Advanced Data Science and Computational Intelligence

(Annual (One issue per year))

ABSTRACT

This theoretical article proposes an LLM-assisted multimodal hybrid recommendation framework for the Amazon Electronics domain. Traditional recommendation methods that rely primarily on user–item interactions often degrade under sparse-data and cold-start conditions. The proposed architecture combines three complementary sources of information: Alternating Least Squares (ALS) collaborative factors, lightweight MiniLM-based semantic embeddings derived from product meta-data and reviews, and visual product features projected into the collaborative embedding space. These signals are integrated through a weighted hybrid scoring mechanism intended to preserve behavioural personalisation while improving semantic and multimodal coverage. The article synthesises 21 recent studies on collaborative, LLM-based, multimodal, graph-based, and hybrid recommendation, identifies unresolved integration and deployment gaps, and specifies a methodology for future implementation and empirical evaluation using Precision@K, Recall@K, NDCG@K, and cold-start cohort analysis. The con-tribution is therefore a research hypothesis and system design rather than a report of completed experimental results.

Keywords: Recommender Systems; Collaborative Filtering; Large Language Models; Multimodal; Semantic Embeddings; MiniLM; ALS; Cold-Start; Amazon Electronics