A BMRC Algorithm Based on Knowledge Enhancement and Case Regularization for Aspect Emotion Triplet Extraction

Aspect Emotion Triplet Extraction (ASTE) is a vital branch in the field of NLP that strives to identify triads consisting of aspect words, opinion words, and emotional polarity from sentences. Each triplet contains three elements, namely, aspect words, opinion words and their emotional polarity. How to effectively extract attribute and opinion words and detect the connection between them is extremely important. Early work often focused on only one of these tasks and could not extract triples simultaneously in the same framework. Recently, researchers proposed using the bidirectional machine reading comprehension model (BMRC) to obtain triples of aspects, opinions, and emotions in the same framework. However, the existing BMRC model ignored professional domain knowledge, and the equal treatment training ignored the special contribution of individual instances in sentences to sentences. To this end, this paper proposes a BMRC with knowledge enhancement, which integrates the knowledge of specialized fields into the model, strengthens the connection between specific fields and open fields, increases instance regularization, and pays attention to the contribution of individual instances to sentences. Our experiments on several benchmark data sets have shown that our model has reached the most advanced performance.