下一個購物籃推薦旨在根據用戶過去的購物籃歷史序列資料，推薦用戶下一個購物籃的商品。然而，過去的購物籃推薦方法大多只是在學習用戶的購買歷史後，直接推薦預測的前 N 個商品，未考慮到商品之間的互補和替代關係。互補關係指的是兩個商品常常一起購買的關係，而替代關係則表示兩個商品之間可以相互替換。為了考慮商品的關聯性，過去的研究通常使用基於關聯規則的推薦系統，但這些方法僅考慮了商品之間的關聯，並未考慮到用戶購物籃之間的序列性關係和用戶的偏好。因此，本研究提出一種基於關聯的購物籃序列推薦模型ASBRec，使用基於序列的模型來構建購物籃推薦系統，以預測用戶在下一個購物籃中可能購買的商品。特別地，在模型的前端和後端加入了購物籃的關聯性考量。在前端，我們使用 Item2Vec 方法對商品進行編碼，將關聯性融入項目嵌入中。這樣做有助於捕捉商品之間的相似性和互補性。在後端，我們利用關聯規則建立信賴度矩陣，以獲得所有項目之間的信賴度。然後，我們將信賴度矩陣與輸出序列聚合，通過調整信賴度和輸出序列之間的權重，修正購物籃推薦的結果，並增強其關聯性。這種基於序列的購物籃推薦方法結合了商品關聯性和用戶偏好，能夠提供更精確和個性化的推薦結果。在本研究中，我們對一個真實世界的資料集進行了實驗，結果顯示我們提出的方法在推薦性能上優於先前的方法。;The next basket recommendation aims to recommend items for the user′s next basket based on their past basket history. However, previous basket recommendation methods often simply recommend the top N predicted items directly after learning from the user′s purchase history, without considering the complementary and substitution relationships between items. Complementary relationship refers to the frequent co-purchasing of two items, while substitution relationship indicates that two items can be interchangeably chosen. To consider the interplay between items, previous research has typically relied on association rule-based recommendation systems. However, these methods only consider item associations and do not consider the sequential relationship between user baskets and user preferences. Therefore, this study proposes an Association-based Sequential Basket Recommendation model, ASBRec, which utilizes sequence-based models to construct a basket recommendation system for predicting the items that users are likely to purchase in the next basket. Notably, we incorporate considerations of basket associations in both the frontend and backend of the model. At the frontend, we encode items using the Item2Vec method, incorporating associations into item embeddings. This helps capture the similarity and complementarity between items. At the backend, we utilize association rules to establish a confidence matrix that captures the confidence between all items. We then aggregate the confidence matrix with the output sequence, adjusting the weights between confidence and the output sequence to refine the basket recommendation results and enhance their associations. This sequence-based basket recommendation approach combines item relationships and user preferences, providing more accurate and personalized recommendations. In this study, we conducted experiments on a real-world dataset, and the results demonstrate that our proposed method outperforms previous approaches in terms of recommendation performance.
