Tapping into the vast reservoir of insights from pre-established language models, my study introduces a novel strategy for recommendation engines. The primary challenge tackled is the initial data scarcity in recommendation scenarios, especially prevalent in nascent enterprises or platforms lacking substantial user activity logs. Instead of relying on past user-item interactions, my approach morphs the recommendation mechanism into a sentiment analysis of languages reflecting user characteristics and item features. Sentiment tendencies are deduced via prompt education. While recommendation tools are pivotal in guiding users towards content resonating with their preferences, formulating tailor-made tools becomes arduous without prior interactions. Earlier research predominantly revolves around initial scenarios for either users or items and these methods, relying on past interactions in the same field, do not address my concern. I also introduce a standard to evaluate my method in the context of initial data scarcity, with outcomes underscoring its potency. To my understanding, this research stands as a pioneering effort in confronting the challenges of recommendations without prior data.
