Multimodal Classification on User-generated Content During Wildfires in Canada
Bibliographic record
Abstract
The ability to swiftly and efficiently assess information is essential for a successful response in the case of a natural disaster, such as wildfires. Access to accurate and timely information is critical for first responders and the public to make informed decisions within tight timelines. Although traditional sources of information such as satellite images, field surveys, and contact centers are costly and time consuming, social media content is more affordable and immediately available, making it a useful tool for real-time updates. However, extracting insights from the abundance of user-generated content on social media can be challenging. Machine learning can automatically categorize posts, filter noise and highlight crucial information for responders and the public. In this work, we introduce, WildFireCan-MMD, a new multimodal dataset tailored to Canadian wildfires, collected from user-generated posts on X during the 2022, 2023, and 2024 British Columbia and Alberta wildfires. We applied topic clustering followed by manual annotation to label this dataset into thirteen wildfire-relevant themes. We tested two Generative Large Vision-Language Models (GLVLMs), namely, gpt-4o-mini and open-source LLaVA, for the classification of WildFireCan-MMD in zero-shot setting. We also developed custom classifiers, including RoBERTa and ViT transformer combinations with early and late fusion. Our results show that if no training data is available, gpt-4o-mini might serve as a practical zero-shot classifier. However, with labeled training data, even a basic Gaussian Naive Bayes classifier outperforms gpt-4o-mini. Fine-tuned transformers outperform gpt-4o-mini by a large margin of 23%. This study underscores the value of curated, task-specific datasets and training customized classifiers despite recent advancements in zero-shot GLVLMs.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".