ScholarWorks@숙명여자대학교

초록

As personalized consumption trends spread in modern society, companies are producing products that provide more diverse choices to meet the diverse needs of consumers. However, as the number of choices increases, consumers experience difficulties in the selection process, which creates a contradiction. Therefore, in this study, we propose a real-time product recommendation service based on image analysis and Retrieval-Augmented Generation(RAG). The proposed system recognizes an image using the GPT-4o model, and then utilizes RAG to suggest the most appropriate choice among the recognized choices. By comparing the existing Optical Character Recognition(OCR) model and the GPT-4o model, we selected GPT-4o, which is superior in contextual understanding and execution time, and presented the maximum label recognition limit by evaluating the recognition accuracy according to the number of labels of the GPT-4o model. This allows consumers to quickly and accurately find the optimal choice and is expected to contribute to improving AI-based label recognition rate in the future based on the results of label recognition experiments.

키워드