🤖 AI Summary
This work investigates the degradation of recurrent neural network (RNN) performance in intent detection under class-imbalanced data, a phenomenon whose underlying mechanism remains poorly understood. Leveraging dynamical systems theory, we model the evolution of sentence representations in the RNN hidden state space as trajectories and reveal that the network performs intent classification through clustering on low-dimensional manifolds. We propose a novel framework—decoupling geometric separation from readout alignment—and, for the first time, elucidate from a dynamical systems perspective how class imbalance disrupts the geometric structure of hidden states. On the SNIPS dataset, clear clustering structures emerge, whereas in ATIS, clusters corresponding to low-frequency intents significantly deteriorate, thereby uncovering how data distribution shapes the network’s computational solution.
📝 Abstract
Intent detection, a fundamental text classification task, aims to identify and label the semantics of user queries, playing a vital role in numerous business applications. Despite the dominance of deep learning techniques in this field, the internal mechanisms enabling Recurrent Neural Networks (RNNs) to solve intent detection tasks are poorly understood. In this work, we apply dynamical systems theory to analyze how RNN architectures address this problem, using both the balanced SNIPS and the imbalanced ATIS datasets. By interpreting sentences as trajectories in the hidden state space, we first show that on the balanced SNIPS dataset, the network learns an ideal solution: the state space, constrained to a low-dimensional manifold, is partitioned into distinct clusters corresponding to each intent. The application of this framework to the imbalanced ATIS dataset then reveals how this ideal geometric solution is distorted by class imbalance, causing the clusters for low-frequency intents to degrade. Our framework decouples geometric separation from readout alignment, providing a novel, mechanistic explanation for real world performance disparities. These findings provide new insights into RNN dynamics, offering a geometric interpretation of how dataset properties directly shape a network's computational solution.