在電子商務中，推薦系統所扮演的角色，其重要性不言而喻。然而，在傳統的基於Session或序列推薦模式中，通常使用先前的項目來推薦可能出現的下一個項目。這種方法面臨一個主要的挑戰，那就是人們傾向於重複購買。當用戶喜歡某項商品時，他們很可能再次選擇同樣的商品，這導致推薦系統多數推薦的項目其實已在之前的序列中出現過。 然而，本論文認為一個準確、與吸引力兼具的推薦系統應該向用戶推薦他們還未發現但可能會喜歡的產品，一旦用戶知曉該產品，他們就有可能進行購買。因此，在預備資料時，本文會根據每個用戶的目標項目，將序列中先前出現過的目標項目全部刪除，確保用戶對該項目一無所知，而我們所提出的模型將在此情況下對目標項目進行預測。 本文根據序列中相鄰項目的後續項目概念，設計了一個深度網路架構，此架構能夠整合局部與全局資訊，以實現成功預測目標項目的目標。我們已對多個真實世界資料集的實驗進行廣泛的測試，結果顯示本文提出的深度學習架構在性能上優於幾種最先進的推薦方法。 ;In e-commerce, the importance of recommendation systems is increasing day by day. However, in traditional session-based or sequence-based recommendation modes, previous items are usually used to recommend the possible next item. This approach faces a significant challenge as people tend to repeat purchases. When users like a certain item, they are likely to choose the same item again, resulting in the majority of items recommended by the recommendation system having already appeared in previous sequences. However, this paper argues that an attractive recommendation system should recommend products to users that they don′t know but might like, and once they are aware of such products, they may choose to purchase them. Therefore, when preparing the data, this paper will delete all the target items that have previously appeared in the sequence based on each user′s target item, ensuring that users are unaware of the item. The model proposed in this paper will then predict the target item under this circumstance. This paper designs a deep network architecture based on the concept of the subsequent item of adjacent items in the sequence, incorporating both global and local information to achieve the goal of successfully predicting the target item. Through experiments with multiple real-world datasets, we conducted extensive testing, and the results show that the deep network architecture proposed in this paper outperforms several state-of-the-art recommendation methods in performance.
