Mutual character dialogue generation with semi‑supervised multitask learners and awareness

Khalaf, Ayesheh Ahrari and Hassan Abdalla Hashim, Aisha and Olowolayemo, Akeem (2024) Mutual character dialogue generation with semi‑supervised multitask learners and awareness. International Journal of Information Technology, 16 (3). pp. 1357-1363. ISSN 2511-2104 E-ISSN 2511-2112

	PDF - Published Version Restricted to Registered users only Download (1MB) | Request a copy
Preview	PDF - Supplemental Material Download (187kB) | Preview

Abstract

Consistent efforts have been ongoing to improve the friendliness and reliability of informal dialogue systems. However, most research focuses solely on mimicking human-like answers. Therefore, the interlocutors’ awareness features of the dialogue system are left unexplored. Meanwhile, cognitive science research reveals that awareness is a crucial indicator of an effective, high-quality informal conversation. This research aims to boost the quality of the conversational generation system by factoring in awareness of the interlocutors in the design and training of the dialogue system model. The Generative Pre-Trained Transformer-2 (GPT-2) model was implemented into the Persona Perception(P2) Bot to achieve the objectives of this study. This was to precisely develop model’s understanding, P2 Bot was implemented using a transmitter–receiver-based structure. The P2 Bot leverages mutual persona awareness to improve the quality of customized dialogue generation. GPT-2 is a1.5B parameter transformer model that produces state-ofthe-art accuracy in a zero-shot setting on seven of the eight evaluated language modeling datasets. The observations of the proposed model on a sizable open-source dataset, PERSONA-CHAT, proved successful, with improvement above the state-of-the-art baselines in both automatic measures and human assessments. The model has achieved 82.2% accuracy on Hits@1 performance metrics in the original data and 68.8% on the revised data. On the human evaluation the model scored an average of 2.66, pointing out that the responses provided were coherent and informative. A dialogue generation model with character and awareness which can communicate like an informative human expert was introduced. This study presents the submerging of GPT-2 model on a mutual persona perception dialogue generating model.

Item Type:	Article (Journal)
Uncontrolled Keywords:	Dialogue generation · Conversational agent · Cognitive science · Natural language understanding (NLU) · Generative Pre-trained Transformer 2 (GPT-2)
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800 Electronics. Computer engineering. Computer hardware. Photoelectronic devices > TK7885 Computer engineering
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Engineering > Department of Electrical and Computer Engineering Kulliyyah of Information and Communication Technology > Department of Computer Science Kulliyyah of Information and Communication Technology > Department of Computer Science Kulliyyah of Engineering Kulliyyah of Information and Communication Technology Kulliyyah of Information and Communication Technology
Depositing User:	Dr Akeem Olowolayemo
Date Deposited:	18 Mar 2024 15:32
Last Modified:	18 Mar 2024 15:32
URI:	http://irep.iium.edu.my/id/eprint/107097

Actions (login required)

View Item