AGHE - Approach for Generating Enhanced Heterogeneous Embeddings from Heterogeneous Graphs

Embeddings represent a viable solution to address the challenge of data and information generation in Heterogeneous Graphs. This paper presents the approach for generating and processing heterogeneous embeddings (AGHE), which are built from various data types such as text, images, and subgraphs embedded in nodes. The AGHE comprises several steps, from graph creation to generating embeddings using metapaths and aggregating information from neighboring nodes. The experiments conducted investigated the performance of Recommender System tasks applied to the generated embeddings: node-local text-based, neighbor-aggregated text-based, metapath-based, and text and metapath composition. Results underscore the effectiveness in representing data heterogeneity in Deep Learning systems based on Heterogeneous Graph.