1. Expanded Reproduction of Socially Shared Opinions via Qtag A presentation about a Master Thesis KAIST Graduate School of Culture Technology Affiliation Lee, Sung Eob Written & Presented by Han, ‘Steve’ SangKi Advisor Professor
2.
3.
4.
5.
6. From Previous Studies Preliminaries People tag to rate & express their opinions However, this kind of tags hardly shared among users Because, users tend to use diversified expressions
7.
8.
9. We propose Qtag Objective Qtag expands user experience of collaborative tagging Q stands for Qualitative, Qtag is made to… 1. To Rate and to express Opinions 2. To produce more sharable tags 3. To provide fast & intuitive interpretation
10. Simple Idea of Qtag Proposed Scheme Qtag is simple augmentation of Plus(+) and Minus(-) signs Positive Tags Positive Rating & Opinions Negative Tags Negative Rating & Opinions Neutral Tags Conventional Tags Style+ Sound+ KFed+ Talent- Life- music- Pop Music KFed Britney Simple
11. How Qtag Makes Tags more Sharable Proposed Scheme Qtag will reduce diversification of expression via formulated expressions (Qtag) Music-
12. Qtag Experiments Experiments In order to evaluate Qtag, we built a conceptual model 1042 distinct tags were tagged 4083 times by 126 participants A series of questionnaire is also conducted in order to elaborate Adoption and Satisfaction of Qtag In order to aggregate Qtags, four digital product reviews were chosen from ZDNet Korea (http://www.zdnet.co.kr/) and four articles concerning celebrities published on Joins (http://www.joins.com/), one of the major internet newspapers in Korea. 1 st Experiment Each participant is asked to tag conventionally & Qtag 2 articles & 2 reviews 2 nd Experiment Participants answered a series of questionnaires The same tagging interface is applied as del.icio.us ( Free form / Bag-model / *Blind Tagging )
13. Qtag Cloud (1) Experiments Total Tag Count(709) / Positive Tag Count(344) / Negative Tag Count(210) Total Tag Count Rating: +189 After conducting experiments, we visualized Qtag clouds A review article about Sanyo Xacti
14. Qtag Cloud (2) Experiments Total Tag Count(414) / Positive Tag Count(35) / Negative Tag Count(176) Total Tag Count Rating: -141 A celebrity article about Britney Spears After conducting experiments, we visualized Qtag clouds
15. Tagging vs. Qtagging (1) 1 1 2 Results Changes in quantity and quality of tags Shared tags increased, It elaborates that Qtag filters meta-noise Total Distinct Tags Total Shared 400 128 1841 Conventional Tagging 642 219 2242 Qtagging Increased By Tagging Frequency 60.0% 71.1% 21.8% Entropy of tag data increased However, shared tags increased 1 2
16. Tagging vs. Qtagging (3) Results 1 2 The difference of probabilities between QT & CT is calculated Calculating Normal Distribution for ‘Conventional Tagging’ Calculation for normal distribution Calculating Normal Distribution for ‘Qtagging’ Calculation for normal distribution Qtagging has absolutely higher 'Shared Tags' 2 1 Qtag improves tag sharing
17. Implementation & Contribution Conclusion Qtag expands user experience of collaborative tagging 2. Qtag encourages users to tag more sharable tags in case of rating & self-expression 3. Participants generally accepted & Qtag system 1. Proposed a new formulated means of tagging for rating and self-expression
18.
19. Collective Intelligence Preliminaries Nobody knows everything, but everybody knows something Collaborative Tagging & Qtagging is also tools for harnessing Collective Intelligence or Wisdom of crowds Collective Intelligence is a form of intelligence that emerges from the collaboration and competition of many individuals. Appears in a wide variety of forms of consensus decision making in Bacteria, Animals, Human and Computers
20. New Domain of Qtag Preliminaries Qtag can be a tool of extracting public opinions from comments We define this data aggregating as “Collective Opinion” Qtag will expend the domain of collective intelligence There must be dominant opinions about this article If public opinions could be revealed, more productive debate via VLSC would be possible However, readers put their eyes on comment lists about 3 to 15 seconds Current interface makes hard to extract dominant opinions among VLSC
21. Collective Opinion vs. Intelligence Preliminaries Collective Opinion is subsidiary to Collective Intelligence According to degree of interactivity, Collective Opinion is the link between Collective Intelligence & Wisdom of Crowds Collective Intelligence Collective Opinion Wisdom of Crowds Abstraction & Enhancement Abstraction & Grouping Social Proofing Problem Solving High degree of Interactivity Diversified degree of interactivity Minimum degree of Interactivity Independency Among Participants Similarity Data size matters / Shares similar process (Data Aggregation & Mining) Knowledge Production (Wikipedia) Aggregation of public opinions (Comments) Problem Solving (Recommendation) Applications
22. What is Collective Opinion Preliminaries Collective Opinion Collective Intelligence Collective opinion is mining public opinions from a very large scale conversation (Comments) Qtag will harness Collective Opinion & encourage debates among a Very Large Scale Conversation (VLSC) Problem Solving / Data Producing Oriented Omni-directional Approach Data Aggregation Oriented Multi-directional Approach
23. Sharing Opinions via Qtag Objective Qtag expands user experience of collaborative tagging Qtag is applied in order to enhance VLSC environment 1. To provide breadcrumbs to access old comments 2. To aggregate & represent public opinions 3. To provide guides for writing new comments
24. User Scenario Proposed Scheme Massing process for dominant public opinions may be similar to following example Qtags JYP Wondergirls- 가창력 - Qtags 박진영 Wondergirls+ 완소희 + Comments 난 상관없다 . 원더걸스 너무 좋아 . 특히 완소희… Comments JYP 의 원더걸스 노래 너무 못한다 . Comments 원더걸스가 뭐가 좋다고…짜증해서 태깅한다 Qtags JYP Wondergirls- Tell Me- Qtag may encourage debates & represent dominant public opinions for a very large scale conversation
25. 2 nd Set of Qtag Experiments Experiments This set of experiments will evaluate Qtag for accumulating & representing Collective Opinion In order to aggregate Qtags, A number of internet articles which can trigger debates will be picked from renowned sources Experiments is openly deployed to attract wide range of internet users 1 st Experiment Qtag will be collected and a set of questionnaire will be conducted 2 nd Experiment Participants will answer a series of questionnaires about usability Tagging interface will be different from the first set of experiments ( Free form / Bag-model / *Open Tagging / synonym) 1 st Experiments conducted for one week from 25 Nov to 2 Dec 12 popular contents are posted, 401 participants participated 1531 comments & 5014 times tagged
26. Implementation & Contribution Conclusion VLSC environment can be improved via Qtag 2. Public opinions among VLSC can be aggregated & represented by Qtag 3. We speculate that comment writing may assisted by Qtag 1. Access to old comments can be improved via Qtag (breadcrumbs)
29. An Extreme Case Tags for ratings and opinion-expression Is a common phenomenon
30. Tag Sharing Status We analyzed tag data from mar.gar.in. To elaborate whether tags for rating can be shared or not Introducing mar.gar.in, Korean replica of del.icio.us 3000 Registered Users 73,001 bookmarks 24,000 distinct tags
31. Tag Data Analysis Our tag classification is based on Golder & Huberman’s Sen et al. customized Golder & Huberman’s tag classifications for their own use. So did we Functions Tag Perform for bookmarks Classified by Golder and Huberman Types of Tags optimized for social shopping and networking Identifying What (or Who) it is about Identifying What It is Identifying Who Owns It Refining Categories Identifying Qualities or Characteristics Self Reference Task Organizing Identifying Tags Descriptive Tags Personal Tags
32. Sharable tag distribution We analyzed tag data which is shared by two or more users to elaborate whether tags for ratings and expressing opinion can be shared or not We assigned two coders, and they distributed tags manually and reached consensus for tags failed to yield agreement
33. Answer for 1 st Research Question RQ1. Are tags for rating and expressing opinions shared easily? Tags for rating and expression are hardly shared 4.8% Of sharable tags are “Descriptive Tags”
34. Qtag Design Basically Qtag system is not much different from current tagging system Qtag is consisted of four following components Featured Item or People A homogeneous kinds of products or people which shares the same qualities Related Contents Reviews or Articles about featured items (Actual Tagging Source) Qtags KFed - Music- Talent- Life- Dance+ Pop Music Fashion+ Dance Music Positive / Negative / Neutral Tags Qtag Counts Demonstrate overall ratings and reputation Positive Tag Neutral Tag Negative Tag 10 5 20 -10 (35)
35. Domains of Qtag A homogeneous product line-up & people which & who shares the same qualities for comparison Qtag may perform well for Social Shopping & SNS A homogeneous product line-up shares the same specifications CCD+ Grip- Dslr Weight- Possible Qtags People who shares similar qualities Look+ personality- Students Possible Qtags
36. Qtag May Control Wild Expressions of Tags Proposed Scheme The causes of expression may be limited Music Privacy Appearance Dance General Reputation Qtag will provide a guideline of formulated expression via augmenting signs
37. Tagging vs. Qtagging (2) Results When there are 642 distinct tags When there are 400 distinct tags A Simple speculation about the probability of sharing tags Qtag System Conventional Tagging System When the number of distinct tag increases, the probability of sharing tags among users decreases. The same number of participants 126 Personnel Although there were more distinct tags, There were more shared tags among Qtag dataset This proves Qtag provides more chance of sharing tags
38. Research Questions Will Qtag work? We will answer following research questions RQ1. Are tags for rating and expressing opinions shared easily? RQ2. Are users able to apply the Qtag without difficulties? RQ3. Are users able to interpret valid information from Qtag? RQ4. Are there changes in quantity and quality of tags? We analyzed conventional tag data & conducted experiments with a Qtag Conceptual model
39. Answer for 2 nd Research Question RQ2. Are users able to apply the Qtag without difficulties? From both aspects of understanding the concept of Qtag & the actual tag count, participants applied Qtag without difficulties The answer based on Questionnaire After a text-based introduction of Qtag was provided, participants responded on a five-point scale as to whether the concept of Qtag was easy to understand. 4.15 stars The answer based on aggregated tag data Comparison of tag counts between conventional neutral tags and tags augmented with positive(+) or negative(-) signs Augmented Tag Count Neutral Tag Count
40. Answer for 3 rd Research Question RQ3. Are users able to interpret valid information from Qtag? Participants generally interpret valuable information from Qtag Clouds Participants responded on a five-point scale as to whether Qtag clouds convey valuable information or not 3.59 stars Participants requested for better visualization of Qtag clouds
41. Answer for 4 th Research Question 1 1 2 RQ4. Are there changes in quantity and quality of tags? Shared tags increased, It elaborates that Qtag filters meta-noise Total Distinct Tags Total Shared 400 128 1841 Conventional Tagging 642 219 2242 Qtagging Increased By Tagging Frequency 60.0% 71.1% 21.8% Entropy of tag data increased However, shared tags increased 1 2
42. Extreme Tags Tags for ratings and opinion-expression Is a common phenomenon We often tag to rate & express
43.
44. Massing Snow into Balls Preliminaries Proposed Scheme Qtag is the tool for making dominant public opinion standout Snow = A Very Large Scale Conversation snowballs = Dominant Public Opinions Dominant public opinions can be more noticeable via Qtag (Qtag: Tagging as Means of Massing Public Opinions)
45. Tagging vs. Qtagging (2) Results Let be a total set such that each element (tagging object) has at least tags on it. Then we have a nested property such that In our case, the 'Total distinct Tag' set is , and 'Shared Tag' set is . And from the nested property, . The number of total distinct tags : (tags) The number of participants : (person) The average number of each participants tagging frequency : (tags/person)
46. Tagging vs. Qtagging (2) Problem Definition Q: What is the probability of any random object is also a element in ? Solution We will think about the complimentary case. What is the probability of with ? Then only one person should pick selected object and others don't.
47. Extreme Case Checking If ,then , : it means that if gets larger, less chance of 'Shared' case. If , then , : it means that if gets smaller, high chance of 'Shared' Tagging vs. Qtagging (2) From the extreme cases, we could see that the derivation of probability is reasonable. And since is constant for each selection in , we can see that distribution of number of shared tagging follows 'Binomial Distribution' and for large , it can be approximated by 'Normal Distribution‘ .
48. Checking ‘Conventional Tagging’ 1 2 Participants tagged average 14.6 tags for ‘Conventional Tagging Model’ The unshared tag production probability reaches near 1.0. This result means that in the case of random tag selection, most tags should be shared. In the random tag selection case, the number of shared tags supposed to be 383.3. Variance is calculated to calculate normal distribution. 3 4 1 2 3 4 Tagging vs. Qtagging (2) Then
49. Checking ‘Qtagging’ Then 1 2 3 4 Participants tagged average 17.6 tags for ‘Conventional Tagging Model’ The unshared tag production probability reaches near 1.0. This result means that in the case of random tag selection, 90% of tags should be shared. In the random tag selection case, the number of shared tags supposed to be 577.8. Variance is calculated to calculate normal distribution. 1 2 3 4 Tagging vs. Qtagging (2)
50. Calculating Normal Distribution for ‘Conventional Tagging’ Calculating Normal Distribution for ‘Qtagging’ 1 2 Tagging vs. Qtagging (2) Calculation for normal distribution Calculation for normal distribution Qtagging has absolutely higher 'Shared Tags' 2 1
Editor's Notes
Expanded Reproduction of Socially Shared Opinions via Qtag