許永真Hsu, Jane Yung-jen臺灣大學:資訊工程學研究所吳冠鋆Wu, David Kuan-ChunDavid Kuan-ChunWu2010-05-172018-07-052010-05-172018-07-052009U0001-1408200901595100http://ntur.lib.ntu.edu.tw//handle/246246/183353近年來數位內容的蓬勃發展為網路資訊世界帶來了變革，人們需要擷取的資訊量日益龐大，也因此造就了資訊氾濫的問題。傳統上，資訊氾濫的處理方法皆採用推薦系統研究的思維，也就是利用一些預測喜好的經驗法則或機率模型來過濾出對使用者有興趣的資訊。然而，這些方法用在社群媒體上的推薦行為時卻經常不適用。在本論文中，社群媒體推薦系統指的是那些能夠提供真人使用者透過主動推薦的方式達到與他人分享及討論內容並進一步產生社群互動的系統。在這些系統中，單純呈現出符合使用者喜好的內容有可能會阻礙使用者與持相反意見者的互動機會，進而降低使用者對系統的滿意度。本論文中，我們認為社群媒體推薦系統上的資訊氾濫問題除了喜好預測外，還可以由另一個不同面向來嘗試解決。相較於喜好預測，我們提出一個解讀社群媒體上的推薦行為的方法，而這些解讀可以幫助使用者快速地並有效地自行決定哪些推薦的內容是值得進一步消費的。我們的實驗結果顯示這些自動產生出的解讀除了可以增加使用者對系統的信賴感，也的確可以提升滿意度。此外，我們的實驗結果也提供了未來研究者一組可供比較與參考的數據。Historically the information overload problem is addressed with approaches rooted in recommender systems research, that is, functional prediction of user preference using ad hoc heuristics or probabilistic models. These approaches have been shown toe effective to a certain extent, but they fail to address different needs required by users when dealing with social media recommendation systems. These are systems in which content is shared and discussed among community users in the form of social recommendations. On these systems, presentation of only the most preferencematching items may hinder user interaction and demote user satisfaction. This thesis contributes by offering a different perspective into tackling the information overload problem on social media recommendation systems. Instead of preference prediction, we propose an approach to annotating social media recommendations with meaningful explanations. These explanations provide some context to assist the user effectively and efficiently decide which recommendations are of interest. Our experimental results empirically demonstrate that such contextual explanations not only harvest user’s trust in the system but also promote overall user satisfaction. This thesis also contributes by providing our results as a set of benchmarks for comparisons with future research.Acknowledgments iiibstract vist of Figures xiiiist of Tables xivhapter 1 Introduction 1.1 Problem Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Reader’s Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6hapter 2 Background 7.1 Explanations in Historic Intelligent Systems . . . . . . . . . . . . . . 7.2 Early Explanations in Recommender Systems . . . . . . . . . . . . . 8.3 Explanations in Knowledge-based Recommender Systems . . . . . . 11.4 Transparent and Justified Explanations in Recommender Systems . . . 13.5 Trust Building with Explanation Interfaces . . . . . . . . . . . . . . . 14.6 Features and Personalization in Tagsplanations . . . . . . . . . . . . 15.7 Limitations of Existing Approaches . . . . . . . . . . . . . . . . . . 17hapter 3 Study Hypotheses 19.1 Persuasiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19.2 Effectiveness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.4 Justification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24.5 Trust . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25.6 Satisfaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26hapter 4 Contextual Explanations 27.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 27.2 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30.3 Proposed Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3.1 Solution Outline . . . . . . . . . . . . . . . . . . . . . . . . 32.3.2 Topic Generation . . . . . . . . . . . . . . . . . . . . . . . . 33.3.2.1 Data Strcture for Bookkeeping . . . . . . . . . . . 34.3.3 Incremental Updates . . . . . . . . . . . . . . . . . . . . . . 35.3.4 Explanation Generation . . . . . . . . . . . . . . . . . . . . 35.3.4.1 Sharer Reliability . . . . . . . . . . . . . . . . . . 37hapter 5 Experimental Design and Results 41.1 Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . 42.1.1 Recaholic System: The Experiment . . . . . . . . . . . . . . 42.1.2 Experimental Conditions . . . . . . . . . . . . . . . . . . . . 44.1.3 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46.1.3.1 Rating Task . . . . . . . . . . . . . . . . . . . . . 46.1.3.2 Survey . . . . . . . . . . . . . . . . . . . . . . . . 48.1.4 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49.1.4.1 Dependent Variables . . . . . . . . . . . . . . . . . 50.2 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.1 Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.2 The Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . 53.2.3 Effects of Contextual Explanations on Persuasiveness . . . . . 56.2.4 Effects of Contextual Explanations on Effectiveness . . . . . 58.2.5 Effects of Contextual Explanations on Efficiency . . . . . . . 60.2.6 Effects of Contextual Explanations on Justification . . . . . . 61.2.7 Effects of Contextual Explanations on Trust . . . . . . . . . . 62.2.8 Effects of Contextual Explanations on Satisfaction . . . . . . 63hapter 6 Concluding Remarks 65.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . 65.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67ibliography 69application/pdf3346684 bytesapplication/pdfen-US社群媒體推薦系統社群推薦社群媒體推薦推薦解讀Social Media Recommendation SystemSocial RecommendationsContextual ExplanationsInformation Overload ProblemSocial Media Sharingthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/183353/1/ntu-98-R96922075-1.pdf