서울대학교 중앙도서관
S-Space 소개
My S-Space
로그인이 필요합니다.
Login

S-Space

Publications

Detailed Information

User Profiling for Personalized Search & Partnership Match : 개인화 검색 및 파트너쉽 선정을 위한 사용자 프로파일링

DC Field Value Language
dc.contributor.advisor김홍기-
dc.contributor.author하시트-
dc.date.accessioned2017-07-14T05:49:33Z-
dc.date.available2017-07-14T05:49:33Z-
dc.date.issued2014-02-
dc.identifier.other000000017517-
dc.identifier.urihttps://hdl.handle.net/10371/125196-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 치의과학과, 2014. 2. 김홍기.-
dc.description.abstractThe secret of change is to focus all of your energy not on fighting the old, but on building the new. - Socrates

The automatic identification of user intention is an important but highly challenging research problem whose solution can greatly benefit information systems. In this thesis, I look at the problem of identifying sources of user interests, extracting latent semantics from it, and modelling it as a user profile. I present algorithms that automatically infer user interests and extract hidden semantics from it, specifically aimed at improving personalized search. I also present a methodology to model user profile as a buyer profile or a seller profile, where the attributes of the profile are populated from a controlled vocabulary. The buyer profiles and seller profiles are used in partnership match.

In the domain of personalized search, first, a novel method to construct a profile of user interests is proposed which is based on mining anchor text. Second, two methods are proposed to builder a user profile that gather terms from a folksonomy system where matrix factorization technique is explored to discover hidden relationship between them. The objective of the methods is to discover latent relationship between terms such that contextually, semantically, and syntactically related terms could be grouped together, thus disambiguating the context of term usage. The profile of user interests is also analysed to judge its clustering tendency and clustering accuracy. Extensive evaluation indicates that a profile of user interests, that can correctly or precisely disambiguate the context of user query, has a significant impact on the personalized search quality. In the domain of partnership match, an ontology termed as partnership ontology is proposed. The attributes or concepts, in the partnership ontology, are features representing context of work. It is used by users to lay down their requirements as buyer profiles or seller profiles. A semantic similarity measure is defined to compute a ranked list of matching seller profiles for a given buyer profile.		-
dc.description.tableofcontents1 Introduction 1
1.1 User Profiling for Personalized Search . . . . . . . . 9
1.1.1 Motivation . . . . . . . . . . . . . . . . . . . 10
1.1.2 Research Problems . . . . . . . . . . . . . . 11
1.2 User Profiling for Partnership Match . . . . . . . . 18
1.2.1 Motivation . . . . . . . . . . . . . . . . . . . 19
1.2.2 Research Problems . . . . . . . . . . . . . . 24
1.3 Contributions . . . . . . . . . . . . . . . . . . . . . 25
1.4 System Architecture - Personalized Search . . . . . 29
1.5 System Architecture - Partnership Match . . . . . . 31
1.6 Organization of this Dissertation . . . . . . . . . . 32

2 Background 35
2.1 Introduction to Social Web . . . . . . . . . . . . . . 35
2.2 Matrix Decomposition Methods . . . . . . . . . . . 40
2.3 User Interest Profile For Personalized Web Search Non Folksonomy based . . . . . . . . . . . . . . . . 43
2.4 User Interest Profile for Personalized Web Search Folksonomy based . . . . . . . . . . . . . . . . . . . 45
2.5 Personalized Search . . . . . . . . . . . . . . . . . . 47
2.6 Partnership Match . . . . . . . . . . . . . . . . . . 52

3 Mining anchor text for building User Interest Profile: A non-folksonomy based personalized search 56
3.1 Exclusively Yours' . . . . . . . . . . . . . . . . . . . 59
3.1.1 Infer User Interests . . . . . . . . . . . . . . 61
3.1.2 Weight Computation . . . . . . . . . . . . . 64
3.1.3 Query Expansion . . . . . . . . . . . . . . . 67
3.2 Exclusively Yours' Algorithm . . . . . . . . . . . . 68
3.3 Experiments . . . . . . . . . . . . . . . . . . . . . . 71
3.3.1 DataSet . . . . . . . . . . . . . . . . . . . . 72
3.3.2 Evaluation Metrics . . . . . . . . . . . . . . 73
3.3.3 User Profile Efficacy . . . . . . . . . . . . . 74
3.3.4 Personalized vs. Non-Personalized Results . 76
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . 80

4 Matrix factorization for building Clustered User Interest Profile: A folksonomy based personalized search 82
4.1 Aggregating tags from user search history . . . . . 86
4.2 Latent Semantics in UIP . . . . . . . . . . . . . . . 90
4.2.1 Computing the tag-tag Similarity matrix . . 90
4.2.2 Tag Clustering to generate svdCUIP and modSvdCUIP 98
4.3 Personalized Search . . . . . . . . . . . . . . . . . . 101
4.4 Experimental Evaluation . . . . . . . . . . . . . . . 103
4.4.1 Data Set and Experiment Methodology . . . 103
4.4.1.1 Custom Data Set and Evaluation Metrics . . . . . . . . . . . . . . . 103
4.4.1.2 AOL Query Data Set and Evaluation Metrics . . . . . . . . . . . . . 107
4.4.1.3 Experiment set up to estimate the value of k and d . . . . . . . . . . 107
4.4.1.4 Experiment set up to compare the proposed approaches with other approaches . . . . . . . . . . . . . . . 109
4.4.2 Experiment Results . . . . . . . . . . . . . . 111
4.4.2.1 Clustering Tendency . . . . . . . . 111
4.4.2.2 Determining the value for dimension parameter, k, for the Custom Data Set . . . . . . . . . . . . . . . 113
4.4.2.3 Determining the value of distinctness parameter, d, for the Custom data set . . . . . . . . . . . . . . . 115
4.4.2.4 CUIP visualization . . . . . . . . . 117
4.4.2.5 Determining the value of the dimension reduction parameter k for the AOL data set. . . . . . . . . . . . 119
4.4.2.6 Determining the value of distinctness parameter, d, for the AOL data set . . . . . . . . . . . . . . . . . . 120
4.4.2.7 Time to generate svdCUIP and modSvd-CUIP . . . . . . . . . . . . . . . . 122
4.4.2.8 Comparison of the svdCUIP, modSvd-CUIP, and tfIdfCUIP for different classes of queries . . . . . . . . . . 123
4.4.2.9 Comparing all five methods - Improvement . . . . . . . . . . . . . . 124
4.4.3 Discussion . . . . . . . . . . . . . . . . . . . 126

5 User Profiling for Partnership Match 133
5.1 Supplier Selection . . . . . . . . . . . . . . . . . . . 137
5.2 Criteria for Partnership Establishment . . . . . . . 140
5.3 Partnership Ontology . . . . . . . . . . . . . . . . . 143
5.4 Case Study . . . . . . . . . . . . . . . . . . . . . . 147
5.4.1 Buyer Profile and Seller Profile . . . . . . . 153
5.4.2 Semantic Similarity Measure . . . . . . . . . 155
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . 160
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . 162

6 Conclusion 164
6.1 Future Work . . . . . . . . . . . . . . . . . . . . . . 167
6.1.1 Degree of Personalization . . . . . . . . . . . 167
6.1.2 Filter Bubble . . . . . . . . . . . . . . . . . 168
6.1.3 IPR issues in Partnership Match . . . . . . . 169

Bibliography 170

Appendices 193
.1 Pairs of Query and target URL . . . . . . . . . . . 194
.2 Examples of Expanded Queries . . . . . . . . . . . 197
.3 An example of svdCUIP, modSvdCUIP, tfIdfCUIP 198		-
dc.formatapplication/pdf-
dc.format.extent6426947 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subjectUser Modelling-
dc.subjectUser Interests-
dc.subjectUser Preferences-
dc.subjectPersonalized Search-
dc.subjectPartnership Match-
dc.subject.ddc617-
dc.titleUser Profiling for Personalized Search & Partnership Match-
dc.title.alternative개인화 검색 및 파트너쉽 선정을 위한 사용자 프로파일링-
dc.typeThesis-
dc.contributor.AlternativeAuthorHarshit Kumar-
dc.description.degreeDoctor-
dc.citation.pagesxiv, 205-
dc.contributor.affiliation치과대학 치의과학과-
dc.date.awarded2014-02-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

Show Simple Item Record
Find it @ SNU

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share

qrcode
SNS Share