Xsb8864/Culture-SupCon-Rec-Dataset

Culture-SupCon Multi-Modal Recommendation Dataset

This dataset contains the processed 5-core interaction graphs and pre-extracted multi-modal features for the paper "Your Paper Title".

Dataset Structure

Amazon-XLocale (Abo)

• —paper_interactions_5core.csv: User-Item interactions.
• —paper_item_metadata.csv: Item metadata including locale (Country) as Cultural Priors.
• —paper_text_feats.npy: 1024-dim text embeddings extracted via BGE-M3.
• —paper_vision_feats.npy: 4096-dim vision embeddings extracted via InternVL2.5-8B.

Yelp

• —paper_interactions_5core.csv: User-Item interactions.
• —paper_item_metadata.csv: Item metadata including locale (City) as Cultural Priors.
• —paper_text_feats.npy: 1024-dim text embeddings extracted via BGE-M3.
• —paper_vision_feats.npy: 4096-dim vision embeddings extracted via InternVL2.5-8B.

Steam

• —paper_interactions_5core.csv: User-Item interactions.
• —paper_item_metadata.csv: Item metadata including locale (Tags) as Cultural Priors.
• —paper_text_feats.npy: 1024-dim text embeddings extracted via BGE-M3.
• —paper_vision_feats.npy: 4096-dim vision embeddings extracted via InternVL2.5-8B.

How to Load in PyTorch

import numpy as np
vision_feats = np.load("yelp/paper_vision_feats.npy")
print(vision_feats.shape) # (33719, 4096)