基於個性化項目頻率和時間注意力的下一個購物籃推薦混合模型;PIFTA4Rec: A Hybrid Model of Personalized Item Frequency and Temporal Attention for Next-basket Recommendation

NCU Institutional Repository > 管理學院 > 資訊管理研究所 > 博碩士論文 > Item 987654321/95464

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95464

题名:	基於個性化項目頻率和時間注意力的下一個購物籃推薦混合模型;PIFTA4Rec: A Hybrid Model of Personalized Item Frequency and Temporal Attention for Next-basket Recommendation
作者:	林莉庭;Lin, Li-Ting
贡献者:	資訊管理學系
关键词:	下一個購物籃推薦;個性化項目頻率;深度學習;K Nearest Neighbor;Item2Vec;Next Basket Recommendation;Personalized Item Frequency;Deep Learning;K Nearest Neighbor;Item2Vec
日期:	2024-07-09
上传时间:	2024-10-09 16:52:58 (UTC+8)
出版者:	國立中央大學
摘要:	在資訊量爆炸的數位時代中，推薦系統扮演著至關重要的角色。它們不僅緩解資訊過載的問題，還能顯著提升用戶體驗並增加電子商務的收入潛力。因此，推薦系統在多種場景中得到廣泛應用，尤其是在下一個購物籃推薦任務上。然而，傳統推薦方法傾向於僅關注短期的局部關係，而忽略了長期的全局關係。雖然基於神經網絡的推薦方法已改善了推薦性能，但仍存在一些侷限與進步空間，特別是在利用個性化項目頻率（PIF進行推薦。 PIF是一種基於用戶對商品重複購買行為的衡量指標，在多個推薦任務中已證明其有效性和價值。本研究認為，將這種個人化重複購買行為的向量模型結合進現有的神經網絡模型中，將有助於進一步提升推薦效率與準確性。有鑑於此，本研究提出了一個結合個性化項目頻率資訊與時間注意力機制的混合模型，稱為 PIFTA4Rec。該模型整合傳統資料探勘和當代深度學習技術的優勢，充分利用 KNN 的局部敏感性對帶有 PIF 資訊的向量數據點進行精確預測。同時，結合了考慮每次購買時間的時間注意力機制和 Transformer 的多頭注意力機制，以深入學習用戶與項目之間的複雜關聯資訊。最終，本模型整合兩個推薦預測過程的結果，以生成最終的推薦結果。在本研究中，我們將這樣的技術融合模型 PIFTA4Rec 在兩個真實世界資料集中進行了實驗。結果顯示，本研究提出之方法在推薦性能上優於現有的其他下一個購物籃推薦方法，顯著證實了本模型的有效性與穩定性。;In the digital age of information explosion, recommendation systems play a crucial role. They not only alleviate the problem of information overload but also significantly enhance user experience and increase the revenue potential of e-commerce. Consequently, recommendation systems are widely applied in various scenarios, especially in the task of predicting the next basket. However, traditional recommendation methods tend to focus only on short-term local relationships, overlooking long-term global interactions. Although recommendation methods based on neural networks have improved performance, there are still limitations and room for improvement, particularly in the use of Personalized Item Frequency (PIF). PIF, a metric based on the user′s repeated purchase behavior, has proven its effectiveness and value in numerous recommendation tasks. This study suggests that integrating this vector model of personalized repeated purchase behavior into existing neural network models will further enhance the efficiency and accuracy of recommendations. In light of this, the study proposes a hybrid model that combines Personalized Item Frequency information with a Temporal Attention mechanism, called PIFTA4Rec. This model integrates the advantages of traditional data mining and contemporary deep learning techniques, leveraging the local sensitivity of K Nearest Neighbor (KNN) to precisely predict vector data points containing PIF information. Simultaneously, it incorporates a temporal attention mechanism that considers the timing of each purchase, along with the multi-head attention mechanism of Transformers, to deeply learn the complex associative information between users and items. Ultimately, the model integrates the results of these two recommendation prediction processes to generate the final recommendation outcomes. In this study, the hybrid model PIFTA4Rec was experimentally tested on two real-world datasets. The results demonstrated that the proposed method outperformed other existing next-basket recommendation methods in terms of recommendation performance, significantly confirming the model′s effectiveness and stability.
显示于类别:	[資訊管理研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	64	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....