Xiang Dai (戴翔)

[CV-full]     [CV-1 page]

I am a research scientist (in Natural Language Processing) at CSIRO, Australia’s national science agency.

My earlier work focused on information extraction, especially identifying entities and relations from domain-specific text and linking these entities to corresponding ontologies. One example application of information extraction in the medical domain is active surveillance in pharmacovigilance: I developed models to detect descriptions of adverse drug events in patient‑written posts. I also created datasets and organised shared tasks on mapping these layman descriptions to standard terms in MedDRA, a medical dictionary used for regulatory purposes.

These days, my research focuses on building cost-effective domain-specific large language models that excel at understanding specialized text and capturing both implicit and explicit domain knowledge. I also work on human-centric AI, aiming to adapt AI systems to users' backgrounds and specific information needs. I am interested in understanding how AI can help people make informed decisions or act on complex information. For example, this may involve supporting a biologist to navigate scientific literature and simulation results to decide which genes to investigate for a particular trait, or helping patients understand their health records and follow medical advice.

I also work(ed) with students and Postdocs on topics including digital health, plain language summarization, multimodal learning, human–robot interaction, and LLM evaluation.

A full list of my publications can be found on the Publications page or on Google scholar.

News

• 2026-06: We are organizing a special section at ACM Transactions on Information Systems (TOIS). The topic is 'Evaluating Multimodal Information Systems: Metrics, Benchmarks, and Reliability'.
• 2026-02: We are organizing the 2nd Workshop on Evaluation for Multimodal Generation. It will be held in Melbourne, Australia on July 24 as part of SIGIR2026.
• 2025-08: Sim's paper 'The More, The Better? A Critical Study of Multimodal Context in Radiology Report Summarization' was accepted to Findings of EMNLP 2025.
• 2025-07: We are organizing a shared task at ALTA 2025, focusing on normalizing adverse drug event descriptions in social media posts to MedDRA.
• 2025-06: I received an Early Career Researcher Award from the Analytics and Decision Sciences program at Data61.
• 2025-05: I gave a guest lecture at Macquarie University on domain-specific VLMs.
• 2025-05: Sim's paper 'Can VLMs Actually See and Read? A Survey on Modality Collapse in Vision-Language Models' was accepted to Findings of ACL 2025.
• 2025-04: I gave a guest lecture at the University of New South Wales on active adverse event surveillance.
• 2025-01: I became a member of the ALTA (Australasian Language Technology Association) executive committee for 2025-2026.
• 2024-11: Our paper 'MultiADE: A Multi-domain Benchmark for Adverse Drug Event Extraction' was accepted to Journal of Biomedical Informatics.
• (more)

Email

dai (dot) xiang (dot) au (at) gmail.com

dai (dot) dai (at) csiro.au

Calendar