Assessing the Impact of Sequence Length Learning on Classification Tasks for Transformer Encoder Models

TL;DR

This paper investigates how sequence length biases in Transformer models can lead to misleading classification results, especially in sensitive domains, and proposes methods to mitigate this issue.

Contribution

It empirically demonstrates the sequence length learning problem in Transformer classifiers and introduces approaches to reduce its impact on model predictions.

Findings

Sequence length can be exploited as a predictive feature.

Biases are more prevalent in private datasets like medicine and insurance.

Proposed mitigation strategies can lessen length-based biases.

Abstract

Classification algorithms using Transformer architectures can be affected by the sequence length learning problem whenever observations from different classes have a different length distribution. This problem causes models to use sequence length as a predictive feature instead of relying on important textual information. Although most public datasets are not affected by this problem, privately owned corpora for fields such as medicine and insurance may carry this data bias. The exploitation of this sequence length feature poses challenges throughout the value chain as these machine learning models can be used in critical applications. In this paper, we empirically expose this problem and present approaches to minimize its impacts.