Mining similar users using semantic properties of annotation tags

Abstract

在這篇論文中，提出了一個嶄新的方法來測量Location-Based Services(LBS)下的用戶間的相似度，此方法會利用到使用者和標註之地理資訊間的關係。本方法主要的概念為在收集資料之後，利用Singular Value Decomposition (SVD) 混合其他分解方法以找出標籤(tag)之間的語義相似性。其後，建構本篇論文所提出的User Attendance Graph (UAG)圖形架構，UAG代表了使用者的打卡紀錄和tag轉換的重要程度。最後，本篇論文提出一個新的演算法Sematic Behavior Similarity(SBS)，用於測量UAG之間的相似性。在實驗章節中，本篇採用Whrrl網站收集的真實使用情形之資料集為測試資料集，其後實驗中採用nDCG測量標準來評估本篇論文提出的演算法。結果顯示本方法能有效找出在LBS情況下的相似使用者，並且可以在不同的應用中使用，如，朋友推薦系統。

In this thesis work, I propose a novel method of measuring user similarity in Location-Based Services (LBS) via relationships between users and annotation tags of locations they attended. Collecting all check-in data together, I apply Singular Value Decomposition (SVD) and other decomposition methods in order to find semantic similarity between tags. Next, I propose an idea of User Attendance Graph (UAG) to represent user check-in history and describe importance of each tag together with transitions between them. Further, I propose a novel Semantic Behavior Similarity (SBS) algorithm for measuring likeness between UAG. I evaluated our approach with a real dataset collected from Whrrl website using nDCG measure. Results show efficiency of proposed method for finding LBS users with similar behavior, and it can be used in different applications, e.g. friend recommender systems.