labeling, groups, vocabulary, development.

Uploaded on:
Top: excellent, romantic comedy, entertaining, exaggerated, girlie motion picture. 22% of label applications ... it, it simply gives more data about the film's class. ...
Slide 1

labeling, groups, vocabulary, development Shilad Sen S. Lam, D. Cosley, A. Rashid, D. Frankowski, F. Harper, J. Osterhouse, J. Riedl GroupLens Research University of Minnesota Dept. of Computer Science

Slide 2

from labels: dayton, graffiti, ohio, spray_paint, tag Tagging Background A tag is a freestyle word or expression used to portray an item Although clients may tag for egotistical reasons, other individuals advantage from their labels Tag-based data can be more adaptable, instinctive and versatile than conventional ontologies

Slide 3

Example Vocabulary #1 FACTUAL

Slide 4

Example Vocabulary #2 SUBJECTIVE

Slide 5

Example Vocabulary #3 PERSONAL

Slide 6

Research Questions How do labeling vocabularies advance? Could we control vocabulary advancement? Are particular sorts of vocabularies pretty much important to the group?

Slide 7

MovieLens Started ~1995 Users rate motion pictures ½ to 5 stars Users get proposals

Slide 8

MovieLens Tagging Features Tagging Intro Page portrayal call-to-tag

Slide 9

MovieLens Tagging Features Movie Lists

Slide 10

MovieLens Tagging Features Adding Tags

Slide 11

Experimental Data 29 days 3366 clients 18.8% of clients labeled no less than one film 3,623 one of a kind label phrases … . Bringing about 11,433 label applications label application = <user, motion picture, tag phrase>

Slide 12

Tag Classes Factual Top: activity, show, comic drama, disney, teenager, james security 64% of label applications Subjective Top: great, romantic comedy, interesting, misrepresented, girlie motion picture 22% of label applications Personal Top: bibliothek, in netflix line, settled, dvd, my dvds 11% of label applications

Slide 13

Research Questions How do labeling vocabularies develop? Will we control vocabulary advancement? Are particular sorts of vocabularies pretty much profitable to the group?

Slide 14

How do Vocabularies Evolve? We consider two components in labeling conduct: Investment and propensity – People label comparatively to how they have labeled previously. Group Influence – People label likewise to how they see other individuals tag. (more intricate model in paper)

Slide 15

Tag Class Similarities

Slide 16

Research Questions How do labeling vocabularies develop? Could we control vocabulary development? Are particular sorts of vocabularies pretty much important to the group?

Slide 17

Can We Affect Vocabulary? Four trial assembles Each gathering\'s arrangement of labels were kept up independently Experimental gatherings use diverse label choice calculations to show group labels for a film.

Slide 18

Unshared Selection Algorithm Tags for Pulp Fiction in the unshared gathering: connected labels exemplary film fascinating better approach for film bibliothek NO people group labels showed

Slide 19

Shared Selection Algorithm Tags for Pulp Fiction in the mutual gathering: mate motion picture, Bruce Willis, exemplary (2), incredible motion picture, John Travolta, need to purchase, tarantino (2), tarantino is god (3), this is the most odd motion picture I ever enjoyed connected labels showed group labels mate motion picture, tarantino, this is the strangest motion picture I ever loved

Slide 20

Shared-Pop Selection Algorithm Tags for Pulp Fiction in the common pop gathering: 90s, wrongdoing (2), discourse, dramatization (2), amusingness, hit men (2), John Travolta (2), murder, Quentin Tarantino (4), Samuel L., Samuel L. Jackson, ultra-savagery, USA connected labels showed group labels Quentin Tarantino (4), dramatization (2), hit men (2)

Slide 21

Shared-Rec Selection Algorithm Tags for Pulp Fiction in the mutual rec bunch: Pulp Fiction Fight Club Reservoir Dogs The Godfather activity, satire, dvd (2), dvd-r, extraordinary, horde, strange, eccentric (2), dreamlike, tarantino (7), theater, vhs dreamlike strange tarantino dvd connected labels showed group labels tarantino, dvd, dreamlike

Slide 22

verifiable subjective individual Vocabularies Evolve Differently! Relative label class extents after some time: unshared shared-rec shared-pop

Slide 23

Research Questions How do labeling vocabularies develop? Can we control vocabulary advancement? Are particular sorts of vocabularies pretty much important to the group?

Slide 24

Vocabulary Value Post-test client overview (365 clients) Asked general inquiries concerning labeling I make my own particular labels to have a fabulous time. (concur/deviate) Overall, I think the labeling highlights help me express my sentiments. and so on… Asked clients about particular utilizations of labels to films: Overall I might want to see the label "telephone corner" for The Matrix. I think the tag "ultra-viciousness" helps me choose whether or not to watch Pulp Fiction . and so on…

Slide 25

Value of Factual Tags: Regarding the tag " anime" for " Spirited Away" : This is an incredible tag, since it has no inclination connected with it, it only gives more data about the class of the motion picture. Incredible tag.. Taggers concurred that accurate labels are: Useful generally speaking (56% assention) Useful for finding out about motion pictures (60%) Useful for discovering films (59%)

Slide 26

Value of Subjective Tags: Regarding the tag " strange" for "Everlasting Sunshine of the Spotless Mind" … "dreamlike" is fitting. what\'s more, from another client… … some person saying a motion picture is "funny" or "surreal" doesn\'t let me know anything - in light of the fact that I don\'t know whether their tastes are like mine. 44% understanding subjective labels were valuable, however… 87% assention clients\' own particular subjective labels communicated their conclusions.

Slide 27

Value of Personal Tags: Regarding the tag " My DVDs" for " Dawn of the Dead" : The individual who made this tag was either confused that the tag would be shared or couldn\'t have cared less. In any case, how can it help a solitary other individual on the planet to realize that this film is in his DVD accumulation? 17% assention that individual labels were helpful, yet.. 87% assention that individual labels accommodating in arranging for the taggers themselves

Slide 28

Conclusion Post test use: 33,238 label applications by 1759 clients Recap: Users label likewise to: How they have labeled in the past The people group labels they have seen Vocabulary development is an "unpredictable" procedure Different label classes have altogether different utilities

Slide 29

Open Questions/Future Work Are the label class contrasts between trial bunches deterministic? Would more unpredictable labeling models better catch labeling conduct? By what means can a framework recognize "great" labels and "awful" labels?

Slide 30

Thank You! John Riedl GroupLens research National Science Foundation (IIS 03-24851 and IIS 05-34420) MovieLens clients! reach me at

View more...