Reproduce “Expanding Knowledge Graphs with Humans in the Loop” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Emaad Manzoor, Jordan Tong, Sriniketh Vijayaraghavan, Rui Li
A novel method for knowledge graph expansion integrates "human-friendliness" into its learning objective, ensuring that even incorrect algorithmic predictions are located close to the true answer within the graph structure. This approach achieved a 400% reduction in time for Pinterest curators to add new interests and contributed to a 20% increase in shopping ad revenue, alongside demonstrating a 97% higher total score in controlled human-in-the-loop tasks.
Key insights
- Designing algorithms with "human-friendliness", where incorrect predictions are close to the true answer, significantly improves human-algorithm collaboration speed and accuracy in error correction.
- The method provides theoretical guarantees that the true parent of a new concept will be located within a bounded shortest path distance from the algorithm's top (even incorrect) prediction.
- Optimizing for traditional accuracy metrics in other methods does not automatically lead to improved human-friendliness, indicating a potential trade-off that the proposed method specifically addresses.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.