Citation

Shannaq, Boumedyen and Farhan, Yasir Hadi and Tareq, Mustafa and Ali, Oualid and AlMaqbali, Said (2025) Overcoming Challenges in AI-Powered NLP Models: Enhancing the Capabilities of ChatGPT and DeepSeek. In: 2025 3rd International Conference on Cyber Resilience (ICCR), 03-04 July 2025, Dubai, United Arab Emirates.

Text 454.pdf - Published Version Restricted to Repository staff only Download (888kB)

Abstract

AI-driven Natural Language Processing (NLP) models have experienced substantial advancements, propelled by transformer architectures and pretraining methodologies. Notable examples include ChatGPT and DeepSeek, which have demonstrated exceptional performance in various tasks such as text generation, comprehension, and reasoning. Nevertheless, ChatGPT and DeepSeek lack factual consistency, contextual awareness, and interpretability, which implies persistent deficiencies. In this research, one critically examines the architectural points, operations, and limitations of these models. It presents a novel architecture based on a hybrid that combines symbolic reasoning, quantum-based attention, and domain-adaptive fine-tuning to improve the existing flaws. According to that, we have done a comparison analysis by ROUGE-L and BERTScore to assess the efficiency of AI-generated content. DeepSeek performed a little better than ChatGPT with the RoadUncover score of 0.91 and 0.47 ROUGE-L and BERTScore respectively, as compared to 0.89 and 0.43 in ChatGPT respectively. Also, DeepSeek demonstrated better facts and consistency (81.3%) and less hallucination (18.7%) as compared to ChatGPT (78.6 facts and consistency and 21.4 hallucination). These findings confirm the effectiveness of our cross-comparison approach and its capability to identify errors generated by AI. The study provides a viable method of validation of AI outcomes based on real data sets and may be used in the applications of audience profiling, influencer analysis and sentiment-based development of content in business, media analysts and researchers.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	ChatGPT, DeepSeek, AI-Powered NLP Models, Transformer Architectures, Text Generation, Text Comprehension.
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD28-70 Management. Industrial Management > HD30.2 Electronic data processing. Information technology. Including artificial intelligence and knowledge management
Divisions:	Others
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	19 Mar 2026 02:28
Last Modified:	19 Mar 2026 02:28
URII:	http://shdl.mmu.edu.my/id/eprint/15616

Downloads

Edit (login required)