This paper was accepted on the Workshops on Information Science with Human within the Loop at EMNLP 2022
Figuring out and integrating lacking info is a vital activity for data graph completion to make sure robustness in the direction of downstream purposes akin to query answering. Including new info to a data graph in actual world system typically entails human verification effort, the place candidate info are verified for accuracy by human annotators. This course of is labor-intensive, time-consuming, and inefficient since solely a small variety of lacking info could be recognized. This paper proposes a easy however efficient human-in-the-loop framework for reality assortment that searches for a various set of extremely related candidate info for human annotation. Empirical outcomes introduced on this work demon- strate that the proposed resolution results in each enhancements in i) the standard of the candidate info in addition to ii) the flexibility of discovering extra info to develop the data graph with out requiring further human effort.
