Call for Papers -- Transfer Learning for Low-Resource Dialogue System Development

2024-08-23

Summary:

 Transfer learning has proven to be a successful approach in addressing the challenges of developing cognitive machine translation systems in low-resource environments. While existing methods often rely on shared target languages, linguistic similarities, or specific training techniques, they may not fully address the complexities of dialogue system development. Identifying user intentions and related categories is a crucial aspect of utterance processing in task-oriented collaborative AI systems. To tackle this, a proposed method involves treating a group of user information as the original sector and a single user's data as the target domain. This allows for the transfer of learning from the data itself to the objective field, providing a viable solution. Leveraging existing information in high-resource languages to train algorithms for low-resource languages is advantageous, as it significantly reduces the time required for data gathering. Overcoming the scarcity of dialogue data in low-resource languages is achieved through the use of importance-frequency and translating-confidence metrics in a word-level aligned technique that creates code-mixed information. Additionally, predictive spatial improvement employs word-encoding models and code-mixed samples to train a new conversation interpretation model, enhancing the overall performance of the system.

Developing a personalized task-based dialogue system poses challenges due to insufficient information gathered from each participant, leading to potential overfitting and difficulty adapting to diverse user demands with small datasets. The deficiency of heterogeneous training data further hampers algorithmic growth. However, employing simultaneous ambient information representations proves superior to cross-lingual stable integration in very low-resource environments. While new algorithmic generators demonstrate improved performance even with limited annotated data, constructing a system efficiently utilizing the maximum learnings from low-resource scenarios remains a critical concern. In such scenarios, a well-defined ontology becomes crucial, providing a data-structured depiction for the conversation system to discuss specific topics. However, gathering precisely specified semantic datasets for these systems is both costly and time-consuming. Despite achieving strong performance with sufficient labelled datasets in transfer learning approaches, models in lower resource settings often outperform their counterparts. Addressing this challenge involves training for domain adaptation and developing models tailored for low-resource training.

This special issue explores a personalized transfer learning method that selects distinct activities based on user needs. Empirical findings, utilizing both simulated and real-world data, demonstrate that the customized conversation system effectively selects optimal actions for various users, thereby enhancing the quality of discourse in personalized contexts.

Relevant topics include, but are not limited to, the following:

  • Low-resource linguistic generation using a variational framework for dialogue systems.
  • Contrary cross-lingual transfer learning in low-resource languages for module allocation.
  • Language-to-language transfer learning for international task-oriented dialogue systems.
  • A collection of task-oriented conversation technologies for low-resource languages across various domains.
  • The impact of sentiment analysis through transfer learning in resource-constrained environments.
  • Low-resource Natural Dialogue Systems: Content Synthesis and unattended Module Detection.
  • Indirect curriculum-based learning from multiple sources for low-resource dialogue generation.
  • An overview of current approaches for natural language processing in low-resource settings.
  • Multi-faceted cross-lingual transfer learning using common and specialized communication abilities.
  • Focus-driven bilingual guidance for zero-shot, task-oriented, cross-lingual dialogue systems
  • Directional Dialogue Mechanism for Multiple Categories based on Transfer Learning.

 

Guest Editors:
Prof. Mahmud Iwan Solihin, Faculty of Engineering, UCSI University, Malaysia
Prof. Lin Guoping, Department of Industrial Engineering and Enterprise Information, Tunghai University, Taiwan
Prof. Slamet Riyadi, Department of Information Technology, Universitas Muhammadiyah Yogyakarta, Indonesia

 

Deadlines:
Article Submission Deadline: January 10, 2025
Authors Notification Date: March 20, 2025
Revised Papers Due Date: May 20, 2025
Final notification Date: July 20, 2025