曹承礎Seng-Cho, Chou臺灣大學:資訊管理學研究所王宇頎Wang, Yu-chiYu-chiWang2010-05-052018-06-292010-05-052018-06-292008U0001-1512200708561400http://ntur.lib.ntu.edu.tw//handle/246246/179850隨著網際網路的發展以及音樂創作的普及，音樂推薦系統成為主要發展中的應用服務之一，嘗試提供符合使用者需求或心情的音樂。為了達成這個目標，傳統的推薦技術被廣泛的使用在這個領域，大部分現有的音樂推薦系統專注在探索使用者偏好、META-DATA、聆聽紀錄、以及音樂的內容來產生可能讓使用者滿意的個人化音樂推薦功能。而，聆聽紀錄是一種主題式的心理認知經驗，這種經驗會在特殊的時間點與個人意向高度相關，因此，情境因素諸如時間、地點、氣候、與溫度等常被納入推薦系統中來增加推薦結果的精確性；心理因素是另一個影響使用者對推薦結果滿意度的重要因素。鑑於此，本研究結合音樂聆聽者的情感因素與情境資訊，先依據Kate Hevner的情緒循環模型、ConceptNet的語意網、以及音樂學原理，計算在使用者、情緒、與情境等因素間的相似度，做為共通性的音樂基礎；再依照使用者的音樂偏好、聆聽音樂時的行為、以及使用者回報的資訊，透過以使用者為基礎的協同過濾演算法，找出不同使用者對音樂的個人差異，來建構一更為符合使用者情境與情緒因素的音樂推薦系統。Music recommendation systems are emerging applications that attempt to provide music to suit users’ needs or moods. To achieve this goal, traditional recommendation techniques are widely used in this field. Most of the music recommendation system exploits user interest, metadata, listening history, and audio signals of music to generate a personalized function that can predict songs the user may like.owever, listening experience is a type of subjective cognitive experience that is highly dependent on the individual’s intention at a particular time. Thus, contexts such as time, location, weather, and temperature have been added to systems to improve their accuracy. Psychological influences represent another important aspect that determines the user’s satisfaction with the recommended results.n the proposed approach, listeners’ emotional information is used in conjunction with context information. We first gather the explicit similarity between human, emotion, context, and music based on Kate Hevner’s Adjective Cycle, the semantic network of ConceptNet, and musicology as the common fundamental. Then, we adjust the individual differences according to the user’s musical taste, listening behavior, and feedback through user-based collaborative filtering in order to generate a more individual intentional music recommendation system.THESIS ABSTRACT IIST OF CONTENTS VIST OF FIGURES VIIIST OF TABLES IX INTRODUCTION 1.1 MOTIVATION 1.1.1 Finding suitable music 1.1.2 Context-aware Influences 2.1.3 Individual Intentions 2.1.4 Music Recommendations 3.2 CONTRIBUTION & APPROACH 4.3 THESIS STRUCTURE 5 RELATED WORK 7.1 MUSIC RECOMMENDATION SYSTEM 7.1.1 Hybrid of Content and Metadata 7.1.2 Social Music Networking 9.1.3 Emotional Playlist 11.2 CONTEXT-AWARE COMPUTING 11.2.1 Definition of Context-aware 12.2.2 Context Factors 12.2.3 Context-aware Models 13.3 EMOTIONS AND MUSIC 14.3.1 Definition of Emotions 15.3.2 Emotional Models 15.3.3 Relationship between Music and Emotion 18.4 RECOMMENDATION ALGORITHMS 20.4.1 Content-based Filtering 20.4.2 User-based Collaborative Filtering 21.4.3 Item-based Collaborative Filtering 22.4.4 Model-based Collaborative Filtering 22.4.5 Hybrid Method 23.5 COMMONSENSE REASONING 23.5.1 Cyc 24.5.2 Open Mind Common Sense 24.5.3 ConceptNet 25.5.4 WordNet 26 SYSTEM DESIGN 27.1 SYSTEM CONCEPT 27.2 DESIGN GOAL 28.3 OUR APPROACH 29.4 SYSTEM ARCHITECTURE 31.4.1 User Interface Module 32.4.2 Query Analyze Module 33.4.3 Context Module 34.4.4 Emotion Module 36.4.5 Song Info Module 38.4.6 User Profile Module 39.4.7 Recommendation Module 39 EXPERIMENT AND ANALYSIS 45.1 EXPERIMENT ENVIRONMENT 45.2 EXPERIMENT PROCESS 45.2.1 Data Collection and Processing 47.2.2 Online Experiment 47.2.3 Experiment Evaluation 49.3 EXPERIMENT RESULT 54.3.1 Music Source 54.3.2 Statistic of User Profile 55.3.3 Collected Context Data 56.3.4 Rating Result 58.4 SYSTEM EVALUATION 60.4.1 Recommendation Accuracy 60.4.2 User Satisfaction 63.4.3 Comparison with Last.FM 66 CONCLUSION 69.1 CONTRIBUTION 69.2 LIMITATIONS 70.2.1 The Limitation of Recommendation 71.2.2 Limited Data with Fixed Factors 71.3 FUTURE WORK 71 BIBLIOGRAPHY 73application/pdf3093401 bytesapplication/pdfen-US推薦系統情緒協同過濾音樂情境recommender systememotioncontextcollaborative filteringmusichttp://ntur.lib.ntu.edu.tw/bitstream/246246/179850/1/ntu-97-R95725024-1.pdf