Generative AI of Pinecone Vector Retrieval and Retrieval-Augmented Generation Architecture: Financial Data-Driven Intelligent Customer Recommendation System

Abstract

This study proposes a generative recommendation model based on Pinecone vector retrieval and Retrieval-Augmented Generation (RAG), designed for intelligent financial customer recommendation scenarios. Building upon traditional embedding retrieval and deep recommendation methods, the model incorporates multimodal vectorization strategies and the RAG architecture to achieve efficient recall of high-dimensional features and contextually enhanced generation. Experimental results demonstrate that the model achieves Precision@10, Recall@50, and NDCG@10 scores of 0.177, 0.436, and 0.463 respectively, representing improvements of approximately 12%, 11%, and 10% over BERT4Rec, exhibiting superior accuracy and interpretability.