Scholarlink

<p>Conversational assistants (CAs), particularly task-oriented ones, are designed to engage users in natural language interactions, assisting them in completing specific tasks or accessing relevant information. These systems leverage advanced natural language understanding (NLU) and dialogue management techniques to interpret user inputs, infer intentions, and respond effectively. Over time, CAs have diversified, serving sectors such as e-commerce, health care, and tourism. Central to these systems is intent detection (ID), a core process that identifies the user’s primary goal based on their utterances. This paper explores the evolution of ID methodologies, transitioning from recurrent neural networks (RNNs) to transformer-based models, analyzing their performance, limitations, and prospective advancements. Additionally, it highlights hybrid approaches, emerging paradigms, and the broader implications of these innovations for next-generation human–computer interaction systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From RNNs to Transformers and Beyond: a Deep Dive into Intent Detection in Goal-oriented Conversational Agents

  • 期刊论文

  • 发表日期：

    2025年6月7日

查看全文
  • Mourad Jbene,
  • Abdellah Chehri,
  • Rachid Saadane,
  • Smail Tigani,
  • Gwanggil Jeon

摘要

Conversational assistants (CAs), particularly task-oriented ones, are designed to engage users in natural language interactions, assisting them in completing specific tasks or accessing relevant information. These systems leverage advanced natural language understanding (NLU) and dialogue management techniques to interpret user inputs, infer intentions, and respond effectively. Over time, CAs have diversified, serving sectors such as e-commerce, health care, and tourism. Central to these systems is intent detection (ID), a core process that identifies the user’s primary goal based on their utterances. This paper explores the evolution of ID methodologies, transitioning from recurrent neural networks (RNNs) to transformer-based models, analyzing their performance, limitations, and prospective advancements. Additionally, it highlights hybrid approaches, emerging paradigms, and the broader implications of these innovations for next-generation human–computer interaction systems.