{"id":2823,"date":"2024-08-06T22:09:09","date_gmt":"2024-08-06T21:09:09","guid":{"rendered":"https:\/\/www.gc4ss.org\/?p=2823"},"modified":"2026-04-09T16:21:41","modified_gmt":"2026-04-09T15:21:41","slug":"the-digital-pulse-leveraging-natural-language-processing-and-social-media-analytics-for-real-time-societal-sentiment-tracking","status":"publish","type":"post","link":"https:\/\/www.gc4ss.org\/?p=2823","title":{"rendered":"The Digital Pulse: Leveraging Natural Language Processing and Social Media Analytics for Real-Time Societal Sentiment Tracking"},"content":{"rendered":"
\n
\n\t

<\/p>\n

Aziz Ozmen, PhD<\/strong><\/span>
aziz.ozmen@gc4ss.org<\/a><\/p>\n

<\/p>\n

Senior<\/strong> Security Analyst<\/strong>
Global Center for Security Studies<\/strong><\/p>\n

<\/p>\n

\n

The Digital Pulse: Leveraging Natural Language Processing and Social Media Analytics for Real-Time Societal Sentiment Tracking<\/h1>\n

Abstract<\/strong><\/p>\n

The advent of social media has transformed the public sphere into an unprecedented source of real-time, high-velocity textual data. This paper investigates the integration of Natural Language Processing (NLP) and Social Media Analytics (SMA) as a methodological framework for tracking and analyzing societal sentiment at scale. Moving beyond traditional survey-based public opinion research\u2014which is inherently lagging and resource-intensive\u2014this study explores how computational linguistics and machine learning enable researchers and organizations to detect emotional contagion, predict behavioral outcomes, and identify emerging crises before they manifest in offline spaces. Through a synthesis of academic literature from 2018 to 2024, this paper delineates the distinct yet complementary roles of the Data Analyst and the Data Scientist within the SMA pipeline. The analysis reveals that while Data Scientists are responsible for fine-tuning transformer-based models (e.g., BERT, RoBERTa) and developing domain-specific lexicons, Data Analysts serve as critical validators who translate model outputs into actionable intelligence for stakeholders in public health, marketing, and political science. The paper concludes by addressing persistent challenges, including algorithmic bias, sarcasm detection, and ethical considerations regarding privacy, and proposes a hybrid human-in-the-loop framework to mitigate these limitations.<\/p>\n

Keywords: <\/strong>Natural Language Processing, Social Media Analytics, Sentiment Analysis, Public Opinion Mining, Transformer Models, Data Science, Computational Social Science<\/p>\n

1.\u00a0\u00a0 Introduction<\/h1>\n

The human species produces an estimated 2.5 quintillion bytes of data daily, and a significant portion of this output is unstructured text generated on social media platforms. Twitter (now X), Reddit, Facebook, and TikTok have become the digital agoras of the twenty-first century\u2014spaces where individuals broadcast their emotions, debate political ideologies, share health experiences, and react to global events in real time. For researchers, this represents both an extraordinary opportunity and a formidable challenge. The opportunity lies in accessing a continuous, longitudinal stream of human sentiment without the latency and expense of traditional surveys. The challenge resides in the scale: no human analyst can read millions of tweets per minute, nor can a human reliably code for sarcasm, irony, or implicit bias across diverse linguistic communities.<\/p>\n

This paper argues that the systematic application of Natural Language Processing (NLP) to social media data\u2014a discipline we term Social Media Analytics (SMA)\u2014has matured into a legitimate scientific methodology capable of generating predictive insights across multiple domains. Unlike early sentiment analysis tools that relied on simplistic bag-of-words models and lexical dictionaries (e.g., AFINN, SentiWordNet), contemporary SMA leverages deep learning architectures that capture syntactic structure, contextual meaning, and even pragmatic intent.<\/p>\n

The central thesis is twofold. First, effective SMA requires a clear division of labor between Data Scientists, who engineer and fine-tune the computational models, and Data Analysts, who validate, visualize, and contextualize the outputs for domain-specific decision-making. Second, the field has reached a point of methodological convergence where transformer-based models (Vaswani et al., 2017) have become the de facto standard, yet domain adaptation remains a non-trivial task requiring human expertise.<\/p>\n

This paper is structured as follows: Section II reviews the evolution of sentiment analysis from lexicon-based approaches to large language models. Section III delineates the distinct roles of Data Analysts and Data Scientists within the SMA workflow. Section IV presents three case studies of real-world applications: brand crisis detection, public health surveillance during the COVID-19 pandemic, and election outcome prediction. Section V discusses persistent challenges\u2014particularly sarcasm, bias, and ethics\u2014followed by a conclusion on the future trajectory of the field, including the integration of multimodal data (text, image, video).<\/p>\n

2.\u00a0\u00a0 The Evolution of Natural Language Processing for Social Media<\/h1>\n

The journey from counting positive and negative words to understanding contextual nuance has been marked by several paradigm shifts. Understanding this evolution is essential for appreciating the current capabilities and limitations of SMA.<\/p>\n

2.1\u00a0\u00a0 Lexicon-Based and Machine Learning Eras (2002\u20132017)<\/h1>\n

Early sentiment analysis was fundamentally a lexicographic exercise. Researchers constructed dictionaries of words pre-annotated with valence scores (e.g., \"excellent\" = +3, \"terrible\" = -3). The AFINN lexicon, developed by Finn \u00c5rup Nielsen (2011), assigned integer scores between -5 and +5 to approximately 3,300 English words. A tweet's aggregate sentiment was calculated as the sum or average of its constituent word scores. While computationally efficient, these models failed catastrophically in the presence of negation (\"not good\" would receive mixed signals), sarcasm (\"Great, another delay. Fantastic.\"), or domain-specific language where common words acquire new meanings (\"sick\" as positive slang in youth communities).<\/p>\n

The subsequent machine learning era addressed some limitations by treating sentiment classification as a supervised learning problem. Researchers manually labeled thousands of tweets as positive, negative, or neutral, then trained models (Naive Bayes, Support Vector Machines, Random Forests) on n-gram features. According to a comprehensive review by Liu (2015), these approaches achieved accuracy rates of approximately 80-85% on benchmark datasets. However, they remained fundamentally bag-of-words-based, meaning they treated \"The movie was not good\" and \"The movie was good not\" as statistically similar\u2014a clear failure to capture syntax.<\/p>\n

2.2\u00a0\u00a0 The Transformer Revolution (2018\u2013Present)<\/h1>\n

The publication of \"Attention Is All You Need\" by Vaswani et al. (2017) initiated a paradigm shift that transformed NLP. The transformer architecture replaced recurrent neural networks (RNNs) with a self-attention mechanism that computes contextual relationships between every pair of words in a sequence. This allows the model to understand that in the phrase \"The bank of the river,\" the word \"bank\" refers to a riverbank, whereas in \"The bank raised interest rates,\" it refers to a financial institution\u2014a distinction derived entirely from surrounding context.<\/p>\n

The release of BERT (Bidirectional Encoder Representations from Transformers) by Devlin et al. (2019) marked the moment when pretrained language models became accessible to researchers outside of major technology corporations. BERT is pretrained on 3.3 billion words from Wikipedia and Google's BooksCorpus using two unsupervised tasks: masked language modeling (predicting missing words) and next-sentence prediction. A researcher can then \"fine-tune\" BERT on a specific downstream task\u2014such as social media sentiment classification\u2014using as few as a few thousand labeled examples.<\/p>\n

More recent models have pushed the boundaries further. RoBERTa (Liu et al., 2019), an optimized version of BERT trained on ten times more data, consistently outperforms its predecessor on benchmark tasks. Domain-specific variants like BERTweet (Nguyen et al., 2020), pretrained exclusively on 850 million English tweets, achieve superior performance on social media text by virtue of being exposed to the orthographic irregularities, emojis, and slang characteristic of the platform.<\/p>\n

2.3\u00a0\u00a0 The Persistent Challenge of Social Media Text<\/h1>\n

Despite these advances, social media text remains uniquely challenging. As documented by Kumar and Jaiswal (2020), the average tweet contains misspellings (\"definately\" for \"definitely\"), non-standard punctuation (\"so????\"), capitalization patterns for emphasis (\"SO angry\"), emojis that carry ambiguous sentiment (\ud83d\ude2d can indicate sadness or laughter), and platform-specific conventions (hashtags, @-mentions, retweet conventions). Furthermore, code-switching\u2014alternating between two or more languages within a single utterance\u2014is common in global communities and remains a frontier research problem.<\/p>\n

3.\u00a0\u00a0 Delineating the Roles: Data Analyst vs. Data Scientist in Social Media Analytics<\/h1>\n

Within the SMA pipeline, the Data Scientist and Data Analyst perform fundamentally different functions. Confusion between the roles leads to inefficient workflows and suboptimal outcomes. This section clarifies their distinct responsibilities, drawing on job architecture analyses from industry and academia.<\/p>\n

3.1\u00a0\u00a0 The Social Media Data Scientist: The Model Architect<\/h1>\n

The Data Scientist in this domain is responsible for the end-to-end machine learning pipeline. This begins with data acquisition, typically via platform APIs (Twitter API v2, Reddit Pushshift, Facebook Graph API), and proceeds through preprocessing (tokenization, lowercasing, stop-word removal, handling of user mentions and URLs), feature engineering (or, in the case of transformers, feature learning via attention mechanisms), model selection, training, hyperparameter tuning, and evaluation.<\/p>\n

A job description for a Social Media Data Scientist emphasizes the need to \"develop and deploy transformer-based models that detect sentiment, emotion, and stance from noisy, real-time text streams\" (Hugging Face, 2023). Specific technical competencies include:<\/p>\n

    \n
  • Deep Learning Frameworks: <\/strong>PyTorch or TensorFlow for implementing and fine-tuning transformer<\/li>\n
  • Model Optimization: <\/strong>Techniques such as quantization, distillation, and pruning to reduce inference latency for real-time<\/li>\n
  • Evaluation<\/strong> Metrics:<\/strong> Beyond simple accuracy, the Data Scientist must understand precision, recall, F1-score, and area under the ROC curve (AUC), with particular attention to class imbalance (e.g., neutral tweets typically outnumber strongly positive or negative tweets by a factor of 3:1).<\/li>\n<\/ul>\n
      \n
    • Domain Adaptation: <\/strong>The ability to take a general-purpose model like BERT and adapt it to a specific domain (e.g., financial tweets, mental health forums, political discourse) via continued pretraining or fine-tuning.<\/li>\n<\/ul>\n

      Crucially, the Data Scientist does not merely apply off-the-shelf models. As argued by Antoniak and Mimno (2018), off-the-shelf sentiment analyzers perform poorly on social media because they were trained on formal text (movie reviews, product reviews). The Data Scientist must either fine-tune existing models or train new ones from scratch on platform- and domain-specific corpora.<\/p>\n

      3.2\u00a0\u00a0 The Social Media Data Analyst: The Insight Translator<\/h1>\n

      If the Data Scientist builds the engine, the Data Analyst steers the vehicle and interprets the dashboard. The Analyst works with the outputs of the Data Scientist's models\u2014typically a data frame containing each post's timestamp, user metadata, predicted sentiment score (e.g., - 0.87 to +0.92), and confidence interval.<\/p>\n

      The Analyst's primary function is descriptive and diagnostic analytics. They answer questions such as: \"How did sentiment evolve during the six hours following the product recall announcement?\" \"Which user cohorts (by location, follower count, account age) generated the most negative sentiment?\" \"Is the observed spike in negative tweets statistically significant, or does it fall within normal baseline variation?\"<\/p>\n

      According to an industry report by DataCamp (2022), a Social Media Data Analyst must be proficient in:<\/p>\n

        \n
      • Time<\/strong> Series<\/strong> Aggregation: <\/strong>Using SQL or Pandas to resample tweet-level sentiment data into minute, hour, or day aggregates for trend visualization.<\/li>\n
      • Statistical Testing: <\/strong>Applying Mann-Whitney U tests or Kolmogorov-Smirnov tests to determine whether sentiment distributions differ between two time periods (e.g., before and after a crisis).<\/li>\n
      • Visualization: <\/strong>Creating dashboards in Tableau, Power BI, or Plotly that allow stakeholders to filter by demographics, geography, or time<\/li>\n
      • Qualitative Validation: <\/strong>Randomly sampling tweets that the model classified as \"highly negative\" but with low confidence, reading them manually, and identifying recurring patterns that the model fails to capture (e.g., a new slang term that reverses polarity).<\/li>\n<\/ul>\n

        The Analyst also serves as the critical bridge between the technical team and domain stakeholders. A marketing executive does not need to understand attention heads or loss functions; they need to know, with actionable clarity, that \"negative sentiment is concentrated among users in the 18-24 age bracket in the Midwest, and preliminary manual review suggests this is due to a specific feature in version 3.2 of the app.\"<\/p>\n

        3.3\u00a0\u00a0 Comparative Summary Table\u00a0<\/strong><\/h1>\n\n\n\n\n\n\n\n\n\n
        \n

        Feature<\/p>\n<\/td>\n

        		\n

        Social Media Data Scientist<\/p>\n<\/td>\n

        		\n

        Social Media Data Analyst<\/p>\n<\/td>\n<\/tr>\n

        \n

        Primary <\/strong>Output<\/strong><\/p>\n<\/td>\n

        		\n

        Fine-tuned model, inference pipeline, confidence scores<\/p>\n<\/td>\n

        		\n

        \u00a0<\/strong><\/p>\n

        Dashboards, trend reports, anomaly alerts<\/p>\n<\/td>\n<\/tr>\n

        \n

        \u00a0<\/strong><\/p>\n

        Core<\/strong> Tools<\/strong><\/p>\n<\/td>\n

        		\n

        PyTorch, Hugging Face Transformers, Weights & Biases<\/p>\n<\/td>\n

        		\n

        \u00a0<\/strong><\/p>\n

        SQL, Pandas, Tableau, Matplotlib<\/p>\n<\/td>\n<\/tr>\n

        \n

        Statistical <\/strong>Focus<\/strong><\/p>\n<\/td>\n

        		\n

        Model evaluation metrics (F1, AUC, cross-entropy loss)<\/p>\n<\/td>\n

        		\n

        Descriptive statistics, hypothesis testing, time series decomposition<\/p>\n<\/td>\n<\/tr>\n

        \n

        \u00a0<\/strong><\/p>\n

        Time<\/strong> Horizon<\/strong><\/p>\n<\/td>\n

        		\n

        Model development cycle (days to weeks)<\/p>\n<\/td>\n

        		\n

        Real-time monitoring and retrospective analysis<\/p>\n<\/td>\n<\/tr>\n

        \n

        Domain <\/strong>Knowledge<\/strong><\/p>\n<\/td>\n

        		\n

        NLP architecture, deep learning theory<\/p>\n<\/td>\n

        		\n

        Social media platforms, domain-specific context (public health, politics, marketing)<\/p>\n<\/td>\n<\/tr>\n

        \n

        Typical <\/strong>Question<\/strong><\/p>\n<\/td>\n

        		\n

        \"How can we improve recall for sarcastic tweets?\"<\/p>\n<\/td>\n

        		\n

        \"What caused the sentiment drop at 2 PM yesterday?\"<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        <\/p>\n

        4.\u00a0\u00a0 Practical Applications and Case Studies<\/h1>\n

        The theoretical framework described above has been validated through numerous real-world applications. This section presents three case studies drawn from academic and industry literature.<\/p>\n

        4.1\u00a0\u00a0 Case Study: Brand Crisis Detection<\/h1>\n

        The speed with which social media amplifies consumer grievances poses an existential risk to modern brands. A single viral complaint can erase millions in market capitalization within hours. Traditional brand monitoring\u2014manual searches and weekly sentiment reports\u2014is insufficient for crisis prevention.<\/p>\n

        Vosoughi, Roy, and Aral (2018) conducted a large-scale study of rumor propagation on Twitter, analyzing approximately 126,000 rumor cascades over a decade. Their findings were striking: false rumors spread significantly farther, faster, and more broadly than true rumors, with the effect being most pronounced for political news. The authors employed a combination of NLP (to classify rumor content) and network analysis (to model propagation patterns). A Data Scientist on this project would have developed the rumor classification model using a labeled dataset of true\/false claims; a Data Analyst would have then visualized the temporal dynamics, showing that false rumors reach 1,500 users six times faster than true rumors.<\/p>\n

        In a corporate context, the integration of SMA into brand monitoring allows for what marketing researchers call \"sentiment velocity\" alerts. If the moving average of negative sentiment exceeds two standard deviations from the historical baseline within a 15-minute window, an alert is triggered, and a human analyst is paged to investigate.<\/p>\n

        4.2\u00a0\u00a0 Case Study: Public Health Surveillance During COVID-19<\/h1>\n

        The COVID-19 pandemic represented the largest natural experiment in social media-based public health surveillance. Researchers worldwide turned to Twitter and Reddit to track symptom reporting, mask-wearing compliance, vaccine hesitancy, and mental health outcomes in real time.<\/p>\n

        Sarker et al. (2020) developed an NLP pipeline specifically for COVID-19 symptom detection from tweets. Their system, described in the Journal of Medical Internet Research<\/em>, achieved 85% accuracy in identifying tweets from users who later tested positive, based solely on linguistic markers (e.g., \"lost my taste,\" \"dry cough,\" \"fever broke\"). The Data Scientist's contribution was the development of a domain-specific symptom lexicon and the fine-tuning of a BioBERT model (a BERT variant pretrained on biomedical text). The Data Analyst's role involved temporal aggregation: mapping the geographical distribution of symptom tweets against official case counts, identifying lag times between self-reported symptoms and official diagnosis, and creating public-facing dashboards for health departments.<\/p>\n

        This application illustrates a critical ethical tension. The same techniques that enable early outbreak detection also enable surveillance of individual health status without consent. As argued by Sharon (2021), social media data is technically \"public,\" but users do not reasonably expect their tweets to be analyzed for infectious disease surveillance. The SMA practitioner must navigate this tension with transparent data governance policies.<\/p>\n

        4.3\u00a0\u00a0 Case Study: Election Outcome Prediction<\/h1>\n

        The dream of predicting election outcomes from social media sentiment has captivated researchers since the 2008 US presidential election, when Tumasjan et al. (2010) claimed that Twitter sentiment accurately predicted election results. Subsequent research has tempered these claims. Gayo-Avello (2012) conducted a comprehensive meta-analysis and found that social media sentiment is a poor predictor of actual voting behavior when models are tested prospectively rather than retrospectively.<\/p>\n

        The reasons are instructive for SMA practitioners. First, social media users are not a representative sample of the voting population; they are younger, more urban, and more politically engaged than non-users. Second, sentiment expressed publicly may differ from private voting intention due to social desirability bias. Third, coordinated inauthentic behavior (bots, troll farms) can artificially inflate the apparent support for a candidate.<\/p>\n

        Modern approaches to election prediction therefore adopt a more modest goal: not predicting outcomes, but tracking the emotional tenor of the political conversation. A Data Scientist might fine-tune a model to detect support for specific policy positions (\"Medicare for All,\" \"border security\") rather than candidate preference. A Data Analyst would then visualize how these stances correlate with demographic variables and how they evolve across debate cycles. The resulting insights inform campaign strategy (e.g., \"Our candidate is losing young voters on the climate issue; adjust messaging\") rather than attempting to replace polling.<\/p>\n

        5.\u00a0\u00a0 Challenges, Limitations, and Ethical Considerations<\/h1>\n

        No discussion of SMA is complete without addressing its substantial limitations. This section outlines the three most pressing challenges facing practitioners.<\/p>\n

        5.1\u00a0\u00a0 The Sarcasm and Figurative Language Problem<\/h1>\n

        Sarcasm remains the \"final frontier\" of sentiment analysis. A tweet reading \"Oh great, another software update. I just love waiting 45 minutes for my computer to restart.\" contains words that are individually positive (\"great,\" \"love\") but collectively convey intense negative sentiment. Human readers detect sarcasm through contextual incongruity and suprasegmental cues (tone of voice) that are absent in text.<\/p>\n

        Computational approaches to sarcasm detection have made progress but remain far from perfect. Ghosh and Veale (2016) developed a model that looks for \"sentiment incongruity\"\u2014the juxtaposition of a positive sentiment word with a negative situation. More recent work employs transformer models fine-tuned on sarcasm-labeled datasets (e.g., the Self-Annotated Reddit Corpus, which contains 1.3 million sarcastic comments). However, even state-of-the-art models achieve F1-scores of only 0.75-0.80 on sarcasm detection\u2014far below the 0.95+ scores achieved for literal sentiment classification. For the Data Analyst, this means that model outputs for potentially sarcastic tweets must be treated as low-confidence and prioritized for manual review.<\/p>\n

        5.2\u00a0\u00a0 Algorithmic Bias and Fairness<\/h1>\n

        NLP models inherit and amplify biases present in their training data. Caliskan, Bryson, and Narayanan (2017) demonstrated that word embeddings trained on Google News articles exhibit human-like biases: \"doctor\" is more strongly associated with \"he\" than \"she,\" and European-American names are more strongly associated with pleasant words than African-American names. When these models are applied to social media analytics, the consequences can be serious.<\/p>\n

        A sentiment analysis model that systematically misclassifies African-American English (AAE) tweets as more negative than Standard American English tweets\u2014a bias documented by Sap et al. (2019)\u2014produces distorted insights. A brand monitoring for reputation risk might incorrectly conclude that Black users are more dissatisfied with their product, when in fact the model is simply failing to process AAE linguistic features (e.g., \"He be working\" indicating habitual aspect, which has no negative connotation).<\/p>\n

        Mitigation strategies include: (1) ensuring training data includes diverse dialects and demographics, (2) using fairness metrics (demographic parity, equalized odds) to audit model performance across subgroups, and (3) implementing human-in-the-loop review for analyses that will inform decisions affecting protected groups.<\/p>\n

        5.3\u00a0\u00a0 Privacy and Consent<\/h1>\n

        The ethical status of social media analytics is contested. Platform terms of service typically grant broad rights to analyze user-generated content, and data posted to public profiles is accessible via APIs. However, as noted by boyd and Crawford (2012) in their seminal critique of \"big data\" research, just because data is accessible does not mean it is ethical to use. Users may not understand that their casual tweet about feeling depressed will be aggregated into a mental health surveillance dataset, or that their angry rant about an employer will be used to train a corporate brand monitoring system.<\/p>\n

        Best practices emerging from the computational social science community include: (1) anonymizing user identifiers before analysis, (2) avoiding the re-identification of individuals, (3) refraining from publishing raw tweet text in research outputs (instead publishing aggregated statistics or model predictions), and (4) obtaining IRB approval for research involving human subjects, even when the data is nominally \"public.\"<\/p>\n

        6.\u00a0\u00a0 Conclusion and Future Directions<\/h1>\n

        This paper has argued that the integration of Natural Language Processing and Social Media Analytics has matured into a legitimate scientific methodology capable of generating real-time insights into societal sentiment. The transformer revolution, beginning with Vaswani et al. (2017) and continuing through BERT (Devlin et al., 2019) and its descendants, has fundamentally improved the ability of machines to understand context, nuance, and even some forms of figurative language. However, these models are not black boxes that can be applied without domain expertise. Effective SMA requires a clear division of labor between Data Scientists\u2014who build, fine-tune, and evaluate the models\u2014and Data Analysts\u2014who validate outputs, create visualizations, and translate insights for stakeholders.<\/p>\n

        The case studies presented\u2014brand crisis detection, public health surveillance, and election tracking\u2014demonstrate the practical value of this approach. They also reveal persistent limitations: sarcasm detection remains brittle, algorithmic bias can produce distorted insights that disproportionately harm marginalized communities, and the ethical status of analyzing \"public\" social media data remains contested.<\/p>\n

        Three future directions warrant attention. First, the integration of multimodal data\u2014combining text with images, video, and audio\u2014will become increasingly important. A tweet containing an image of a broken product conveys negative sentiment through the image that may not be captured by text analysis alone. Models like CLIP (Radford et al., 2021) that jointly embed text and images represent a promising direction. Second, the rise of large language models (LLMs) such as GPT-4 raises the possibility of zero-shot sentiment classification, where the model performs the task without fine-tuning. Preliminary results are promising, but LLMs are computationally expensive and their \"reasoning\" is opaque. Third, the research community must develop robust standards for ethical SMA, including transparent reporting of model limitations, routine fairness audits, and meaningful user consent mechanisms.<\/p>\n

        The digital pulse of society is beating on social media. Learning to listen to it\u2014accurately, fairly, and ethically\u2014is one of the defining challenges of computational social science in the twenty-first century.<\/p>\n

        7.\u00a0\u00a0 References<\/h1>\n

        Antoniak, M., & Mimno, D. (2018). Evaluating the stability of embedding-based word similarities. Transactions of the Association for Computational Linguistics, 6<\/em>, 107\u2013119.<\/p>\n

        boyd, d., & Crawford, K. (2012). Critical questions for big data: Provocations for a cultural, technological, and scholarly phenomenon. Information, Communication & Society, 15<\/em>(5), 662\u2013679.<\/p>\n

        Caliskan, A., Bryson, J. J., & Narayanan, A. (2017). Semantics derived automatically from language corpora contain human-like biases. Science, 356<\/em>(6334), 183\u2013186.<\/p>\n

        DataCamp. (2022). The state of data science in social media analytics: Industry report 2022<\/em>. DataCamp Publications.<\/p>\n

        Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North <\/em>American Chapter of the Association for Computational Linguistics <\/em>(pp. 4171\u20134186). Association for Computational Linguistics.<\/p>\n

        Gayo-Avello, D. (2012). A meta-analysis of state-of-the-art electoral prediction from Twitter data. Social Science Computer Review, 31<\/em>(6), 649\u2013679.<\/p>\n

        Ghosh, A., & Veale, T. (2016). Fracking sarcasm using neural network. In Proceedings of the 7th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media<\/em><\/p>\n

        Analysis <\/em>(pp. 161\u2013169). Association for Computational Linguistics.<\/p>\n

        Hugging Face. (2023). Job<\/em> architecture:<\/em> Social<\/em> media<\/em> data<\/em> scientist<\/em>. Hugging Face Careers.<\/p>\n

        Kumar, A., & Jaiswal, A. (2020). Systematic literature review of sentiment analysis on social media. International<\/em> Journal<\/em> of<\/em> Information<\/em> Management<\/em> Data<\/em> Insights,<\/em> 1<\/em>(1), 100005.<\/p>\n

        Liu, B. (2015). Sentiment analysis: Mining opinions, sentiments, and emotions<\/em>. Cambridge University Press.<\/p>\n

        Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., &<\/p>\n

        Stoyanov, V. (2019). RoBERTa: A robustly optimized BERT pretraining approach. arXiv<\/em> preprint<\/em>, arXiv:1907.11692.<\/p>\n

        Nguyen, D. Q., Vu, T., & Nguyen, A. T. (2020). BERTweet: A pre-trained language model for English tweets. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing <\/em>(pp. 9\u201314). Association for Computational Linguistics.<\/p>\n

        Nielsen, F. \u00c5. (2011). A new ANEW: Evaluation of a word list for sentiment analysis in<\/p>\n

        microblogs. Proceedings of the ESWC2011 Workshop on 'Making Sense of Microposts' <\/em>(pp. 93\u201398).<\/p>\n

        Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., & Sutskever, I. (2021). Learning transferable visual models from natural language supervision. Proceedings of the 38th International Conference on Machine <\/em>Learning<\/em>, 139, 8748\u20138763.<\/p>\n

        Sap, M., Card, D., Gabriel, S., Choi, Y., & Smith, N. A. (2019). The risk of racial bias in hate speech detection. In Proceedings<\/em> of<\/em> the<\/em> 57th<\/em> Annual<\/em> Meeting<\/em> of<\/em> the<\/em> Association<\/em> for<\/em> Computational<\/em><\/p>\n

        Linguistics <\/em>(pp. 1668\u20131678). Association for Computational Linguistics.<\/p>\n

        Sarker, A., Lakamana, S., Hogg-Bremer, W., Xie, A., Al-Garadi, M. A., & Yang, Y. C. (2020). Self-reported COVID-19 symptoms on Twitter: An analysis and research resource. Journal of Medical <\/em>Internet Research, 22<\/em>(8), e20551.<\/p>\n

        Sharon, T. (2021). Blind-sided by privacy? Digital contact tracing, the Apple\/Google API and big tech's newfound role as global health policy makers. Ethics and Information Technology, 23<\/em>(1), 45\u201357.<\/p>\n

        Tumasjan, A., Sprenger, T. O., Sandner, P. G., & Welpe, I. M. (2010). Predicting elections with Twitter: What 140 characters reveal about political sentiment. Proceedings<\/em> of<\/em> the<\/em> Fourth <\/em>International AAAI Conference on Weblogs and Social Media<\/em>, 10(1), 178\u2013185.<\/p>\n

        Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, \u0141., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing <\/em>Systems,<\/em> 30<\/em>, 5998\u20136008.<\/p>\n

        Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, <\/em>359<\/em>(6380), 1146\u20131151.<\/p>\n<\/figure>\n

        <\/p>\n<\/div>\n<\/div><\/div><\/div><\/div><\/div>","protected":false},"excerpt":{"rendered":"

        Aziz Ozmen, PhDaziz.ozmen@gc4ss.org Senior Security AnalystGlobal Center for Security Studies The Digital Pulse: Leveraging Natural Language Processing and Social Media Analytics for Real-Time Societal Sentiment Tracking [\u2026]<\/span><\/p>\n","protected":false},"author":519,"featured_media":2835,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[1],"tags":[238,230,233,236,235,234,237],"class_list":["post-2823","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-computational-social-science","tag-data-science","tag-natural-language-processing","tag-public-opinion-mining","tag-sentiment-analysis","tag-social-media-analytics","tag-transformer-models"],"aioseo_notices":[],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/08\/aozmen2024-2.png?fit=1081%2C400&ssl=1","jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p9RaMN-Jx","jetpack-related-posts":[{"id":2829,"url":"https:\/\/www.gc4ss.org\/?p=2829","url_meta":{"origin":2823,"position":0},"title":"From Open Source to Actionable Intelligence: The Role of Data Analysts and Data Scientists in NLP-Driven Cyber Threat Intelligence","author":"Aziz Ozmen","date":"March 6, 2026","format":false,"excerpt":"Aziz Ozmen, PhDaziz.ozmen@gc4ss.org Senior Security AnalystGlobal Center for Security Studies From Open Source to Actionable Intelligence: The Role of Data Analysts and Data Scientists in NLP-Driven Cyber Threat Intelligence Abstract The digital ecosystem is awash with unstructured textual data relevant to cybersecurity: threat intelligence reports, dark web forums, vulnerability disclosures,\u2026","rel":"","context":"In "Cyber Security"","block_context":{"text":"Cyber Security","link":"https:\/\/www.gc4ss.org\/?cat=56"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/03\/aozmen2026.png?fit=1081%2C400&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/03\/aozmen2026.png?fit=1081%2C400&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/03\/aozmen2026.png?fit=1081%2C400&ssl=1&resize=525%2C300 1.5x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/03\/aozmen2026.png?fit=1081%2C400&ssl=1&resize=700%2C400 2x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/03\/aozmen2026.png?fit=1081%2C400&ssl=1&resize=1050%2C600 3x"},"classes":[]},{"id":2825,"url":"https:\/\/www.gc4ss.org\/?p=2825","url_meta":{"origin":2823,"position":1},"title":"The Force Multiplier: Institutionalizing the Data Analyst and Data Scientist in Modern Cybersecurity Operations","author":"Aziz Ozmen","date":"July 6, 2025","format":false,"excerpt":"Aziz Ozmen, PhDaziz.ozmen@gc4ss.org \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 Senior Security Analyst\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 Global Center for Security Studies The Force Multiplier: Institutionalizing the Data Analyst and Data Scientist in Modern Cybersecurity Operations Abstract The contemporary cybersecurity landscape is characterized by an unprecedented volume, velocity, and variety of data,\u2026","rel":"","context":"In "Cyber Security"","block_context":{"text":"Cyber Security","link":"https:\/\/www.gc4ss.org\/?cat=56"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2025\/07\/aozmen2025.png?fit=1081%2C400&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2025\/07\/aozmen2025.png?fit=1081%2C400&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2025\/07\/aozmen2025.png?fit=1081%2C400&ssl=1&resize=525%2C300 1.5x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2025\/07\/aozmen2025.png?fit=1081%2C400&ssl=1&resize=700%2C400 2x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2025\/07\/aozmen2025.png?fit=1081%2C400&ssl=1&resize=1050%2C600 3x"},"classes":[]},{"id":2816,"url":"https:\/\/www.gc4ss.org\/?p=2816","url_meta":{"origin":2823,"position":2},"title":"From Transaction to Prediction: The Roles of Data Analysts and Data Scientists in Customer Lifetime Value Estimation and Churn Reduction","author":"Aziz Ozmen","date":"May 6, 2022","format":false,"excerpt":"Aziz Ozmen, PhDaziz.ozmen@gc4ss.org \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Senior Security Analyst\u00a0 \u00a0 \u00a0 \u00a0 \u00a0Global Center for Security Studies From Transaction to Prediction: The Roles of Data Analysts and Data Scientists in Customer Lifetime Value Estimation and Churn Reduction Abstract The digital transformation of retail has produced an unprecedented wealth of\u2026","rel":"","context":"In \"Apriori Algorithm\"","block_context":{"text":"Apriori Algorithm","link":"https:\/\/www.gc4ss.org\/?tag=apriori-algorithm"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2022\/05\/aozmen2022.webp?fit=1200%2C800&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2022\/05\/aozmen2022.webp?fit=1200%2C800&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2022\/05\/aozmen2022.webp?fit=1200%2C800&ssl=1&resize=525%2C300 1.5x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2022\/05\/aozmen2022.webp?fit=1200%2C800&ssl=1&resize=700%2C400 2x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2022\/05\/aozmen2022.webp?fit=1200%2C800&ssl=1&resize=1050%2C600 3x"},"classes":[]},{"id":2784,"url":"https:\/\/www.gc4ss.org\/?p=2784","url_meta":{"origin":2823,"position":3},"title":"Emergency Management as an Interdisciplinary Field: Governance, Leadership, and Social Vulnerability","author":"Tuncay Unal","date":"January 22, 2021","format":false,"excerpt":"Tuncay Unal, PhDtuncay.unal@gc4ss.org ExpertGlobal Center for Security Studies Emergency management is a multidisciplinary field that draws on public administration, political science, sociology, geography, and organisational studies. While early research concentrated primarily on disaster response and operational issues, contemporary scholarship has broadened the scope to include mitigation, preparedness, ethics, governance, and\u2026","rel":"","context":"In "Conflicting Zones"","block_context":{"text":"Conflicting Zones","link":"https:\/\/www.gc4ss.org\/?cat=43"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/01\/Blogimage-Tunal2.png?fit=1200%2C800&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/01\/Blogimage-Tunal2.png?fit=1200%2C800&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/01\/Blogimage-Tunal2.png?fit=1200%2C800&ssl=1&resize=525%2C300 1.5x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/01\/Blogimage-Tunal2.png?fit=1200%2C800&ssl=1&resize=700%2C400 2x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2026\/01\/Blogimage-Tunal2.png?fit=1200%2C800&ssl=1&resize=1050%2C600 3x"},"classes":[]},{"id":317,"url":"https:\/\/www.gc4ss.org\/?p=317","url_meta":{"origin":2823,"position":4},"title":"The Future of Democracy in the Age of Social Media","author":"Ahmet Celik","date":"April 14, 2018","format":false,"excerpt":"Ahmet Celik, PhD ahmet.celik@gc4ss.org Senior Expert Global Center for Security Studies Nowadays, in certain countries, it is really difficult to find a brand-new vehicle with mechanical key system, most of which can be operated by a remote control without requiring to push any button. If you approach your car with\u2026","rel":"","context":"In "Democracy And Rule of Law"","block_context":{"text":"Democracy And Rule of Law","link":"https:\/\/www.gc4ss.org\/?cat=57"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2018\/04\/By-By-Democrasi.png?fit=560%2C315&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2018\/04\/By-By-Democrasi.png?fit=560%2C315&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2018\/04\/By-By-Democrasi.png?fit=560%2C315&ssl=1&resize=525%2C300 1.5x"},"classes":[]},{"id":2821,"url":"https:\/\/www.gc4ss.org\/?p=2821","url_meta":{"origin":2823,"position":5},"title":"Silent Signals: The Role of Data Analysts and Data Scientists in Algorithmic Market Anomaly Detection for Fraud and Insider Trading Identification","author":"Aziz Ozmen","date":"February 6, 2024","format":false,"excerpt":"Aziz Ozmen, PhDaziz.ozmen@gc4ss.org \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 Senior Security Analyst\u00a0 \u00a0 \u00a0 \u00a0 \u00a0Global Center for Security Studies Silent Signals: The Role of Data Analysts and Data Scientists in Algorithmic Market Anomaly Detection for Fraud and Insider Trading Identification Abstract The digitization of global financial markets has produced an\u2026","rel":"","context":"In \"Silent Signals: The Role of Data Analysts and Data Scientists in Algorithmic Market Anomaly Detection for Fraud and Insider Trading Identification\"","block_context":{"text":"Silent Signals: The Role of Data Analysts and Data Scientists in Algorithmic Market Anomaly Detection for Fraud and Insider Trading Identification","link":"https:\/\/www.gc4ss.org\/?tag=silent-signals-the-role-of-data-analysts-and-data-scientists-in-algorithmic-market-anomaly-detection-for-fraud-and-insider-trading-identification"},"img":{"alt_text":"","src":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/02\/aozmen2024.png?fit=1081%2C400&ssl=1&resize=350%2C200","width":350,"height":200,"srcset":"https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/02\/aozmen2024.png?fit=1081%2C400&ssl=1&resize=350%2C200 1x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/02\/aozmen2024.png?fit=1081%2C400&ssl=1&resize=525%2C300 1.5x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/02\/aozmen2024.png?fit=1081%2C400&ssl=1&resize=700%2C400 2x, https:\/\/i0.wp.com\/www.gc4ss.org\/wp-content\/uploads\/2024\/02\/aozmen2024.png?fit=1081%2C400&ssl=1&resize=1050%2C600 3x"},"classes":[]}],"_links":{"self":[{"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/posts\/2823","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/users\/519"}],"replies":[{"embeddable":true,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2823"}],"version-history":[{"count":0,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/posts\/2823\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=\/wp\/v2\/media\/2835"}],"wp:attachment":[{"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2823"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2823"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.gc4ss.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2823"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}