Hierarchical Interpretation of Neural Text Classification

Authors

  • Hanqi Yan University of Warwick
  • Lin Gui University of Warwick
  • Yulan He University of Warwick

Abstract

Recent years have witnessed increasing interests in developing interpretable models in Natural Language Processing (NLP). Most existing models aim at identifying input features such as words or phrases important for model predictions. Neural models developed in NLP however often compose word semantics in a hierarchical manner. Interpretation by words or phrases only thus cannot faithfully explain model decisions. This paper proposes a novel Hierarchical INTerpretable neural text classifier, calledHint, which can automatically generate explanations of model predictions in the form of label-associated topics in a hierarchical manner. Model interpretation is no longer at the word level, but built on topics as the basic semantic unit. Experimental results on both review datasets and news datasets show that our proposed approach achieves text classification results on par with existing state-of-the-art text classifiers, and generates interpretations more faithful to model predictions and better understood by humans than other interpretable neural text classifiers.

Author Biographies

  • Hanqi Yan, University of Warwick
    Hanqi Yan is currently working toward the Ph.D. degree with the Department of Computer Science, University of Warwick, UK. Her current research interests include natural language processing and information extraction, and he also focuses on several application fields, including emotion-cause pair extraction and explainable AI.
  • Lin Gui, University of Warwick
    Lin Gui received the Ph.D. degree in computer science and technology from the Harbin Institute of Technology, Shenzhen, China. He is currently a Marie Curie Research Fellow with the Department of Computer Science, University of Warwick, Coventry, U.K. His specific research interests include the development of text classification algorithms and the models for natural language understanding, sentiment analysis, stance detection, emotion cause detection, and topic modeling.
  • Yulan He, University of Warwick
    Yulan He is a Professor of Computer Science at the University of Warwick. She has published over 180 papers in the areas of natural language understanding, sentiment analysis and opinion mining, question-answering, topic/event extraction from text, biomedical text mining, and social media analytics. 

Published

2024-11-15