Knowledge Graph Completion-based Question Selection for Acquiring Domain Knowledge through Dialogues
summary
Framework selects missing knowledge-graph triplets to turn into natural questions by ranking KGC scores and asking high-scoring items; model uses tensor-factorization ComplEx and two training tweaks called SUB modification (linking entities with shared name substrings) and NF negative sampling (restricting negatives). Cross-validation shows improved link-prediction performance, and an open user study with 200 crowdworkers finds the modified model yields more correct questions and that picking only top-scored predictions raises subjective correctness.
AI-written summary · about