ESTD: Empathy Style Transformer with Discriminative Mechanism

• Published in: Advanced Data Mining and Applications Vol. 13726; pp. 58 - 72
• Main Authors: Zhang, Mingzhe, Yue, Lin, Xu, Miao
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer 2022; Springer Nature Switzerland
• Series: Lecture Notes in Computer Science
• Subjects: Empathy; Natural language processing; Text style transfer; Transformer
• ISBN: 3031221362; 9783031221361
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-22137-8_5

Language expressions without empathy can neither effectively convey the expresser’s concern and goodwill, but also have a negative effect on the emotional and mental health of the recipients of the information. Compared to harsh or aggressive expressions, expressions with a high empathetic level can produce positive emotions. Unfortunately, non-empathetic expressions are generated daily without intention, causing negative feelings. Existing work has achieved certain success on style transfer, however, there are still limitations in language style selection. This paper addresses this challenge by using a corpus with multiple language styles. To this end, we employ ESTD to transfer a lower-empathetic expression to a higher-empathic expression. Experimental results on empathy style transfer task shows that our model outperforms some currently available baseline methods.