{"resourceId":"tamu-crisissense-llm-disaster-social-media-classifier-20261106","versions":[{"version":"external-8db0d339a73a855ad4b5ec19033169012c222f429dbb39eebaea0a9b0eac4d23","resource":{"id":"tamu-crisissense-llm-disaster-social-media-classifier-20261106","title":"Texas A&M announces CrisisSense-LLM, a disaster social-media classifier still in development and not yet deployed in an active disaster","organization":"Texas A&M University Zachry Department of Civil and Environmental Engineering","sector":"Emergency management / crisis informatics","geography":"United States (Texas)","publishedAt":"Texas A&M Stories, Nov. 6, 2026; underlying paper dated May 26, 2026","publicationDate":"2026-11-06","eventDate":null,"sourceName":"Texas A&M Stories","sourceLabel":"Researchers develop AI tool to improve crisis communication","sourceUrl":"https://stories.tamu.edu/news/2026/11/06/researchers-develop-ai-tool-to-improve-crisis-communication/","evidenceClass":"academic-research","outcomeClass":"emerging","topics":["knowledge-work","data-security"],"finding":"A university press release describes an NSF-funded large language model that filters and categorizes social media posts during crises such as hurricanes into disaster-related, weather-related and humanitarian information, and further by event type, informativeness and humanitarian information type. It was tested on a benchmark of human-annotated posts from multiple disaster events. The release states the system is under development and has not been deployed during an active disaster, and the team plans an interactive AI agent for emergency professionals. The related paper (CrisisSense-LLM, DOI 10.1016/j.aei.2026.104836, May 26, 2026) was not opened, and the release gives no accuracy figures.","sledRelevance":"Shows research interest in LLM triage of crisis information flows for emergency operations centers; it is a signal of direction, not evidence of operational benefit.","evidence":"Opened the Texas A&M Stories release via HTTPS curl (User-Agent SLED-research-bot/1.0) and saved the text. The journal article was not opened.","architectureImplications":"Would sit between social-media ingestion and EOC situational-awareness tools as a classification layer; ingestion, rate limits and platform access are unaddressed.","governanceImplications":"Social-media-derived situational awareness needs rules on verification, misclassification handling and when staff may act on classified posts.","securityPrivacyImplications":"Processing public posts raises privacy and misinformation concerns not discussed in the release.","caveats":"University press release with no performance metrics; benchmark evaluation only; not deployed; paper not opened; social media data bias toward connected, English-speaking users is not discussed.","streamIds":["emergency-services"],"roles":{"sales":"Sales framing: Texas A&M’s announcement shows where emergency-operations tooling is heading, LLM sorting of social posts into weather, damage and aid-request categories. It is early research with no published accuracy and no field use. Use it to open a conversation about situational-awareness workload, not to sell a product. A bounded engagement is a requirements and evaluation workshop. Do not claim speed gains, accuracy or readiness for hurricane response; the release states the system has not been used in an active disaster.","engineering":"Engineering view: this would be a classification layer ingesting public posts and tagging event type, informativeness and humanitarian category for an operations centre dashboard. The release gives no metrics, latency, model size or hosting detail. Any evaluation should use the agency’s own past events, measure misclassification of urgent requests and test duplicates and false rumors. Proof of value is precision and recall on labeled local posts, not the benchmark alone. Treat as research-stage and unvalidated.","delivery":"Delivery plan: if a pilot is considered, define the decision the classified feed informs, an owner in the operations centre and a human verification step before action. Dependencies are social media data access, annotators for local labels and EOC staff time. Acceptance should be measured misses of urgent requests and analyst time saved in an exercise. Risks include misinformation, privacy complaints and over-reliance when networks fail."},"roleDetails":{"sales":{"customerProblem":"Operations centers face floods of unverified reports during disasters and struggle to find actionable items quickly.","stakeholders":["Emergency management director","EOC planning section chief","Public information officer","University research partner"],"discoveryQuestions":["How do you triage social-media reports during activations today?","What share of analyst time goes to sorting?","Could you supply labeled past-event posts for testing?"],"valueHypothesis":"LLM classification may reduce sorting time in exercises, as interpretation; not shown by the source.","potentialEngagement":"Requirements and exercise-based evaluation design for social-media triage tools.","unsupportedClaims":["That the tool improves response times","That it has been used in a real disaster","Any accuracy figure"]},"engineering":{"fit":"Research-stage fit for EOCs exploring social-media situational awareness.","architectureIntegration":"Ingestion of public posts, LLM classification into categories, output to an EOC dashboard with human review.","prerequisites":["Social media data access","Labeled local events","Hosting and model governance"],"constraints":["No published metrics","Not deployed","Platform access limits"],"security":"Control data retention of ingested posts and log classifications; assess prompt injection from post content.","proofOfValue":["Precision and recall on local labeled posts","Miss rate on urgent aid requests","Analyst time saved in a tabletop exercise"]},"delivery":{"work":"Scope a small exercise-based pilot with human verification and measured analyst workload.","dependencies":["Data access","Annotators","EOC staff participation"],"ownership":"EOC planning section lead.","skillsAdoption":"Analysts need training on reviewing and correcting classifications.","governanceCheckpoints":["Use policy for social data","Pilot results review","Decision on continuation"],"acceptanceCriteria":["Urgent-request miss rate below agreed threshold in exercise","Documented analyst time change","Verification step used on all actioned items"],"risks":["Misinformation amplified","Privacy objections","Dependence when networks fail"]}},"retrievedAt":"2026-11-09T03:58:33Z","enrichedAt":"2026-11-09T04:01:55Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Aims to reduce analyst workload during surges; no workforce evidence given.","procurementImplications":"No product exists; agencies should treat it as research and require benchmark results on their own event types before any purchase.","operatingModelImplications":"Aimed at reducing time spent sorting reports so staff focus on decisions; not measured.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://stories.tamu.edu/news/2026/11/06/researchers-develop-ai-tool-to-improve-crisis-communication/","referenceExcerpt":"The system is under development and has not been deployed during an active disaster.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}