We consider this work as a first step towards the definition of Social Games With A Purpose: games that could take advantage of the specific properties of social networks.
Boost PC performance: How more available memory can improve productivity
Social Tagging Revamped: Supporting the Users' Need of Self-promotion through Social Filtering
1. Social Tagging Revamped
Supporting the Users’ Need of
Self-promotion through
Persuasive Techniques
Mauro Cherubini, Alejandro Gutiérrez (UIUC),
Rodrigo de Oliveira, and Nuria Oliver
2. How many ‘friends’ do you have on Facebook?
~30 ~100 ~300
information
overload
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
3. Facebook statistics
• 100 Million photos uploaded to the site each day
• More than 5 billion pieces of content (web links, news
stories, blog posts, notes, photo albums, etc.) shared each
week
• Average user has 130 friends on the site
• Average user clicks the Like button on 9 pieces of content
each month
• Average user writes 25 comments on Facebook content
each month
source: Facebook
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
10. social mechanisms
social
filtering
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
11. social mechanisms
social
filtering
social navigation
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
12. problem space
probability to miss
relevant content
auto. filtering
auto. tagging
social filtering
social navigation
effort required from user
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
13. problem space
probability to miss
relevant content
auto. filtering
auto. tagging
social filtering
social navigation
social tagging
effort required from user
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
14. social tagging?
the activity of producing collaboratively metadata
in the form of keywords to shared content
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
15. social tagging?
the activity of producing collaboratively metadata
in the form of keywords to shared content
the activity of producing with peers in a SN metadata in
the form of descriptors to shared content
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
16. social tagging?
the activity of producing collaboratively metadata
in the form of keywords to shared content
the activity of producing with peers in a SN metadata in
the form of descriptors to shared content
- mutual modeling
- persuasive techniques
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
18. methodology
study 2:
study 1:
can we take advantage
is it really
of the fact that peers
a problem?
know each-other?
19. methodology
study 2:
study 1:
can we take advantage
is it really
of the fact that peers
a problem?
know each-other?
study 3:
can social tagging
support
self-promotion?
20. study 1
• 48 Facebook users
• m: 36, f: 12, median age: 26.5 years
• 3 independent social networks
• exploratory questionnaire
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
21. Do you feel overwhelmed by the amount of
updates from your friends’ activity on Facebook?
22. Do you feel overwhelmed by the amount of
updates from your friends’ activity on Facebook?
66.6% of the respondents perceived
the ‘Facebook fatigue’
23. What strategies do you use to keep up with what
your friends are doing on Facebook?
24. What strategies do you use to keep up with what
your friends are doing on Facebook?
only 31% of the respondents declared to have a
strategy to deal with information overload
25. summary study 1
These results confirm that information overload is a
problem in social networking sites.
Most of the respondents felt overwhelmed by the
amount of updates in FB and they adopted few
strategies to overcome it.
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
26. study 2: playing with
mutual-modeling
• 9 participants (m: 8, f:1) were recruited by
mail advertisement
• average 31y (age ranged from 23 to 46)
• CS students, researchers, administrative
assistants
• tagging exercise (comprising 3 phases)
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
29. c) rate the comments
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
30. People tend to like photo comments from their
peers, mostly when they include jokes.
“I REALLY liked to see my friends’ comments on my picture”
(participant 2)
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
31. Commenting is typically a communication
activity directed to a person,
while tagging is impersonal.
“I think that keyword based tagging is better for content retrieval
because the addressee of the communication is the anonymous
world and thus the terms are often chosen in order to explain the
picture. As for the comment, I directed my communication to the
author of the image”
(participant 4)
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
32. Relationships between peers are different and
affect comment appreciation.
“Knowing who the owner of the picture is may result in different
comments or tagging”
(participant 6)
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
33. 1- most of the specific single-word keywords were
contained in the commentaries
2- descriptors of peers were more specific
34. 1- most of the specific single-word keywords were
contained in the commentaries
2- descriptors of peers were more specific
35. 1- most of the specific single-word keywords were
contained in the commentaries
2- descriptors of peers were more specific
36. 1- most of the specific single-word keywords were
contained in the commentaries
2- descriptors of peers were more specific
37. 1- most of the specific single-word keywords were
contained in the commentaries
2- descriptors of peers were more specific
38. summary study 2
this study suggested the idea that commentaries
could be used as a source of metadata instead of
single-word tags
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
39. summary study 2
this study suggested the idea that commentaries
could be used as a source of metadata instead of
single-word tags
tagging and commenting has a potential beyond
search and retrieval
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
40. study 3: self-promotion experiment
• 51 Facebook users (m:40, f:11)
• 3 separate social networks
- SN1: 17 part. US residents (av. age 27y)
- SN2: 14 part. US residents (av. age 28y)
- SN3: 20 part. CS researchers Spain (31y)
• controlled experiment
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
41. persuasive techniques
PhotoBest CommBest
➀ user X
➀ user X
uploads an album uploads a picture
➁
➁
for best picture
peers write a
peers vote
comment
➂
user X
➂ the 3 best pictures are
published on the main feed selects the best comment
➃ the picture plus the best comment
are published on the main feed
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
50. take home message 1:
there is a need for
signaling strategies
the abundance of less relevant content
might “crowd out” more relevant content
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
51. take home message 2:
peers generate
descriptors which are
more specific
users naturally take advantage of their implicit knowledge
about their peers when tagging content
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
52. take home message 3:
designing quality-control
mechanisms
contextualized comments might allow the users of social
networking sites to gain access to interpretative information
they could not gain elsewhere
intro → definitions → exp 1 → exp 2 → exp 3 → conclusion
53. Q&A
Social Tagging Revamped
take away messages:
1) there is a need for signaling strategies
2) peers generate descriptors which are more specific
3) SGWAP
mauro@tid.es
54. Mauro Cherubini, Alejandro Gutiérrez (UIUC),
Rodrigo de Oliveira, and Nuria Oliver
end
thanks
mauro@tid.es
http://www.i-cherubini.it/mauro/blog/
http://research.tid.es/multimedia/
Notas do Editor
I would like to acknowledge my colleagues: Alejandro Gutierrez, a PhD student at University of Illinois at Urbana Champaign and my colleagues at Telefonica Research: Rodrigo de Oliveira and Nuria Oliver.
In this talk I am going to focus on the problem of information overload in social networks. I am going to show the few solutions that users of SNs can adopt to overcome it. Then, I am going to move to the core of the presentation. We basically redesigned the concept of social tagging with the hope of providing another tool to users to fight this information abundance.
In CHI 2008, Joinson showed that people use social networking sites, such as Facebook, to monitor what the friends are doing and as a self-presentation tool. A typical way to maintain one’s virtual presence involves sharing multimedia content with peers. This feature is indeed popular and users transfer an increasing amount of content to their social network. Given this constant growth, navigating this quantity of information might become a problem for some users (particularly to those who have many friends).
To give you a sense of the scale of the growth I am telling you about, let me show some statistics. Here you can see that facebook users upload more than 100 million pictures to the site each day. Plus, they upload more than 5 billion pieces of content on the website each week.
So, what can we do about it? Let’s look at some “off-the-shelf” solutions.
FB allows its users to enable some filters that supposedly should reduce the amount of news updates with the goal of allowing the users to focus on the most important content. Unfortunately, users employing these solutions can still miss relevant content. It is difficult to understand what are the criteria that are used to operate the filtering and they are difficult to adjust dynamically to the users’ level of commitment.
FB allows its users to enable some filters that supposedly should reduce the amount of news updates with the goal of allowing the users to focus on the most important content. Unfortunately, users employing these solutions can still miss relevant content. It is difficult to understand what are the criteria that are used to operate the filtering and they are difficult to adjust dynamically to the users’ level of commitment.
Another solutions might be that of generating metadata for this multimedia content and then using this metadata to retrieve and filter in a more dynamic way. For example crowdsourcing techniques as the Google Image Labeler can produce lots of metadata in a small amount of time. Unfortunately, the game mechanics of this solution encourages players to enter generic labels, as demonstrated by Robertson and colleagues.
There are other solutions to the problem. One is that of using peers’ recommendations (i.e., social filtering). The other is that of preselecting the content that was already appreciated by other peers (i.e., social navigation). However, in both cases this requires to manually parse for this information in the list of updates.
There are other solutions to the problem. One is that of using peers’ recommendations (i.e., social filtering). The other is that of preselecting the content that was already appreciated by other peers (i.e., social navigation). However, in both cases this requires to manually parse for this information in the list of updates.
So, if we summarize, the automatic solutions can lower significantly the effort required from the user. However they do not offer any guarantee that relevant content could be overlooked. On the other hand, social filtering and navigation offer better chances to explore relevant content. However these solutions require more effort from the user.
So, in this study we focus in the middle. We try to design interaction mechanism which sits in the middle of this design space and that we call social tagging.
The tagging activity might be supported by the mutual model of the peers. Mutual acquaintances possess accurate information about each other (e.g., the work they do, where they live, what they like, and so forth). This information is organized into a mental model that is usually referred to as mutual because of its reciprocal nature.
The tagging activity might be corroborated by persuasive techniques. Persuasive techniques are interactive computing products created for the purpose of changing positively people’s attitudes or behaviors
The tagging activity might be supported by the mutual model of the peers. Mutual acquaintances possess accurate information about each other (e.g., the work they do, where they live, what they like, and so forth). This information is organized into a mental model that is usually referred to as mutual because of its reciprocal nature.
The tagging activity might be corroborated by persuasive techniques. Persuasive techniques are interactive computing products created for the purpose of changing positively people’s attitudes or behaviors
So, we conducted three studies: first we wanted to understand whether information overload in social networking sites was a real problem. Second, we tried to understand whether we could take advantage of the fact that peers know each other to design a better social tagging mechanism. Finally, we tested wether our design was actually useful to support self-promotion in SN sites.
So, we conducted three studies: first we wanted to understand whether information overload in social networking sites was a real problem. Second, we tried to understand whether we could take advantage of the fact that peers know each other to design a better social tagging mechanism. Finally, we tested wether our design was actually useful to support self-promotion in SN sites.
One of the core questions that we asked in the questionnaire was whether respondents felt overwhelmed by the amount of updates from their friends’ activity in Facebook. Two thirds of the respondents affirmed that this was the case.
respondents reported using three basic solutions when deal- ing with the constantly changing information on sites like FB: 48% checking the feeds often (23 respondents), (2) 23% using manual filters (11 respondents) such as social filtering or navigations discussed in the previous slides, and (3) 3% using automatic filters (4 respondents).(4)21% reported not having any strategy to deal with information overload.
So, we first asked the participants to generate single-words tags for their friends’ pictures. We asked them to come out with 3 synonyms, 3 verbs, and 3 adjectives.
Then, in the second phase we asked participants to write a full sentence describing their friends’ pictures.
Finally, we asked participants to review the comments that were generated by peers to describe their picture and we asked them to tell us whether they liked, disliked, or felt neutral with regard to their friends’ comments. We also asked them to explain us why.
Next we analyzed the commentaries and we found some recurring themes in the answers provided by participants.
As a final step in the analysis we recruited a number of extra participants that did not know the people in the social network that participated in this experiment. We asked these other participants to also generate single-word tags for the pictures.
Then we categorized the keywords that were generated by all the participants and we noticed two interesting facts.
As a final step in the analysis we recruited a number of extra participants that did not know the people in the social network that participated in this experiment. We asked these other participants to also generate single-word tags for the pictures.
Then we categorized the keywords that were generated by all the participants and we noticed two interesting facts.
As a final step in the analysis we recruited a number of extra participants that did not know the people in the social network that participated in this experiment. We asked these other participants to also generate single-word tags for the pictures.
Then we categorized the keywords that were generated by all the participants and we noticed two interesting facts.
As a final step in the analysis we recruited a number of extra participants that did not know the people in the social network that participated in this experiment. We asked these other participants to also generate single-word tags for the pictures.
Then we categorized the keywords that were generated by all the participants and we noticed two interesting facts.
In fact, we would like to suggest that good commenting could be used to signal the quality and relevance of the multimedia content that is shared in web sites such as facebook. To prove this idea we conducted the third study that I am going to explain next.
They did not receive any monetary incentives.
To test the idea that emerged during the second study, we designed 2 persuasive techniques. The first PhotoBest did not involve any commenting, while the second CommBest reflected more the results of the second study.
Photo best works as follows ...
Commentary best works as follows ...
Given that we propose to use commentaries, somebody in the audience might be wondering why the comments that users leave in facebook are not good already. Well, we think that because these comments are generated outside of a structured activity, such as that of a game, they tend to focus more on the relationship between the peers and less on the actual content that is being shared.
So, to come back to the experiment, we basically recruited 3 separated social networks. Each SN was segmented so that one part of the peers participated in the description and sharing of some pictures through the persuasive techniques that I introduced in the previous slide. The rest of the peers did not see any of this commenting part. The “viewers” were asked to log into a prototype page which was replicating some functionalities of FB and click on the content of their friends that they wanted to explore further.
The prototype page looked like this. We made an effort to basically replicate the look and feel of Facebook. The page contained 4 different news feeds. Two were taken from FB: the first ...
The last two items were the experimental items that we introduced with this experiment. Each “viewer” was asked to go through four pages like this. In each page they could see content of a different peer and the order of presentation of these four methods was changed to compensate for any bias that could be due to the presentation of the feeds. We recorded the sequence of clicks that the viewers generated on this page.
These are the main results. The graph shows the cumulative number of clicks across the three social networks per experimental methods. Comments that have been generated by means of an entertaining structure attracted more attention than those usually posted in social networks. Our CommBest scored as the random selection of picture that FB uses to generate a preview of an album.
Photos are more persuasive than comments to promote one’s photo album. However, the impact of introducing a quality filter in the comment’s choosing process can highly improve it’s appeal towards acquiring the user’s preference.
These are the main results. The graph shows the cumulative number of clicks across the three social networks per experimental methods. Comments that have been generated by means of an entertaining structure attracted more attention than those usually posted in social networks. Our CommBest scored as the random selection of picture that FB uses to generate a preview of an album.
Photos are more persuasive than comments to promote one’s photo album. However, the impact of introducing a quality filter in the comment’s choosing process can highly improve it’s appeal towards acquiring the user’s preference.
These are the main results. The graph shows the cumulative number of clicks across the three social networks per experimental methods. Comments that have been generated by means of an entertaining structure attracted more attention than those usually posted in social networks. Our CommBest scored as the random selection of picture that FB uses to generate a preview of an album.
Photos are more persuasive than comments to promote one’s photo album. However, the impact of introducing a quality filter in the comment’s choosing process can highly improve it’s appeal towards acquiring the user’s preference.
These are the main results. The graph shows the cumulative number of clicks across the three social networks per experimental methods. Comments that have been generated by means of an entertaining structure attracted more attention than those usually posted in social networks. Our CommBest scored as the random selection of picture that FB uses to generate a preview of an album.
Photos are more persuasive than comments to promote one’s photo album. However, the impact of introducing a quality filter in the comment’s choosing process can highly improve it’s appeal towards acquiring the user’s preference.
I would like to conclude this talk with three take away messages. First ...
Given the rate at which social networks are growing and the rate at which content is increasingly shared through them [1], we believe the HCI community should pay more attention to designing more effective solutions that would allow users to signal the quality of the content uploaded in the network. For this reason, we conducted the second study reported in this paper.
Second ...
Participants in our second experiment really liked the ability of creating conversations, (i.e., stories), around pictures and being able to vote for their friends’ best comments. More importantly, we observed that by imposing a structure around the commenting activity, we obtained comments that are more relevant descriptions of the content than those that are typically found in social networking sites.
Third and final message...
Pictures that have been selected by peers to represent an album are better signals to promote one’s multimedia content than random pictures taken from the album. Similarly, comments (of pictures) generated by the SN peers and through a structured activity are a better signal to promote multimedia content than standard commentaries.
We believe this can give raise to a new class of Games With A Purpose, which we called SGWAP. Games that could take advantage of the unique features of social networks.
With that I would like to conclude my presentation, take your questions and hear your comments.
Thanks.