New method KG-Neighbor improves embedding visualization by using knowledge graphs

Researchers have developed KG-Neighbor, a novel method for constructing neighborhood graphs in embedding visualizations. This approach uses relational knowledge from knowledge graphs (KGs) to adjust embedding distances, aiming to better reflect content relationships rather than just input similarities. The method, which does not require class labels, also includes IDProfile to analyze residual mismatches between assigned classes and neighborhood majorities. Experiments across various datasets and encoder conditions showed that KG-Neighbor improved metrics like NMI and LT_2D compared to standard k-nearest neighbors baselines, indicating its effectiveness in aligning embedding neighborhoods with semantic content. AI

IMPACT Enhances the interpretability of embedding visualizations, potentially improving downstream tasks that rely on understanding semantic relationships.

RANK_REASON The cluster contains an academic paper detailing a new method for embedding visualization. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• embedding visualization
• IDProfile
• KG-Neighbor
• k-nearest neighbors algorithm
• knowledge graph
• Knowledge-Guided Neighborhood Graph Construction
• LT_2D

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →