ProtoTEx: Explaining Model Decisions with Prototype Tensors

TL;DR

ProtoTEx is a new interpretable NLP classification model that uses prototype tensors to explain decisions, matching state-of-the-art accuracy and aiding non-experts in recognizing propaganda.

Contribution

Introduces ProtoTEx, a prototype network-based architecture with a novel training algorithm for explainable NLP classification.

Findings

ProtoTEx achieves accuracy comparable to BART-large and surpasses BERT-large.

ProtoTEx provides faithful, interpretable explanations for its decisions.

User study shows prototype explanations improve propaganda recognition by non-experts.

Abstract

We present ProtoTEx, a novel white-box NLP classification architecture based on prototype networks. ProtoTEx faithfully explains model decisions based on prototype tensors that encode latent clusters of training examples. At inference time, classification decisions are based on the distances between the input text and the prototype tensors, explained via the training examples most similar to the most influential prototypes. We also describe a novel interleaved training algorithm that effectively handles classes characterized by the absence of indicative features. On a propaganda detection task, ProtoTEx accuracy matches BART-large and exceeds BERT-large with the added benefit of providing faithful explanations. A user study also shows that prototype-based explanations help non-experts to better recognize propaganda in online news.