Table of Contents: :: Library Catalog

Saved in:

Bibliographic Details
Main Author:	Lyu, Huanzhu
Format:	Recurso digital
Language:
Published:	Zenodo 2026
Online Access:	https://doi.org/10.5281/zenodo.18366640
Tags:	Add Tag No Tags, Be the first to tag this record!

Table of Contents:

<h1>Dataset Description</h1> Data for "Based on Data Balancing and Model Improvement for Multi-Label Emotion Recognition". This repository contains the comprehensive data and experimental results supporting our study on multi-label emotion recognition using the GoEmotions dataset. The dataset and materials are shared under the CC-BY 4.0 license. <h2>Core Dataset</h2> <ul> <li> balanced_emotion_dataset.csv <ul> <li> Final balanced multi-label sentiment dataset used for training and evaluation. </li> <li> Renamed from final_balanced_df_output.csv. </li> <li> Columns: text, sentiment (list of emotion labels). </li> </ul> </li> <li> goemotions_original.csv <ul> <li> Original GoEmotions data after removing example_very_unclear. </li> </ul> </li> <li> sentiment140_auto_labels.csv <ul> <li> Sentiment140 tweets labeled into the 28 GoEmotions categories. </li> <li> Columns include text and model_labels. </li> </ul> </li> <li> gpt4mini_generated_texts.csv <ul> <li> GPT-4 mini generated texts with target emotion prompts. </li> </ul> </li> </ul> <h2>Original Submission Data (Version 1)</h2> <h3>Data for Figures</h3> <ul> <li> balanced_label_counts.csv <ul> <li> Renamed from fig2_balanced_label_counts.csv. </li> <li> Counts of each of the 28 emotion labels in the final balanced dataset. </li> <li> Columns: Sentiment Labels, Counts. </li> </ul> </li> <li> training_history.csv <ul> <li> Training history log for figures (loss and accuracy per epoch). </li> <li> Columns: epoch, accuracy, loss, val_accuracy, val_loss. </li> </ul> </li> </ul> <h3>Source Code</h3> <ul> <li> model_pipeline.ipynb <ul> <li> Renamed from model (1).ipynb. </li> <li> Full notebook for data processing, model training, and evaluation. </li> </ul> </li> </ul> <h2>Updated Experimental Results (Version 2)</h2> In response to reviewer feedback, we conducted ablation studies and baseline comparisons. The following ablation archives are included: <ol> <li> ablation_unbalanced_attn.tar.gz <ul> <li> CNN + BiLSTM + Attention on original unbalanced GoEmotions. </li> </ul> </li> <li> ablation_unbalanced_noattn.tar.gz <ul> <li> CNN + BiLSTM (no attention) on original unbalanced GoEmotions. </li> </ul> </li> <li> ablation_balanced_attn.tar.gz <ul> <li> CNN + BiLSTM + Attention on oversampled balanced GoEmotions. </li> </ul> </li> </ol> <h3>Key Updates in Version 2</h3> <ul> <li> Extended Training: all models trained for 34 epochs (no early stopping). </li> <li> Validation-Only Threshold Optimization: thresholds tuned on validation only. </li> <li> Comprehensive Metrics: <ul> <li> Subset accuracy </li> <li> Jaccard index </li> <li> Hamming loss </li> <li> Micro/Macro Precision, Recall, F1-score </li> <li> Macro AUC </li> <li> Per-label metrics for all 28 emotion categories </li> </ul> </li> </ul> <h3>File Structure (inside each ablation archive)</h3> <ul> <li> *_loss.png </li> <li> *_precision.png </li> <li> *_recall.png </li> <li> per_label_metrics_thr0.5.csv </li> <li> per_label_metrics_thr_opt.csv </li> <li> f1_thr05.png </li> <li> f1_thr_opt.png </li> <li> summary.json </li> </ul> <h2>Quality Control and Audits</h2> <ul> <li> sentiment140_audit.csv <ul> <li> Audit samples for Sentiment140 auto-labels. </li> </ul> </li> <li> gpt4mini_annotations.csv <ul> <li> Five-annotator labels with majority vote. </li> </ul> </li> <li> gpt4mini_audit.csv <ul> <li> Audit samples for GPT-4 mini generated texts. </li> </ul> </li> </ul> <h2>Transformer Baseline (Version 3)</h2> <ul> <li> transformer_baseline_train.py <ul> <li> DistilRoBERTa baseline training script. </li> </ul> </li> <li> transformer_baseline_requirements.txt <ul> <li> Python dependencies for the baseline. </li> </ul> </li> <li> transformer_baseline_summary.json <ul> <li> Overall metrics at threshold 0.5 and optimized thresholds. </li> </ul> </li> <li> transformer_baseline_per_label_thr0.5.csv <ul> <li> Per-label metrics at threshold 0.5. </li> </ul> </li> <li> transformer_baseline_per_label_thr_opt.csv <ul> <li> Per-label metrics under validation-tuned thresholds. </li> </ul> </li> <li> transformer_baseline_thresholds_opt.csv <ul> <li> Optimized thresholds per label. </li> </ul> </li> </ul> <h2>Scripts</h2> <ul> <li> data_balancing_pipeline.py <ul> <li> Data integration, filtering, and balancing logic. </li> </ul> </li> <li> cnn_bilstm_training.py <ul> <li> CNN + BiLSTM + Attention training and evaluation script. </li> </ul> </li> <li> transformer_baseline_train.py <ul> <li> Transformer baseline training script. </li> </ul> </li> </ul> <h2>Notes</h2> <ul> <li> Split protocol: 80/10/10 with MultilabelStratifiedShuffleSplit, random_state=42. </li> <li> Threshold optimization: per-label grid search from 0.05 to 0.95 (step 0.05). </li> <li> File names are normalized and do not include timestamps or parentheses. </li> </ul> <h2>Citation</h2> If you use this dataset in your research, please cite the paper associated with this repository. <h2>Contact</h2> For questions about the data or experiments, please contact the corresponding author.

Similar Items