保持內容一致性的學習方法：跨模態和跨項目的表示學習

Abstract

這篇論文的重點是解決推薦系統中學習內容一致性的挑戰。我們提出了一種新穎的模型，旨在學習跨模態和跨項目的表示，有效捕捉相似項目內容的文本和視覺語義。我們在這項研究中將嵌入應用於推薦系統和主題生成。廣泛實驗在三個真實的亞馬遜數據集上的結果表明，與現有的知名模型相比，在這兩個應用中都取得了顯著的改善。

The paper focuses on tackling the challenge of learning content consistency in recommender systems. We introduce a novel model that aims to learn cross-modal and cross-item representations, effectively capturing the textual and visual semantics of similar item contents. We apply the embedding to the recommender system and topic generation in this research. The results of extensive experiments on three real Amazon datasets show significant improvement in both applications, compared to existing well-known models.