DialogConv: A Lightweight Fully Convolutional Network for Multi-view Response Selection

• DOI: 10.18653/v1/2022.emnlp-main.828
• Publication Date: 2023-08-04T20:21:02Z
• Authors: Yongkang Liu; Shi Feng; Wei Gao; Daling Wang; Yifei Zhang
• Keywords: Benchmark (surveying); Convolution (computer science); Word embedding; Utterance

Current end-to-end retrieval-based dialogue systems are mainly based on Recurrent Neural Networks or Transformers with attention mechanisms. Although promising results have been achieved, these models often suffer from slow inference huge number of parameters. In this paper, we propose a novel lightweight fully convolutional architecture, called DialogConv, for response selection. DialogConv is exclusively built top convolution to extract matching features context and response. Dialogues modeled in 3D views, where performs operations embedding view, word view utterance capture richer semantic information multiple contextual views. On the four benchmark datasets, compared state-of-the-art baselines, average about 8.5x smaller size, 79.39x 10.64x faster CPU GPU devices, respectively. At same time, achieves competitive effectiveness