A Neuro Symbolic AI Conversational Recommender System for Recommending a Workout Plan - Dalam bentuk buku karya ilmiah

WIDI SAYYID FADHIL MUHAMMAD

Informasi Umum

Kode

25.05.949

Klasifikasi

006.3 - Special Computer Methods- Artificial intelligence

Jenis

Karya Ilmiah - Thesis (S2) - Reference

Subjek

Artificial Intelligence In Healthcare

Dilihat

36 kali

Informasi Lainnya

Abstraksi

Physical inactivity remains a global health challenge, with existing digital fitness applications often failing to provide guidance that is simultaneously personalized, safe, and engaging. Current recommender systems face a critical dilemma: ontology-based systems are secure but rigid, while LLM-based conversational systems are engaging but unreliable. To address this gap, this study proposes and evaluates a neuro-symbolic framework that integrates the logical consistency of ontology-based reasoning with the conversational fluency of Large Language Models (LLMs). The core of this research is a controlled comparative experiment between a baseline model (Gemma3-Ground) and the proposed model (Gemma3-Chase), which is fine-tuned on a corpus containing explicit, inferred reasoning traces generated by SWRL rules. Results from a multi-faceted evaluation using 20 synthetic user profiles show that Gemma3-Chase consistently and significantly outperforms Gemma3-Ground across key metrics, particularly in Effectiveness, Semantic Relevance, and Groundeness. The neuro-symbolic model also exhibited significantly lower standard deviation, indicating higher reliability. This study validates that embedding logical reasoning into the LLM fine-tuning process is an effective method for creating trustworthy and engaging conversational recommender systems for the health and wellness domain