DiffTSP: Discrete Diffusion Model for Knowledge Graph Triple Set Prediction
A new discrete diffusion model called DiffTSP has been introduced to address the Triple Set Prediction task in knowledge graph completion. Unlike traditional methods that predict missing elements within individual triples given partial information, TSP aims to infer entire sets of missing triples based solely on the observed knowledge graph. Existing approaches process triples one by one, failing to capture dependencies among predictions for consistency. DiffTSP frames TSP as a generative task using a discrete diffusion process that adds noise by masking relational edges in the knowledge graph. The model then gradually recovers the complete graph through a reverse process conditioned on the observed data. This method represents arXiv preprint 2604.18344v1, which announces this new research contribution to the field of knowledge graph completion.
Key facts
- DiffTSP is a discrete diffusion model for Knowledge Graph Triple Set Prediction
- Triple Set Prediction aims to infer missing triples without assuming partial information
- Traditional KGC methods predict missing elements given one or two elements of a triple
- Existing TSP methods predict triples individually without capturing dependencies
- DiffTSP treats TSP as a generative task using discrete diffusion
- The model adds noise by masking relational edges in the knowledge graph
- The reverse process recovers the complete KG conditioned on observed data
- The research is documented in arXiv preprint 2604.18344v1
Entities
Institutions
- arXiv