De-identified Feature Embeddings and Labels for XJTU Campus Facility Maintenance (DAGF-Net)

<p dir="ltr">This dataset contains the de-identified feature embeddings and labels used to train and evaluate the DAGF-Net model in the manuscript "An Intelligent Campus Facility Maintenance Framework: Multimodal Fault Diagnosis and Dynamic Resource Scheduling".<b>File Contents:</b></p><ul><li><code>resnet_features.npy</code>: Image features extracted by ResNet-50 (Shape: 12527 x 2048).</li><li><code>bert_features.npy</code>: Text features extracted by BERT-base-chinese (Shape: 12527 x 768).</li><li><code>labels.npy</code>: Integer labels corresponding to the 11 core maintenance categories.</li><li><code>class_names.txt</code>: Mapping of integer labels to category names.</li></ul><p dir="ltr"><b>Privacy Note:</b> To strictly comply with campus privacy regulations, raw images and raw text logs are withheld.

However, these pre-extracted feature matrices allow researchers to fully reproduce the classification results (Tables 2, 3, and 4) and the ablation studies reported in the manuscript.</p>