Token-wise sentiment decomposition for ConvNet: Visualizing a sentiment classifier

• DOI: 10.1016/j.visinf.2020.04.006
• Publication Date: 2020-04-03T20:49:15Z
• Authors: Piyush Chawla; Subhashis Hazarika; Han-Wei Shen
• Keywords: Interpretability; Sentiment Analysis

Convolutional neural networks are one of the most important and widely used constructs in natural language processing AI general. In many applications, they have achieved state-of-the-art performance, with training time faster than other alternatives. However, due to their limited interpretability, less favored by practitioners over attention-based models, like RNNs self-attention (Transformers), which can be visualized interpreted more intuitively analyzing attention-weight heat-maps. this work, we present a visualization technique that understand inner workings text-based CNN models. We also show how method generate adversarial examples learn shortcomings data.