On Attribution of Recurrent Neural Network Predictions via Additive Decomposition

TL;DR

This paper introduces REAT, a novel additive decomposition method that enhances the interpretability of RNNs by providing faithful, phrase-level attribution scores for predictions, applicable across various RNN architectures.

Contribution

The paper proposes REAT, a flexible attribution method that decomposes RNN predictions into additive contributions of words and phrases, improving interpretability and faithfulness over existing approaches.

Findings

REAT provides faithful, interpretable attributions for RNN predictions.

The method is applicable to various RNN architectures including GRU and LSTM.

Analysis reveals linguistic knowledge captured by RNNs and potential for debugging.

Abstract

RNN models have achieved the state-of-the-art performance in a wide range of text mining tasks. However, these models are often regarded as black-boxes and are criticized due to the lack of interpretability. In this paper, we enhance the interpretability of RNNs by providing interpretable rationales for RNN predictions. Nevertheless, interpreting RNNs is a challenging problem. Firstly, unlike existing methods that rely on local approximation, we aim to provide rationales that are more faithful to the decision making process of RNN models. Secondly, a flexible interpretation method should be able to assign contribution scores to text segments of varying lengths, instead of only to individual words. To tackle these challenges, we propose a novel attribution method, called REAT, to provide interpretations to RNN predictions. REAT decomposes the final prediction of a RNN into additive contribution of each word in the input text. This additive decomposition enables REAT to further obtain phrase-level attribution scores. In addition, REAT is generally applicable to various RNN architectures, including GRU, LSTM and their bidirectional versions. Experimental results demonstrate the faithfulness and interpretability of the proposed attribution method. Comprehensive analysis shows that our attribution method could unveil the useful linguistic knowledge captured by RNNs. Some analysis further demonstrates our method could be utilized as a debugging tool to examine the vulnerability and failure reasons of RNNs, which may lead to several promising future directions to promote generalization ability of RNNs.