Online Review System Using Relational Triple Extraction with Novel Data Augmentation Methods

Abstract

The online review system is a fundamental application system. However, the existing system not only can not automatically check the matching of comments and score ratings but also can not give a fair reference score according to the comments, In this work, we utilize a relational triple extraction method to solve the two problems for the first time. In addition, considering that the existing online review systems are generally characterized by a lack of high-quality labeled data, we present five novel data augmentation techniques for boosting performance specifically on relational triple extraction tasks. The five data augmentation techniques demonstrate particularly strong results for both datasets of the review system and the public datasets of relational triple extraction.