본문 바로가기 주메뉴 바로가기
22대 대한민국 국회 국회정보길라잡이 국회의원검색
회원가입 로그인
HOME

정보검색

소장정보 검색

국회도서관 홈으로 정보검색 소장정보 검색

결과 내 검색

동의어 포함

고급검색

인명/단체명

저자정보 상세정보
인명/단체명을 입력하세요.

키워드

  • 대표어

  • 외국어

-

목차보기

Title Page

Abstract

Contents

1. Introduction 13

1.1. Research motivation and objective 13

1.2. Research issues 14

1.2.1. Issue 1: Deriving implicit rating information 14

1.2.2. Issue 2: Integrating implicit rating-based CF and SPA 15

1.2.3. Issue 3: Integrating multiple perspectives of similarity concept 16

1.2.4. Issue 4: Identifying more reliable neighbors 17

1.2.5. Issue 5: Finding different neighbors for each different target item 18

1.2.6. Issue 6: Utilizing temporal information 19

1.3. Organization 21

2. Literature review on recommendation systems 22

2.1. General classification of recommendation systems 22

2.1.1. Content-based filtering (CBF) 22

2.1.2. Collaborative filtering (CF) 24

2.1.3. Rule-based approach 26

2.2. Limitations of recommendation techniques 27

3. Proposed approach to resolving research issues 1 and 2 33

3.1. Overall framework 33

3.2. Collaborative filtering (CF) - based recommendation 34

3.2.1. Deriving implicit ratings of users on items 34

3.2.2. Calculating similarity score based on implicit ratings 36

3.2.3. Selecting neighbors 38

3.2.4. Calculating CF-based predicted preference (CFPP) 38

3.3. Sequential pattern analysis (SPA) - based recommendation 39

3.3.1. Deriving sequential patterns 39

3.3.2. Matching subsequences of a target user with derived sequential patterns 39

3.3.3. Calculating SPA-based predicted preference (SPAPP) 40

3.4. Integrating CF-based and SPA-based Recommendations 40

3.5. Recommending items 41

3.6. An illustration with an example 41

4. Experiments for research issues 1 and 2 45

4.1. Experimental design 45

4.2. Experimental results and analysis 47

4.2.1. Experiment 1: Similarity function 47

4.2.2. Experiment 2: Minimum support 51

4.2.3. Experiment 3: Weight adjustment 52

4.2.4. Experiment 4: Hybrid of CF and SPA 53

4.3. Discussions 54

5. Proposed approach to resolving research issues 3 to 6 56

5.1. Overall framework 56

5.2. Building user profiles 58

5.3. Calculating similarity scores and selecting neighbors 59

5.3.1. Integrating similarity scores 59

5.3.2. Identifying more reliable neighbors 61

5.3.3. Selecting different neighbors for each different target item 62

5.4. Predicting preference scores on items to recommend 66

5.5. Combining predicted preference scores, considering time periods 67

5.6. Recommending items 69

6. Experiments for research issues 3 to 6 70

6.1. Experimental design 70

6.2. Experimental results and analysis 73

6.2.1. Experiment A. 73

6.2.1.1. Experiment A.1: Similarity integration 74

6.2.1.2. Experiment A.2: Relative degree of match (DOM) 76

6.2.1.3. Experiment A.3: Similarity integration and DOM 78

6.2.1.4. Experiment A.4: Different neighbors of a target user for each different target item 81

6.2.1.5. Experiment A.5: Similarity integration, DOM, and different neighbors of a target user for each different target item 84

6.2.2. Experiment B. 88

6.2.2.1. Experiment B.1: Temporal information 88

6.2.2.2. Experiment B.2: Temporal information, similarity integration, DOM, and different neighbors of a target user for each target item 90

6.3. Discussions 92

7. Conclusions 95

References 97

List of Tables

Table 1. Example of item profiles used in CBF recommendation systems 23

Table 2. Example of a user profile used in CBF recommendation systems 23

Table 3. Example of similarity between a target user and items to recommend 23

Table 4. Example of user-item rating matrix used in CF recommendation systems 25

Table 5. Example of similarity between a target user and other users 25

Table 6. Example of ratings of a target user on items to recommend 25

Table 7. Summary of techniques used in recommendation systems 32

Table 8. An example of implicit rating data derived from an original transaction data 42

Table 9. The integration of results from CF-based and SPA-based recommendation system 42

Table 10. Example of user-item rating matrix 60

Table 11. Example of similarity measure and rank, depending on similarity function 60

Table 12. Example of user-item rating matrix 65

Table 13. Example of similarity and rank, depending on similarity function 66

Table 14. Example of similarity between a target item and other item 66

Table 15. Temporal weight matrix 67

List of Figures

Fig. 1. Overall framework of HOPE system 34

Fig. 2. Different perspectives of three similarity functions 38

Fig. 3. Dataset partitioning 46

Fig. 4. Comparison among CF-based recommendation systems using different similarity function, when the number of neighbors is set to 3 50

Fig. 5. Comparison among SPA-based recommendation systems, each of which is implemented with different minimum support 51

Fig. 6. Adjustment of integration weight 52

Fig. 7. Comparison among three approaches 54

Fig. 8. Overall framework of the proposed recommendation system 58

Fig. 9. Comparison among systems in groups 1-A and 1-B 76

Fig. 10. Comparison among systems in groups 1-A and 1-C 78

Fig. 11. Comparison among systems in groups 1-C and 1-D 80

Fig. 12. Comparison in MAE among systems in groups A to D 80

Fig. 13. Comparison among CF_P and systems in group 1-E 83

Fig. 14. Comparison in MAE among systems in groups 1-A, 1-E, 1-F, and 1-G 84

Fig. 15. Comparison among CF_I(D), CF_P_P, and CF_I(D)_I(D) 86

Fig. 16. Comparison in MAE among CF_I(D), CF_P_D, and CF_I(D)_I(D) 87

Fig. 17. Comparison among systems in groups 2-A and 2-B 90

Fig. 18. Comparison among systems in groups 2-B and 2-C 92

초록보기

 In the age of information overload, information about products and services on the Internet is growing explosively and, as a result, people have difficulties in processing the overwhelming amount of information. As a means to lessen this problem, personalized recommendation systems have been introduced by many studies. The personalized recommendation can be seen usually as a specific kind of information filtering which enables people to filter out unnecessary and uninteresting information. Thanks to the personalized recommendation systems, people can find what they want immediately without wasting their time, and sellers can increase their revenue by providing people with the items they are likely to purchase. Since it is known that high accuracy of the personalized recommendation systems contributes to improve customers' repurchase intention, it is necessary and important to improve the accuracy of the system. So far, a number of studies have provided fundamental knowledge and techniques for developing the recommendation systems and enhancing the accuracy of them. However, there is still much room for improving the accuracy of recommendation systems. Instead of focusing on the issues addressed by many other studies, this paper aims to propose novel approaches to improving the accuracy of recommendation systems, focusing on the following six research issues which previous studies rarely have addressed: 1) deriving implicit rating information, 2) integrating collaborative filtering (CF) and sequential pattern analysis (SPA), 3) integrating multiple perspectives of similarity concept, 4) reflecting relative degree of match, 5) finding different neighbors of a target user for each different target item, and 6) utilizing two kinds of temporal information, purchasing time and item launch time. Through a series of experiments using the transaction data of an online shopping mall in Korea and MovieLens data, this paper verified that taking into account the above six issues when developing recommendation systems is worthwhile and significantly improves the accuracy of recommendation systems.

추천서가 (다양한 추천 자료를 만나보세요)

권호기사보기

권호기사 목록 테이블로 기사명, 저자명, 페이지, 원문, 기사목차 순으로 되어있습니다.
기사명 저자명 페이지 원문 기사목차

권호

기사명 원문보기 기사목차
닫기