Experimental Datasets for A Latent Affective Factor-Aware Dynamic Multi-Path Collaborative Reasoning Model for Image-Text Sentiment Analysis

The paper uses two public image-text datasets, MVSA-SINGLE and HFM. MVSA-SINGLE is used for a three-class image-text sentiment analysis task and contains 4,511 samples, including 1,358 negative, 470 neutral, and 2,683 positive samples.

It is randomly split into training, validation, and test sets at a ratio of 8:1:1 with random seed 55. HFM is used for a binary multimodal sarcasm/sentiment-related task and contains 24,635 samples, including 14,075 non-sarcastic and 10,560 sarcastic samples. The original training, validation, and test splits released by the dataset authors are adopted.