Weakly Supervised Relation Extraction

Relation extraction is crucial for many natural language processing applications, such as question answering and text summarization. Although there are several different approaches for relation extraction, most use a supervised learning approach, which requires a large training dataset. These extensive datasets must be hand-labeled by experts, which makes the annotation process time-consuming and expensive. Another approach utilized in this study is weak supervised relation extraction. By using weak supervised learning, the cost of training data labeling can be reduced. In this study, we propose a weakly supervised relation extraction approach inspired by another weakly supervised model named REPEL. In both REPEL and our relation extraction approach, extraction patterns are derived from unlabeled texts using the given relation seed examples. To extract more useful extraction patterns, we introduced the use of labeling functions in our method. These labeling functions consist of simple rules to analyze the candidate pattern’s syntax and these labeling functions help extract more confident candidate patterns. Our proposed method was tested on the same dataset used by REPEL to compare our results with those obtained by REPEL. Both systems require several relation seed examples to learn patterns from unlabeled data. When fewer relation seed examples were used, our method significantly outperformed REPEL. In the experimental tests, our approach was approximately 15 times more successful than REPEL with few relation seeds. Even with more relation seeds, our approach remains more successful.