# Supervised Domain Enablement Attention for Personalized Domain Classification

## TL;DR

This paper introduces a supervised enablement attention mechanism for personalized domain classification that uses sigmoid activation and supervised learning to enhance attention expressiveness and improve classification accuracy.

## Contribution

It proposes a novel supervised enablement attention method utilizing sigmoid activation and self-distillation, advancing personalized domain classification techniques.

## Key findings

- Significant improvement in domain classification accuracy.
- Effective use of supervised attention with sigmoid activation.
- Enhanced expressiveness of attention weights.

## Abstract

In large-scale domain classification for natural language understanding, leveraging each user's domain enablement information, which refers to the preferred or authenticated domains by the user, with attention mechanism has been shown to improve the overall domain classification performance. In this paper, we propose a supervised enablement attention mechanism, which utilizes sigmoid activation for the attention weighting so that the attention can be computed with more expressive power without the weight sum constraint of softmax attention. The attention weights are explicitly encouraged to be similar to the corresponding elements of the ground-truth's one-hot vector by supervised attention, and the attention information of the other enabled domains is leveraged through self-distillation. By evaluating on the actual utterances from a large-scale IPDA, we show that our approach significantly improves domain classification performance.