My paper reading repository
- NOIGAN: NOISE AWARE KNOWLEDGE GRAPH EMBEDDING WITH GAN
- Proposed a unified GAN framework to jointly combine the two tasks, knowledge graph completion by embedding model and error detection.
- Experimented on FB15K, YOGA3-10, WN18 with noisy triples that are made by randomly substituting the head entity or the tail entity.
- On Understanding Knowledge Graph Representation
- Assume that knowledge graph embeddings have similar properties to word embeddings.
- Conversation Generation with ConceptFlow
- Similar idea to my paper DyKgChat, but using graph neural network and focusing on 2-hops as maximum.
- Experimented on Commonsense Conversation Dataset, primarily used overlapping metrics for evaluation.