Publications
Research Threads
All papers
- 2026
Critical Evaluation of Knowledge Graph-powered Question Answering Systems to Enable Responsible AI for Disaster Management
Venezuela-quake KG-QA: fluent, "grounded" answers mask wrong routing — evaluate the pipeline
- 2026
Data Fragmentation as a Critical Bottleneck in Disaster Response Planning: The EVAC-AI Case Study for Flood Risk Assessment
Neo4j KG of FEMA/NOAA/city/Waze data + local-LLM NL→Cypher chatbot-map
- 2025
Knowledge-guided Continual Learning for Behavioral Analytics Systems
Wiktionary-KG augmentation of replay nearly eliminates forgetting (HateXplain acc 16.7→46.7)
- 2025
Eliciting Socio-Technical Research Needs on Countering Violence-Inducing and Hateful Speech in Cyberspace
Participatory needs study (victims + 20 orgs) → responsible multilingual HS framework
- 2024
Designing a Human-Centered AI Tool for Proactive Incident Detection Using Crowdsourced Data Sources to Support Emergency Response
Interactive tool for practitioners to pick Pareto-optimal crash detectors; CNN 2× KNN F1
- 2024
EVO-LYZER: Social Media Mining System for Evolving Communication Behavior Analytics to Aid Climate Change Programs
Topic-model divergence/consistency show +31% climate posting after Climate Matters training
- 2024
Lifelong Learning Framework for Multilingual Hate Speech Detection in Social Media Streams
Replay-memory lifelong learning over topic×language task stream; best on 7/10 tasks
- 2024
Closing the Knowledge Gap in Designing Data Annotation Interfaces for AI-powered Disaster Management Analytic Systems
LLM "Context" interface rated best by CERT annotators; accuracy on par
- 2023
Citizen-Helper System for Human-Centered AI Use in Disaster Management
Human-centered AI design principles via Citizen-Helper deployments (COVID-19, responder training)
- 2023
Violence-Inducing Behavior Prevention in Social-Cyber Space
Project design for multilingual, KB-augmented violence-inducing behavior detection
- 2023
A Comparative Study of Pre-trained Language Models to Filter Informative Code-mixed Data on Social Media during Disasters
First Persian–English code-mixed crisis dataset; mBERT best (F1 72%)
- 2023
Summarizing Social Media & News Streams for Crisis-related Events by Integrated Content-Graph Analysis: TREC-2023 CrisisFACTS Track
Fact extraction + query-extended filtering + graph centrality ranking of crisis facts
- 2022
PROBER: A System for Real-time Propaganda Behavior Analytics on Social Media and Web Data Streams
Real-time system: tweet stream → BERT propaganda-technique tagger → Kibana dashboard
- 2022
Human-AI Teaming for COVID-19 Response: A Practice & Research Collaboration Case Study
Reflexive case study of a researcher–practitioner–volunteer collaboration; 8 best practices
- 2021
Evaluating the Relevance of UMLS Knowledge Base for Public Health Informatics during Disasters
Only 16% of UMLS concepts relevant for disaster-health terms; 33% of terms missing
- 2021
Mining Risk Behaviors from Social Media for Pandemic Crisis Preparedness and Response
CERT-co-designed risk-behavior schema, 13.6k labels; filter AUC 0.88, behavior F1 0.81
- 2021
Practitioner-Centric Approach for Early Incident Detection Using Crowdsourced Data for Emergency Services
CNN + Pareto optimization trades F1 vs. localization; F1 41 vs 0.6, ~40% crashes caught early
- 2021
Efficient Detection of Multilingual Hate Speech by Using Interactive Attention Network with Minimal Human Feedback
Interactive attention + <4% human feedback; AUC 84.1, attention aligns with semantic frames
- 2021
Online-Computer-Mediated Interviews and Observations: Overcoming Challenges and Establishing Best Practices in a Human-AI Teaming Context
Methods for screen-shared online observation of volunteer annotators; 5 new challenges
- 2021
Towards a Design of Resilience Data Repository for Community Resilience
Practitioner survey (n=35) informs a community-resilience data repository architecture
- 2020
Emergency Incident Detection from Crowdsourced Waze Data Using Bayesian Information Fusion
Bayesian fusion of noisy Waze reports detects crashes ~5.9 min before official records
- 2020
Evaluating Semantic Feature Representations to Efficiently Detect Hate Intent on Social Media
Hatebase/FrameNet/embedding features boost precision and cross-dataset F1 (39.6→56.8)
- 2020
Exploring Deep Neural Networks and Transfer Learning for Analyzing Emotions in Tweets
LSTM-CNN + DeepMoji/sentiment-neuron transfer ≈ SOTA emotion intensity (r 0.798) + SHAP explanations
- 2019
A Hybrid Approach for Aspect Extraction from Customer Reviews
CNN + SVM probability ensemble sets best F1 on SemEval-2016 (77.2 / 54.5)
- 2019
ACTSEA: Annotated Corpus for Tamil & Sinhala Emotion Analysis
First Tamil/Sinhala emotion corpus (~900k tweets) via morphological keyword expansion
- 2018
DataSEARCH at IEST 2018: Multiple Word Embedding Based Models for Implicit Emotion Classification of Tweets with Deep Learning
Stacked LSTM-CNN over 6 embeddings for implicit emotion; macro-F1 68.1, ranked 8th
- 2018
Aspect Extraction from Customer Reviews Using Convolutional Neural Networks
Multi-kernel CNN with theory-sized layers; best restaurant F1 (76.0) at the time
- 2017
Multi-domain Aspect Extraction using Support Vector Machines
Feature-engineered one-vs-rest SVM beats SemEval-2016 best on both domains (F1 74.2 / 52.2)