Text-attributed Graph Condensation via Text Selection and Attribute Matching
Researchers have developed TAGSAM, a novel method for condensing Text-Attributed Graphs (TAGs) to reduce computational costs. TAGSAM employs subgraph text selection and attribute similarity matching to compress both the text descriptions and graph topology. This approach significantly improves accuracy compared to existing methods, even when condensing the TAG to a mere 1% of its original size. AI
IMPACT Reduces computational requirements for processing text-attributed graph data, enabling larger datasets to be handled more efficiently.