基於社群網路分析的模式增強轉換網路;A Novel Social Network Prediction with Augmented Pattern-Based Transformer

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/93257

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93257

Title:	基於社群網路分析的模式增強轉換網路;A Novel Social Network Prediction with Augmented Pattern-Based Transformer
Authors:	凃家甄;Tu, Chia-Chen
Contributors:	資訊管理學系
Keywords:	社群網路分析;動態連結預測;基於特徵的轉換網路;Social Network Analysis;Dynamic Network Link Prediction;Pattern-Based Transformer
Date:	2023-07-24
Issue Date:	2024-09-19 16:50:55 (UTC+8)
Publisher:	國立中央大學
Abstract:	群體意識，是刻在血液裡的記憶。當古人由獨居走向部落，人際關係的概念在交流中形成，並藉由每一次的相互影響，逐漸深化。關係鏈的交織，讓獨立的小團體們發展成難以切割的社群網路。隨著網際網路的發展，社群網站上的分享逐漸取代了現實生活中的交流。倘若人們不再透過見面聯繫感情、認識新朋友不再僅限於線下活動，當人際互動逐漸由實體轉向虛擬，如果我們試圖了解一個人的行為模式與偏好，則應聚焦在社群平台上的社群網路分析。社群網路分析的首要目標，在於透過所有使用者在人際網路中的互動關係，識別出目標使用者在此研究範圍下的潛在關係，而其最常見的應用方法為連結預測。此外，由於關係網路會隨著時間演化，當我們在實踐連結預測時，應考慮其動態性，即動態連結預測。在過去的研究中，時間範圍的切割造成資訊量不足、圖像資料導致運算過於複雜，以及無法兼顧個人與整體關係網路資訊等問題，造成最終的預測效果不彰。基於以上的挑戰，本研究提出了以特徵進行分析的增強式轉換網路架構——NET-APT，以改善已知的分析劣勢，提高社群網路關係預測的成效。同時，我們也將NET-APT運用至真實世界的資料集中，並驗證其表現皆優於基準模型。對於此研究的未來展望，我們認為可以將NET-APT應用至商業領域，以其即時性的優勢結合基於特徵的分析模式，根據使用者當下的行為特質，向其發送可能會感興趣的文章乃至廣告、社團或粉絲專頁建議。;Throughout history, human beings′ social nature has fostered the development and deepening of interpersonal relationships through interactions and collaborations. With the advent of the Internet, people increasingly rely on virtual interactions and relationships, there is a need to shift our focus to social network analysis on social media platforms in order to uncover personal interaction patterns and preferences. The primary objective of social network analysis is to identify latent associations and connections within a social network, particularly with respect to a target user. This is commonly achieved through dynamic network link prediction techniques. However, past researches have faced challenges in achieving accurate predictions due to the lack of information caused by temporal splits, the computational complexity associated with graph data, and the failure to capture both personal and general network information. To address these challenges, we propose NET-APT, an augmented pattern-based Transformer architecture designed to enhance the effectiveness of social network prediction. We evaluate the performance of NET-APT using real-world datasets and compare it against baseline models, demonstrating its superior predictive capabilities. In terms of future development, we envision the application of NET-APT in commercial domains. For example, it could be utilized to send personalized advertisements based on users′ interaction patterns, enabling real-time pattern-based analysis. This potential application highlights the value of NET-APT in enhancing targeted marketing and providing tailored suggestions to users in various contexts.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	12	View/Open

社群 sharing

Loading...