacies of consumer attitudes toward privacy within data‐centric marketing. Our research design is underpinned by a mixed methodology that uses qualitative and quantitative research methods. Initially, we presented text mining of Iranian tweets on X (Twitter). Subsequently, we expounded upon the intricacies of the structural equation model and the clustering process. The research findings highlight cultural disparities in attitudes toward privacy and collectivist countries such as Iran prioritize privacy vigilance, as they are meticulous about information disclosure. Notably, Iranians of varying genders, ages, and educational backgrounds displayed distinct levels of privacy apprehension. Based on the text of Twitter data mining, data confidentiality stands as the paramount pillar of data privacy, and they have expressed dissatisfaction with things like “Leakage of information and sale of users' data” and “Hacking the data of an online business”. Also, four clusters were identified for privacy in data‐driven marketing: (1) Privacy Balancers, (2) Privacy Guardians, (3) Privacy Nomads, and (4) Privacy Advocates. Finally, suggestions were presented to businesses and policymakers. Especially, businesses can use gamification techniques such as interactive challenges to engage customers in reading and understanding privacy policies.

Analysing the Twitter social media network in voicing opinions on money loundering issue in Indonesia

This study aims to examine the Twitter social media network concerning the money laundering issue involving the Indonesian Ministry of Finance. The primary goal is identifying and highlighting the individuals or entities significantly influencing online discussions. Further, the study looks at the intricacies of public perception and communication about this pressing issue within the Ministry. The study gathered and analyzed Twitter data using social network analysis methodologies, consisting of 6742 nodes and 744 edges. The findings demonstrate ten actors exhibit the most prominent clusters, whereas five exhibit the maximum links. The significance of the Twitter user was @seeksixsuck, who highlights the importance of religious people in ongoing conversations. This research study enhances the social media environment's comprehension of money laundering within the Ministry of Finance, providing insights into the interconnected relationships and individuals influencing the ongoing conversation. The study's findings provide significant value in facilitating public engagement initiatives and promoting a more knowledgeable and engaged population in the ongoing battle against money laundering.

AI-Driven Sentiment Analysis of Retail Investor Behavior during Market Volatility: A Study of Twitter Data in Southeast Asia

In recent years, retail investor participation in Southeast Asian capital markets has surged, contributing to increased market volatility and making sentiment analysis a critical factor in understanding price dynamics. This study investigates the relationship between social media sentiment and stock market fluctuations by focusing on Twitter data during periods of market volatility in Indonesia, Thailand, and Malaysia. The objective is to examine how collective investor emotions, as expressed through social media, correlate with daily stock index movements. Employing an exploratory quantitative approach, the study integrates Natural Language Processing (NLP) methods, both lexicon-based tools such as VADER and advanced transformer-based models like BERT and GPT, to classify over 150,000 tweets into positive, negative, and neutral sentiments. Sentiment scores were then aggregated and statistically tested using Pearson correlation with daily stock index returns, specifically the IDX Composite, SET Index, and FTSE Bursa Malaysia. The findings reveal a significant negative correlation between negative sentiment and market returns, particularly in the IDX Composite (r = -0.61, p < 0.05), indicating that pessimistic sentiment is associated with market downturns. Thailand’s SET Index and Malaysia’s FTSE Index showed moderate to weak negative correlations, with r = -0.43 and r = -0.27, respectively. These results highlight the sensitivity of emerging markets to emotionally driven retail behavior. The study concludes that AI-based sentiment analysis offers a valuable early warning tool for market volatility and can complement traditional financial indicators. It recommends developing AI-based sentiment dashboards and enhancing digital financial literacy to mitigate emotional reactivity among retail investors.

Gender differences in tweets on postnatal depression: A corpus linguistic analysis.

Gender differences have been found in the way parents communicate online, however it is unclear whether these differences apply in the context of postnatal depression (PND). This research aimed to evaluate online discourses surrounding PND and explore gender differences in communication style associated with PND. X (formerly Twitter) data (15,850 posts) was identified and collected based on a key term search (e.g. 'PND') and analysed using corpus linguistic analysis. Results showed that female X users were more likely to discuss PND using words with a negative connotation or to use self-referent items, compared to male users who discussed PND more generally. X content related to PND was mostly created by female users and generally revolved around the experiences of mothers. The limited discussion regarding paternal PND suggests a lack of acknowledgement and insufficient online resources available for fathers.

Playing dead in natural disasters: analysing customer reactions to the perceived silence of Starbucks on the earthquake disaster in Türkiye

Purpose This study aims to investigate the antecedents, consequences and severity of brand hate towards Starbucks, which was boycotted on social media due to its post-earthquake behaviour, including its “perceived silence”, in other words, playing dead, by delaying the publication of a condolence message and making an allegedly insufficient donation following the Kahramanmaras earthquakes in Türkiye. Design/methodology/approach Based on Küçük’s (2019) hierarchical model for brand hate, the severity of the brand hate attracted by Starbucks is differentiated. The antecedents and the consequences of brand hate in the aforementioned boycott period are investigated by conducting a thematic analysis of tweets collected from Twitter. Findings Hot hate was found to be the most frequently observed severity of hate towards Starbucks during the boycott period. Besides, brand betrayal (56.83%, n = 262 of 456 tweets) and insincerity (34.49%, n = 159 of 456 tweets) were prominent antecedents of brand hate and were found to be specific triggers of hot hate and cold hate, respectively. This study determined that negative word of mouth (20.54%, n = 205 of 510 tweets) and brand avoidance (11.92%, n = 119 of 510 tweets) were remarkable consequences of brand hate towards Starbucks and were also found to be extensions of hot brand hate. Originality/value This study makes a theoretical contribution to the literature by demonstrating how severity of hate functions as both a consequence and a trigger of brand hate. In addition, it expands the brand hate theory by showing that brand questioning can be a consequence of attitudinal brand hate, ethnocentric tendencies can trigger brand switching behaviour and symbolic incongruity can be associated with both cold hate and cool hate. In addition, by analysing real-time Twitter data, the study reveals how brand hate intensifies in crisis contexts and showcases social media as a powerful tool for understanding consumer backlash.

Quantitative Sentiment Analysis of Women's Safety Using Twitter Data: An NLP Approach

In today’s world of digitalization, social media platforms like Twitter serve as important outlets where people express their concerns and opinions on various societal issues including safety. Women's safety remains a significant concern, particularly in urban areas where incidents are often discussed publicly on social platforms. Despite the growing availability of such data there is a gap between analyzing social media sentiment and applying it to understand real-world safety issues. The issue lies in the underutilization of social media analytics for meaningful insights into public perceptions and potential policy interventions regarding women's safety. Twitter had 335.7 million monthly active users (MAU) in 2024, down 5.14% from 2023. There were 368.4 million users in 2022, the largest amount to date. Between 2018 and 2021, approximately 2% of all tweets originating from India contained elements of misogynistic language or sentiment. This research addresses that gap by conducting a comprehensive sentiment analysis of tweets related to women's safety identifying public sentiment patterns and highlighting the critical areas that require attention. Using NLP to process the data and to classify sentiments, this research paper will provide actionable insights. The paper is organized into several sections including a literature review, methodology, results and implications of Women Safety. Each section is designed to provide a comprehensive understanding of the research process from data collection to the implementation of the code and the interpretation of results. The findings from this analysis can support the development of real-time safety monitoring systems and addressing women’s safety concerns more effectively.

Ukrainian linguistic creativity and decolonial resistance: a Twitter analysis

In this study, I conduct an analysis of Ukrainian-language tweets from 2022, the year of Russia’s full-scale invasion, using the statistical software R. After a brief overview of the linguistic history of Ukraine and how it has been shaped by Russia, resulting in widespread Ukrainian-Russian bilingualism, I discuss the challenges of working with Ukrainian Twitter data and compiling a Twitter corpus concerned with Russia’s war in Ukraine. Subsequently, I conduct the analysis and visualize the results. The analysis reveals that, since Russia’s invasion on February 24, 2022, Ukrainians have employed various means, such as nicknames and slurs for members of the aggressor state, decapitalization of proper nouns, the creative use of prepositions, and creative orthography and ‘Mock Russian.’ Sentiment analysis suggests that Ukrainian Twitter users respond positively to certain war-related events, demonstrating confidence and the will to resist the aggressor. Furthermore, the study demonstrates that linguistic questions are relevant for Ukrainian society at war, showing that the analysis of linguistic behavior on social media helps to shed light on how Ukrainians cope with Russia’s war of aggression and how they use social media as a tool for decolonial resistance.