Predicting Document Coverage for Relation Extraction

Abstract This paper presents a new task of predicting the coverage text document for relation extraction (RE): Does contain many relational tuples given entity? Coverage predictions are useful in selecting best documents knowledge base construction with large input corpora. To study this problem, we present dataset 31,366 diverse 520 entities. We analyze correlation features like length, entity mention frequency, Alexa rank, language complexity, and information retrieval scores. Each these has only moderate predictive power. employ methods combining statistical models TF-IDF BERT. The model BERT, HERB, achieves an F1 score up to 46%. demonstrate utility on two use cases: KB claim refutation.