Evidano home
Commentary on News

Patient Insight: AI Qualitative Analysis of Social Media

Evidano9 min read

Researchers studying patient communities need more than counts of posts, likes, or retweets. They need to understand how patients, advocates, oncologists, and organizations exchange knowledge, form relationships, and influence healthcare discussions over time. AI qualitative analysis of social media can help research teams move from large-scale discovery to close reading of the conversations that explain why patterns matter. The November 20, 2026 study in PLoS One, “Tweeting with precision: A multiscale longitudinal digital ethnography of online cancer communities, ” provides a useful model. The authors analyzed more than 8 million Twitter publications from 2011 to 2022, combining semantic analysis, user classification, retweet networks, and qualitative interpretation. This post explains what the study found, why mixed-method analysis matters for healthcare and policy researchers, and how AI-assisted qualitative workflows can make comparable research more systematic. The findings are research-focused and non-diagnostic; they should not be used as medical advice.

Key Takeaways

According to the November 20, 2026 PLOS One study, AI-assisted qualitative analysis of social media is most useful when researchers combine topic discovery, stakeholder classification, network analysis, and close reading of representative conversations.

The study found that online cancer communities were not only information channels: according to the authors, patient advocates helped connect patients with oncologists, translate biomedical knowledge, and raise questions about clinical trials, treatment access, and healthcare policy.

  • The PLOS One researchers analyzed 8, 614, 284 Twitter publications collected from 2011 to 2022, including 4, 389, 619 tweets and 3, 031, 281 retweets.
  • The researchers identified 834 highly active accounts between March and April 2023; those accounts represented less than 3% of 28, 524 users but produced more than 80% of tweets containing the cancer-community hashtags.
  • The study's topic model began with 116 clusters and consolidated them into 49 topic groups and 13 overarching categories.
  • The PLOS One analysis found that 2.4% of the corpus referred to precision oncology, while cancer-related topics accounted for 50% of the corpus after excluding outliers.
  • The qualitative case study examined 7, 670 tweets organized into approximately 3, 000 conversations involving Janet Freeman-Daily and a wider network of advocates, oncologists, patients, and healthcare professionals.

How AI Qualitative Analysis Mapped Cancer Communities

The PLOS One study used a mixed-method workflow because computational patterns alone could not explain how cancer communities created knowledge and influenced healthcare discussions.

The authors first defined a cancer-related hashtag ontology of approximately 30 hashtags and collected posts from January 1, 2010, through December 31, 2022. The researchers then selected users who had posted more than 200 relevant hashtagged tweets and retrieved the full tweet histories of the selected accounts during data collection in March and April 2023.

The researchers classified 834 accounts into eight stakeholder categories using account descriptions, manual review by four researchers, and interviews with cancer-advocacy participants. The categories were patients, advocates, oncologists, health professionals, health centers, media, nonprofit organizations, and industries.

For semantic analysis, the authors used BERTopic with transformer-based embeddings, UMAP dimensionality reduction, and HDBSCAN clustering. The researchers also reviewed the 50 most representative tweets for each topic, showing why human interpretation remained necessary after automated discovery.

The authors summarized the qualitative value of the approach through a direct finding: “The qualitative approach reveals that, despite the limited nature of Twitter messages, both oncologists and advocates have turned it into a discussion space capable of providing concrete answers to patients’ questions.”

Findings Snapshot for Healthcare Researchers

Date or periodMetricValueResearch implication
2011 to 2022Twitter corpus coverageMore than 10 years and more than 8 million publicationsLongitudinal analysis can show how communities and scientific topics change over time.
March to April 2023Final dataset size8, 614, 284 publications, including 4, 389, 619 tweets and 3, 031, 281 retweetsLarge datasets require automated discovery before researchers conduct detailed qualitative review.
2011 to 2022Topic structure116 initial clusters, consolidated into 49 groups and 13 overarching categoriesHierarchical coding can connect specific terms, such as biomarkers, to broader social and policy themes.
2011 to 2022Precision-oncology discussion2.4% of tweetsA relatively small topic can still be analytically important when linked to treatment access, advocacy, and regulation.
2011 to 2022Cancer-related discussion50% of the corpus after excluding outliersResearchers should preserve contextual and non-technical discussion rather than filtering for medical terms alone.
July 2013 to December 2021Lung cancer social-media chatsMore than 150 chatsRepeated, moderated conversations can provide qualitative evidence about stakeholder priorities and policy concerns.

Implications for AI-Assisted Qualitative Research

Why should researchers combine frequency analysis with thematic analysis?

Researchers should combine frequency analysis with thematic analysis because high-volume topics do not necessarily explain the most consequential community dynamics.

The PLOS One study found that mundane topics represented 25% of labeled tweets, while cancer topics represented 50% after excluding outliers. The finding shows that social support, ordinary interaction, and relationship maintenance can provide context for how health information circulates.

An AI-enabled workflow can identify frequent terms and themes first, then help researchers inspect representative excerpts, exceptions, and cross-stakeholder differences.

How can qualitative researchers identify stakeholder differences?

Researchers can identify stakeholder differences by combining a transparent coding scheme with cross-segment analysis of themes, language, and interactions.

According to the PLOS One study, oncologists authored 40% of biomarker-related tweets, while advocates and media each contributed 17%. The authors also reported that advocates discussed a broader range of issues and appeared to bridge medical and patient communities, although the network finding was descriptive rather than causal.

AI-assisted cross-segment analysis can help compare patients, advocates, clinicians, organizations, and media without reducing their perspectives to a single overall average.

What does the ROS1 case show about qualitative context?

The ROS1 case shows that qualitative context can connect online exchanges with real-world community formation, treatment access concerns, and policy questions.

The PLOS One authors linked conversations among advocates and oncologists to the formation of The ROS1ders in 2015 and examined discussions before and after the U.S. Food and Drug Administration approved crizotinib for ROS1-positive non-small-cell lung cancer on March 11, 2016.

Janet Freeman-Daily described Twitter as “the one place where all stakeholders […] could meet and exchange ideas on something that had life-and-death implications for so many, ” a quotation that captures why researchers must preserve participant language and interactional context.

How Evidano Helps with AI Qualitative Analysis of Social Media

Problem: Large social datasets are difficult to synthesize

Evidano is an AI-powered qualitative data analysis platform that helps researchers analyze social posts, reports, interviews, open-ended survey responses, and other documents.

Researchers can use thematic, content, frequency, and cross-segment analyses to move from a large document collection to recurring themes, stakeholder differences, and evidence-backed summaries. Explore Evidano's qualitative research features.

Problem: Automated topics can lose the meaning of conversations

Solution: Researchers can use Evidano's AI chat over documents and analyses to ask focused questions about representative excerpts, competing interpretations, and the context surrounding a theme.

A research team can use the workflow to test whether a high-frequency term reflects clinical information, emotional support, advocacy, policy debate, or ordinary social interaction. Human researchers should review sensitive health content and make the final analytic judgment.

Problem: Stakeholder differences remain hidden in aggregate results

Solution: Evidano supports cross-segment analysis, enabling researchers to compare themes and language across groups such as patients, advocates, clinicians, media, and nonprofit organizations.

Researchers can also use visualizations such as word clouds, co-occurrence networks, and hierarchical codes with subcodes to communicate how specific biomarkers connect to broader topics.

Problem: Health research data requires careful handling

Solution: Evidano encrypts data and does not use customer data to train third-party models, which is relevant when researchers work with sensitive transcripts, survey responses, or collected online material. Review Evidano's data-security approach.

Ethical review, platform terms, privacy protections, and responsible reporting remain necessary. The PLOS One authors shared tweet IDs rather than full tweet content because of personal-data concerns and platform terms, while acknowledging that deletion, account suspension, and access changes can limit reproducibility.

FAQ: AI Qualitative Analysis of Social Media

What is AI qualitative analysis of social media?

AI qualitative analysis of social media uses machine-learning and language-analysis methods to organize large collections of posts while supporting human interpretation of themes, narratives, and interactions.

The November 20, 2026 PLOS One study combined semantic analysis, user classification, retweet-network analysis, and close reading to study cancer communities over more than 10 years.

How did the PLOS One study analyze cancer communities?

The PLOS One researchers analyzed more than 8 million Twitter publications from 2011 to 2022 and combined topic modeling, manual stakeholder classification, network analysis, and qualitative review.

The researchers identified 834 highly active accounts, classified them into eight categories, and examined 7, 670 tweets in approximately 3, 000 conversations for the detailed ROS1 case study.

What were the main findings about patient advocates?

The PLOS One study found that patient advocates helped connect different stakeholder groups and discussed medical, social, research, and policy issues.

The authors reported that advocates and oncologists both showed homophilic retweeting patterns, but advocates retweeted a wider range of actors and appeared to serve as bridges between patients and medical communities. The authors presented this as a descriptive finding, not proof of causation.

Can AI replace human qualitative researchers in healthcare studies?

AI cannot replace human qualitative researchers in healthcare studies because interpretation, ethics, sampling decisions, and contextual judgment remain essential.

The PLOS One researchers reviewed representative tweets qualitatively after computational topic modeling, and they acknowledged limitations involving active-user sampling, a focal advocate, an Anglo-American context, and changes from Twitter to X.

What privacy issues affect AI analysis of social media research?

Privacy issues include personal health information, re-identification risk, deleted posts, platform terms, and the ethical limits of reproducing public conversations.

The PLOS One authors shared tweet IDs rather than full tweet content and stated that deletion, account suspension, and platform-access changes could limit exact reproducibility.

Conclusion and Next Steps

The November 20, 2026 PLOS One study demonstrates how AI-assisted discovery and human qualitative interpretation can reveal the structure, language, and influence of online cancer communities. The study's 8.6 million-publication corpus shows the value of longitudinal analysis, while its ROS1 case study shows why researchers must examine conversations in context. For healthcare, policy, and patient-research teams, the practical lesson is to combine thematic analysis, frequency analysis, stakeholder comparison, and careful excerpt review rather than relying on volume metrics alone. Evidano helps research teams organize and interpret interviews, documents, surveys, and social-media material in one AI-powered qualitative workflow. Try Evidano for free.

Topics

  • AI qualitative analysis of social media
  • qualitative social media analysis
  • AI analysis of patient communities
  • digital ethnography cancer communities
  • healthcare qualitative research

Keep reading

Browse all articles