Augmenting Visual Information in Knowledge Graphs for Recommendations
summary
Proposes integrating image-derived features into item knowledge graphs and learning hybrid user profiles via Skip-gram node embeddings to boost top-N recommendations. Extracted features include SIFT, SURF, ORB, color histogram, and CNN features from the Caffe reference model. Experiments use Amazon Clothing Shoes and Jewelry and HetRec2011-MovieLens-2K. Visually-augmented knowledge graphs plus visually-annotated meta-paths improve precision and recall versus semantic-only metapath2vec; SIFT wins on Amazon while color histogram and CNN work best on MovieLens; GAT shows gains only for MovieLens with color histograms.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:19Introduction
- 1:28The Proposed Approach
- 2:33Datasets
- 3:30Results (cont.)
- 4:06Conclusions
- 4:52References
related talks
22mExploiting Auxiliary Information for Better Recommendation
38mKnowledge Graphs : Adding a New Dimension to Vector Search and Generative AI
40mKnowledge graphs --enter- the Hype Cycle
19mKnowledge Graphs & Deep Learning at YouTube
01h 40mKnowledge Graphs & GraphRAG: Techniques for Building Effective GenAI Applications
44mBuilding Knowledgeable Machines
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
27mAdaptive Knowledge Graphs for Dynamic RAG Systems