Damage Identification in Social Media Posts using Multimodal Deep Learning

Multimodal (Image, Captioned Image, and Text) | General | Response

This is a general-disaster dataset for damage assessment. It contains 10,875 images, 5,879 captioned images, and 19,031 textual data. It is introduced for multiclass classification.

  • ML task type: Multiclass classification
  • Data Source: Social Media (Instagram and Twitter); Publicly available datasets; Google Images
  • Size: 10,875 Images; 5,879 Captioned Images, 19,031 Text
  • Timespan: N/A
  • Geographical Coverage: Global
  • Baseline Information
  • Evaluated on: Maximum Decision Rule, Weighted Maximum Decision Rule, DFMC with ANN, DFMC with KNN, DFMC with SVM
  • Metrics used: Accuracy (Test)
  • Results as reported in original paper: 92.62 (DFMC with SVM)

Bibtex goes here