Trust-based Collective View Prediction

150 Pages·2013·3.9 MB·English

by  Tiejian Luo

Tiejian Luo · Su Chen
Guandong Xu · Jia Zhou
Trust-Based
Collective View
Prediction
Trust-Based Collective
View Prediction
Tiejian Luo Su Chen
• •
Guandong Xu Jia Zhou
•
Trust-Based Collective
View Prediction
123
Tiejian Luo Guandong Xu
Jia Zhou Universityof Technology
UniversityofChineseAcademyofSciences Sydney, NSW
Beijing Australia
People’s Republic ofChina
SuChen
China MobileResearch Institute
Beijing
People’s Republic ofChina
ISBN 978-1-4614-7201-8 ISBN 978-1-4614-7202-5 (eBook)
DOI 10.1007/978-1-4614-7202-5
SpringerNewYorkHeidelbergDordrechtLondon
LibraryofCongressControlNumber:2013934720
(cid:2)SpringerScience+BusinessMediaNewYork2013
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation,broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionor
informationstorageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purposeofbeingenteredandexecutedonacomputersystem,forexclusiveusebythepurchaserofthe
work. Duplication of this publication or parts thereof is permitted only under the provisions of
theCopyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
CopyrightClearanceCenter.ViolationsareliabletoprosecutionundertherespectiveCopyrightLaw.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publicationdoesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexempt
fromtherelevantprotectivelawsandregulationsandthereforefreeforgeneraluse.
While the advice and information in this book are believed to be true and accurate at the date of
publication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityfor
anyerrorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,with
respecttothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
To wife Xia Chen and daughter Chenxi Luo
From Tiejian
To wife Ms. Jane Zhu, father Mr. Sanqin
Chen, and mother Ms. Xingqiao Luo
From Su
To wife Feixue Zhang and son Jack Xu
From Guandong
Preface
The task of collective view prediction is to reason about the opinion of an indi-
vidual to an item by calculating his or her relevant online community’s attitudes.
More than 20 years’ explorations in this field have made progress in developing
preciseandrobustmodelsandrelatedalgorithmsforpredictingcollectiveviewin
online community. The researchers also learned the current methods’ advantage
and understood their limitations. Our research presents a new perspective and
ideastoaddressthelowperformanceandrobustnessincollectiveviewprediction
tasks.Theapplicationsofthemodels,methods,andalgorithmsinthisbookwillbe
promising and valuable for improving the quality of online information recom-
mendationservices,targetingadvertisementdelivery,word-of-mouthanalysisand
so on.
Investigatingrelatedtheoryandengineeringpracticecanhelpusunderstandthe
prosandconsofconventionalmethodsincompletingthegroupopinionprediction
task. Recommendation methods and sentiment analysis are closely related to the
task. Collaborative filtering is such a typical approach. The advantage of this
method can generate personalized predictions without additional text analysis.
Those approaches have become mainstream models and methods for collective
view prediction in recent years. Early study indicated that the feasibility of col-
laborative filtering is solely based on the reliability of information resources.
However,intherealapplicationenvironment,thingsaregettingcomplicated.The
misconduct behaviors in rating could decrease the information reliability. Those
activities could make the prediction invalid and unreliable. To improve the pre-
diction accuracy andreducethe impact ofnoisedata, the theme ofthisbook isto
review the previous related theoretical foundation and propose a trust-based col-
lective view prediction model and relevant algorithms. Our study shows that
effective model for collective view prediction is attributed to users’ trust rela-
tionships network.
Asking appropriate research questions could motivate us to pursue the right
direction and address hard problems in the right way. From the theoretical per-
spective, this book re-examines the trust definition and quantitatively analyze the
relationship between user’s similarity and their trust network leads us to the right
solutions. From algorithm design perspective, one of the key questions is what
kinds of trust metrics strategies would impact the collective view prediction
vii
viii Preface
accuracy.Fromtheevaluationperspective,weestablishaframeworkforassessing
the model’s robustness and to formally describe the attacks aimed at trust-based
prediction algorithms.
This book studies on the linear correlation of trust and similarity, and the
influence of spread distance to the correlation. To explore the trust network, we
collectmorethan300,000users’datafromthepopularreviewwebsites.Thestudy
results indicate that users’ similarity on opinions is positively correlated to their
distance in trust network and negatively correlated to their trust value. We con-
clude two basic rules that are important to designing effective and efficient col-
lective view prediction algorithms. In order toanalyze how different trustmetrics
influence the prediction accuracy, we further elaborate on two well-known trust
metrics, and based on the new metrics we design new collective view prediction
algorithms. To further improve the accuracy of the trust-based prediction algo-
rithms, we propose a Bayesian fusion model for combining trust and similarity.
Moreover,asecond-orderMarkovrandomwalkmodelisproposedtoalleviatethe
sparse data problem in similarity measurement. These new approaches are more
accurate than the classical collaborative filtering algorithms in our experimental
evaluations.
Trust-based collective view prediction demonstrates more capability to resist
attacksoverthetraditionaltechniques.Buttherewerefewquantitativeanalysisof
this issue in previous studies. We build a robustness analysis framework to
measure the capability of trust-based prediction algorithms to resist attacks.
Simulation results using this framework reveal the key factors which impact the
robustnessoftrust-basedalgorithms,andconfirmthatthehonestusers’feedbacks
can help algorithm recover from attacks. We also give two strategies to improve
the algorithm robustness in real applications.
Keywords Collective view (cid:2) Trust metrics (cid:2) Trust network (cid:2) Social network (cid:2)
Sentimental analysis (cid:2) Recommendation (cid:2) Collaborative filtering
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research Theme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Scope and Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Recommendation Algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Content-Based Recommendation. . . . . . . . . . . . . . . . . . 13
2.1.2 Collaborative Filtering. . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.3 Hybrid Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Sentiment Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Dynamic Network Mechanism . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Statistic Characteristics . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Evolution Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3 Web Information Cascades. . . . . . . . . . . . . . . . . . . . . . 22
2.4 Trust Management. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Collaborative Filtering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 Neighborhood-Based Method . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 User-Based CF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 Item-Based CF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.1.3 Comprehensive Analysis of User-Based CF
and Item-Based CF. . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Latent Factor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.1 Singular Value Decomposition . . . . . . . . . . . . . . . . . . . 30
3.2.2 Regularized Singular Value Decomposition . . . . . . . . . . 31
3.3 Graph-Based Collaborative Filtering . . . . . . . . . . . . . . . . . . . . 32
3.3.1 Bipartite Graph Model of Collaborative Filtering . . . . . . 32
3.3.2 Graph-Based algorithm of Collaborative Filtering. . . . . . 32
ix
x Contents
3.4 Socialization Collaborative Filtering . . . . . . . . . . . . . . . . . . . . 33
3.4.1 Gathering Socialization Data . . . . . . . . . . . . . . . . . . . . 34
3.4.2 Neighborhood-Based Socialization Recommendation
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Graph-Based Socialization Recommendation
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Dynamic Model in Collaborative Filtering . . . . . . . . . . . . . . . . 38
3.5.1 Dynamic Neighborhood-Based Model. . . . . . . . . . . . . . 38
3.5.2 Dynamic Latent Factor Model . . . . . . . . . . . . . . . . . . . 39
3.5.3 Case Study: Dynamic Graph-Based
Collaborative Filtering. . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4 Sentiment Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.1 Sentiment Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.1 Dictionary-Based Opinion Words Generation. . . . . . . . . 55
4.1.2 Corpus-Based Opinion Words Generation . . . . . . . . . . . 56
4.2 Sentiment Orientation Classification . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Counting Opinion Words. . . . . . . . . . . . . . . . . . . . . . . 57
4.2.2 Supervised Learning Approaches . . . . . . . . . . . . . . . . . 58
4.3 Case Study: Sentimental Analysis in Recommender Systems . . . 59
4.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3.2 Related Work. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.3 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.4 Sentiment Enhanced Approach for Tag-Based
Recommendation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.5 Experimental Evaluation . . . . . . . . . . . . . . . . . . . . . . . 66
5 Theoretical Foundations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.1 Traditional Prediction Method. . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2 Trust and Its Measurement. . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.1 Data Source. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2.2 Statistical and Transmission Characteristics . . . . . . . . . . 76
5.3 Analysis of Trust and Collective View. . . . . . . . . . . . . . . . . . . 77
5.3.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.2 Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3.3 Indicator Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.3.4 Result Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6 Models, Methods and Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . 93
6.1 Theoretical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Contents xi
6.2 Trust-Based Prediction Methods . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.1 Dual Factor Trust Metric. . . . . . . . . . . . . . . . . . . . . . . 96
6.2.2 Single Factor Trust Metric. . . . . . . . . . . . . . . . . . . . . . 97
6.2.3 User Similarity Weighted. . . . . . . . . . . . . . . . . . . . . . . 99
6.3 Prediction Algorithm Based on Second-Order Markov Model. . . 100
6.3.1 Random Walk Model . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.3.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.4 Bayesian Fitting Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.5.1 Evaluation Measurement . . . . . . . . . . . . . . . . . . . . . . . 108
6.5.2 Trust-Based Algorithms. . . . . . . . . . . . . . . . . . . . . . . . 108
6.5.3 Second-Order Markov Model. . . . . . . . . . . . . . . . . . . . 110
6.5.4 Bayesian Fitting Model . . . . . . . . . . . . . . . . . . . . . . . . 112
6.5.5 Complexity Analysis of Algorithm . . . . . . . . . . . . . . . . 114
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7 Framework for Robustness Analysis. . . . . . . . . . . . . . . . . . . . . . . 117
7.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.3 Noisy Data Injection Strategy. . . . . . . . . . . . . . . . . . . . . . . . . 120
7.4 Attack Cost. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.5 Feedback Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.6 Application Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.7 Results Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145