基於超圖與知識圖譜協同的降噪會話推薦;Denoising Session Recommendation Based on the Collaboration of Hypergraphs and Knowledge Graphs

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/95490

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95490

Title:	基於超圖與知識圖譜協同的降噪會話推薦;Denoising Session Recommendation Based on the Collaboration of Hypergraphs and Knowledge Graphs
Authors:	何佳馨;He, Jia-Xin
Contributors:	資訊工程學系
Keywords:	推薦系統;知識圖譜;超圖;自監督學習;Recommendation;Knowledge Graph;Hypergraph;Self-Supervised Learning
Date:	2024-07-16
Issue Date:	2024-10-09 16:54:08 (UTC+8)
Publisher:	國立中央大學
Abstract:	基於會話的推薦系統（Session-based Recommendation，SBR）專注於預測短時間內的項目點擊。由於這種情況大多沒有使用者的歷史資訊，因此需要透過會話內隱含的用戶意圖進行推薦，除了用戶意圖外，也可以使用知識圖譜幫助項目學習額外的資訊，但現有的模型存在以下問題：1. 項目交互主導模型 2. 知識圖譜包含噪音 3.會話內部隱含噪音。由於以上問題，在本文中，我們提出了三個通道（知識圖譜通道、會話超圖通道以及會話線圖通道）分別用來捕捉知識圖譜、會話內部及會話外部的關聯訊息，在知識圖譜通道中，我們會自適應的刪除多餘的邊，幫助模型減少知識圖譜帶來的噪音，且知識圖譜形成的項目表示將與會話超圖通道形成的表示共同進行推薦預測，以緩解項目交互佔主導的問題，此外，我們還會為會話內的項目生成各自的注意力，用來進行會話的降噪。最後我們會透過會話產生的超圖通道和線圖通道最大化雙方的互訊息，來作為改善推薦任務的輔助任務。實驗結果顯示，在大多數據集中我們的方法可以提升推薦準確度。;Session-based recommendation systems need to capture implicit user intents from sessions, but existing models suffer from issues like item interaction dominance, noisy knowledge graphs, and session noise. We propose a multi-channel model with a knowledge graph channel, session hypergraph channel, and session line graph channel to capture relevant information from knowledge graphs, within sessions, and beyond sessions respectively. In the knowledge graph channel, we adapt ively remove redundant edges to reduce noise. Knowledge graph representations cooperate with hypergraph representations for prediction to alleviate item dominance. We also generate in-session attention for denoising. Finally, we maximize mutual information between the hypergraph and line graph channels as an auxiliary task. Experiments show our method improves recommendation accuracy on most datasets.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	19	View/Open

社群 sharing

Loading...