Ana səhifə

Computing Point-of-View: Modeling and Simulating Judgments of Taste


Yüklə 0.6 Mb.
səhifə1/28
tarix26.06.2016
ölçüsü0.6 Mb.
  1   2   3   4   5   6   7   8   9   ...   28


Computing Point-of-View: Modeling and Simulating Judgments of Taste
by
Xinyu Hugo Liu
Sc.B., Massachusetts Institute of Technology (2001)

M.Eng., Massachusetts Institute of Technology (2002)


Submitted to the Program in Media Arts and Sciences,

School of Architecture and Planning,

in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Media Arts and Sciences
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2006
© Massachusetts Institute of Technology 2006. All rights reserved.

Author____________________________________________

Program in Media Arts and Sciences

5 May 2006

Certified by________________________________________

Pattie Maes

Associate Professor

Program in Media Arts and Sciences

Thesis Supervisor

Accepted by_______________________________________

Andrew B. Lippman

Chair, Department Committee on Graduate Students

Program in Media Arts and Sciences

Computing Point-of-View: Modeling and Simulating Judgments of Taste
by
Xinyu Hugo Liu
Submitted to the Program in Media Arts and Sciences,

School of Architecture and Planning,

on 1 May 2006, in partial fulfillment of the

requirements for the degree of

Doctor of Philosophy in Media Arts and Sciences
Abstract
People have rich points-of-view that afford them the ability to judge the aesthetics of people, things, and everyday happenstance; yet viewpoint has an ineffable quality that is hard to articulate in words, let alone capture in computer models. Inspired by cultural theories of taste and identity, this thesis explores end-to-end computational modeling of people’s tastes—from model acquisition, to generalization, to application—under various realms.
Five aesthetical realms are considered—cultural taste, attitudes, ways of perceiving, taste for food, and sense-of-humor. A person’s model is acquired by reading her personal texts, such as a weblog diary, a social network profile, or emails. To generalize a person model, methods such as spreading activation, analogy, and imprimer supplementation are applied to semantic resources and search spaces mined from cultural corpora. Once a generalized model is achieved, a person’s tastes are brought to life through perspective-based applications, which afford the exploration of someone else’s perspective through interactivity and play.
The thesis describes model acquisition systems implemented for each of the five aesthetical realms. The techniques of ‘reading for affective themes’ (RATE), and ‘culture mining’ are described, along with their enabling technologies, which are commonsense reasoning and textual affect analysis. Finally, six perspective-based applications were implemented to illuminate a range of real-world beneficiaries to person modeling—virtual mentoring, self-reflection, and deep customization.

Thesis Supervisor: Pattie Maes

Title: Associate Professor, Program in Media Arts and Sciences

Computing Point-of-View: Modeling and Simulating Judgments of Taste
by
Xinyu Hugo Liu

Thesis Committee:


Advisor___________________________________________

Professor Pattie Maes

Associate Professor of Media Arts and Sciences

Massachusetts Institute of Technology

Thesis Reader______________________________________

Professor William J. Mitchell

Alexander W. Dreyfoos, Jr. (1954) Professor

Professor of Architecture and Media Arts and Sciences

Director, Design Laboratory

Massachusetts Institute of Technology

Thesis Reader______________________________________

Professor Warren Sack

Assistant Professor of Film & Digital Media

University of California, Santa Cruz


Acknowledgements
If a point-of-view is interesting, it could only be due to the rich soil whence it grows. Most dearly, thank you Henry, Pattie, Ted, and Push (in memory) for believing in me through the years, for entrusting me with space to create and play in the deep end of the pool, and for giving me the important chances. Thanks Warren, Bill, Pattie, Henry, Walter, and Ted for critiquing versions of this document. Judith and Erik—thanks for generals and for lending me your encyclopedic insight on too many occasions.
To those who’ve inspired or supported my scholarship—Glorianna, Marvin, Jef, Bill, John, Roz, and Prof. Singer—thanks for your generosity. Grazie mille to— in presentia, Rada, Ian, Carson, Josiah, and Greg, and in absentia, Friedrich, Gilles, Erlend, Soren, Henri, Milan, Joseph, Marcus, Jean-Francois, Jacques, Jacques, Erich and Ludwig—for having advanced my point-of-view. To my professors, group mates, friends, detractors, and water-cooler mates at the Media Laboratory—you’re wonderful.
Thank you my N., my nothing-to-do-with-work super friends M., A., T., A., C. and A., J., L., and E. Mom and Dad—this is for you most.
Contents

Abstract 003

Aperitif 015
1 Introduction 016

1.1 Thesis summary 016

1.2 Roadmap and contributions 034
2 Approach 036

2.1 Related work in user modeling 036

2.2 Person modeling from everyday texts 039

2.3 Model acquisition 041

2.4 Model generalization 047

2.5 Model application 051


3 Techniques & Technologies 057

3.1 Reading for affective themes 0 057

3.2 Culture mining 0 067

3.5 Commonsense reasoning 071

3.6 Textual affect analysis 076
4 Acquisition systems: implementations & evaluations 082

4.1 Cultural taste realm: ‘taste fabric’ 082

4.2 Food realm: ‘synesthetic cookbook’ 090

4.3 Perception realm: ‘escada’ 092

4.4 Attitudes realm: ‘what would they think?’ 98

4.5 Humor realm: ‘catharses’ 106


5 Perspective-based applications 109

5.1 Art that’s always tasteful 109

5.2 Facilitating social introductions 115

5.3 An identity mirror 116

5.4 Virtual mentors, and pundits too 121

5.5 Foraging for food with the family 127

5.6 A jocular companion 129
6 Philosophical underpinnings 130

6.1 Bourdieu’s framework from taste 130

6.1 Post-structural aesthetics 132

6.2 The aesthetic consistency hypothesis 135

6.3 The stability of viewpoint 137

6.4 Dialogics 140


7 Conclusion 144
References 146

List of Figures



Figure 1 -1. Triptych summarizing the thesis’s approach to modeling persons in aesthetical realms.

20

Figure 1 -2. A walkthrough for cultural taste modeling

22

Figure 1 -3. Topological features in the learned taste fabric—identity hubs (left) and taste cliques (right)

24

Figure 1 -4. A walkthrough for generalizing attitudes

27

Figure 1 -5. The space of possible ways of perceiving

29

Figure 1 -6. Summary of built applications

33







Figure 2 -7. A semantic sheet representation

43

Figure 2 -8. A semantic diversity matrix

46







Figure 4 -9. Mining algorithm for cultural taste space

83

Figure 4 -10. Taste Fabric’s instance types and ontology sources

83

Figure 4 -11. Evaluation of the generalized cultural taste model, and the effect of identity hubs and taste cliques, in a complete recommendation task

89

Figure 4 -12. Results of ten-fold cross validation showing blog-level classification accuracies

96

Figure 4 -13. Learned feature weightings for single-scale classification.

97

Figure 4 -14. How reflexive memories get recorded from weblog excerpts

100

Figure 4 -15. Political viewpoints of major newspapers

102

Figure 4 -16. Evaluating model-based predictions of persons’ reactions to news stories.

105







Figure 5 -17. A semiotic model of aesthetic transaction

110

Figure 5 -18. The effects of perspective on artistic rendition—the words “sunset” (top-row) and “war” (bottom-row) are rendered with a Thinking-Seeing bias (left-column) versus with an Intuiting-Feeling bias (right-column)

114

Figure 5 -19. Aesthetiscope’s aesthetic impressions of the four season keywords (columns) rendered through the monadic optics of each Reader taken alone (rows)

115

Figure 5 -20, a-d. Reflections in the Identity Mirror—(clockwise from upper-left) as performer approaches the Identity Mirror, his reflection gains descriptive granularity, passing from subcultures (a), into genres and artists (b), into films and albums (c), into foodstuffs, activities, and songs (d)

118

Figure 5 -21. A panel of virtual AI mentors react affectively to a passage of text that the user is reading

122

Figure 5 -22. Clicking on virtual Roz Picard triggers an explanation dialog

123

Figure 5 -23. Evaluation of WWTT in a person-learning task

125

Figure 5 -24. Interactions with the Synesthetic Cookbook

128


List of Tables



Table 4 .1. Excerpt from Synesthetic Cookbook’s lists of ingredients and descriptive keywords

91

Table 4 .2. Political culture: attitudes of the Democratic and Republican parties

101







Table 5 .3. Signalling efficacy of Aesthetiscope’s five dimensions

112


  1   2   3   4   5   6   7   8   9   ...   28


Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©atelim.com 2016
rəhbərliyinə müraciət