Reviewer와 Gatekeeper

MPI에 와서 Krishna와 함께 일하면서 이 사람 엄청나다는걸 하루하루 볼 때마다 느끼고 있다.

특히 조리있게 생각을 정리하고 남을 설득시키는 능력은 세계 최고라고 생각하는데, 얼마 전에 이런 주제로 얘기를 나눴다.

How should a reviewer do his/her job?

물론 난 일방적으로 듣기만 했지만…;  (Krishna랑 같이 이야기하면 누가 와도 듣고만 있게 될거다;;;),

Krishna의 요지는 간단했다.

A reviewer should review a paper. One should not be a gatekeeper of one’s field.

Academia의 목적에 충실해서 의미있는 발견이나 시사점을 제시하는 걸 우선으로 볼 것이냐, 해당 분야에 관련된 내용만을 걸러내는 역할을 할 것이냐라는 선택인데…

gatekeeper의 역할을 최소화하되 아예 없으면 안된다고 생각한다. 아예 다른 분야의 논문을 제출하면야 당연히 걸러내는게 맞지만, 이런 일은 없다고 보면 되고…

어쨌든, 당시에는 무슨 말인지는 이해했어도 느낌을 잘 몰랐는데 이번에 받은 리뷰 중 하나가 아주 gatekeeper의 느낌이 강하게 풍기길래…;

시사하는 바가 큰 논문이나, 해당 분야의 논문으로 부적합

시사하는 바가 크지만 이쪽 분야와는 맞지 않는다는 이야기인데, 다른 부분들도 조목조목 다 반박할 여지가 있어서 문장을 조금 더 명료하게 바꾸는 정도에서 그쳤다.

제출한 논문의 논의점이 분야가 살짝 달라도 accept할 만큼 striking하지 않거나 아니면 이 리뷰어가  conservative한 시선을 가지고 있거나인데…나머지 두 개 리뷰는 별 태클이 없드만 ㅠㅜ

그나저나 쓰다보니 느낀건데 좋은 키보드를 사야겠다는 생각이 강하게 든다. 이번 지름신은 어떻게 물리치나.

[Python] [{}]*10 과 [{} for i in range(10)]

마땅한 파이썬 디버그 툴을 쓰지 않기 때문에 가끔 삽질을 좀 할 때가 있는데, 오늘의 삽질 이야기.

종종 dictionary여러개가 필요해서 dictionary의 list를 U =  [{} for i in rage(10)] 이런식으로 만들어서 사용하곤 했다.

며칠전에 파이썬 레퍼런스를 뒤져보다가 예제 코드에 [{}]*10 이라고 써진 걸 보고는, “오 짧네” 하고 생각없이 썼는데 이게 오늘 아침에 삽질하던 원인이었다.

결론부터 얘기하면

[{}]*10 은 dictionary하나를 생성해서 그 reference 10개의 list를 주고

[{} for i in range(10)] 은 실제로 서로 다른 dictionary 10개를 생성해서 return

실제로 아래처럼 각 object의 주소가 같고 다른걸 확인!

>>> [nx.Graph()]*3
[<networkx.classes.graph.Graph object at 0x7ff74dd4e290>, <networkx.classes.graph.Graph object at 0x7ff74dd4e290>, <networkx.classes.graph.Graph object at 0x7ff74dd4e290>]

>>> [nx.Graph() for i in range(3)]
[<networkx.classes.graph.Graph object at 0x29abdd0>, <networkx.classes.graph.Graph object at 0x29abd90>, <networkx.classes.graph.Graph object at 0x29abe10>]

오늘은 hopefully 삽질 끝.

The 1st International Workshop on Social Web 참가! :)

Attended the first IWSW(International Workshop on Social Web) held at SangamDMC in Seoul last night. I tried to summarize while listening but it turned out that nobody wouldn’t be able to figure out the context of the talk from my note…;;

The most interesting point was on the first talk since this was the first thing I intended to look into. Analyzing people’s opinion on local election in Korea, the result has shown that users on Twitter are more progressive than non-Twitter users. They even said they wouldn’t vote for the current government if they go back in time before the presidential election. (I wouldn’t either)

The third talk was also interesting. Dr.Lee Wonjae presented a conservative view of sociology on social network analysis which is new for me. I didn’t get the exact meaning of all the numbers and stats, but the point was OSN(online social network) could not reflect the “real” social relationships due to some constraints. So, the inequality we observe in real world is not as extreme as  the one that we see from the Power-law distribution.

Another thing I got was from discussion section we had at the last of the workshop.(well, the professors had to be precise ㅋ)

WE(I) MUST STUDY STATISTICS SERIOUSLY!

The rest of this post is my scribble. I don’t guarantee that you would get much from it. There are some missing parts as well.

1. Democracy after Twitter (Dukjin Chang)

Empirical analysis on 1.1million Korean users on Twitter.

It boils down to that users on Twitter are more progressive.

Interesting approaches

• For those who didn’t vote, predict their possible class in terms of…
  • indegree, out degree, income, etc

QnA

• Does all these statistics are actually representative statistically?
  • It is socially representative, but still hard to make it enough. Any good way?
• Why progressive people make less use of new media like Twitter?

2. Social? or parasocial network (Eun-ju Lee)

Empirical study on effects of celebrities.

Parasocial Interaction

• Interpersonal interaction in which one party knows the other very well while the other party does not.

Research Questions

• How does celebrities’ interaction w/ their fans thru sns affect people’s attitudes and behavioral intention toward them?
• Will the communication channel thru which such interaction is publicized matter? and Why?
  • Web site vs. News articleSocial presence as a mediator, leading to PSI(parasocial interaction)
• Any effects of individual differences?

3. Dynamics of Sociological Parameters in Computer Mediated Interactions (Wonjae Lee)

Typical approaches to the sociology (he said it’s fundamental, conservative)

• Web documents, Power Grid, Citation (Barabasi & Albert 1999) – Power law connectivity distribution
• Elizabeth Billington (Alfred Marshall 1943) – Economics of Superstar
• Matthew Effect
• Social status mechanism

Market Consumption  No constraints on resources Independence of the transactions Balanced

Social Exchange  Constraints on resources(Time & Energy) Dependence Unbalanced at individual level

Interesting Approaches

• Inequality is not as extreme as what is known by Power-law distribution
• Need more conservative constraints for “real” social relationships

4. More retweets do not always bring more audience (Sue Moon)

A majority of potential readers receives multiple messages over and over again

More followers do not always translate to more effective readers.

Future picture

• Evolution and flow of information
• From unstructured to structured data

Quick note

• For retweeting, is it about information? or just attention?

5. Is their a science in social networks? (Kavé Salamatian)

Science – a set of organised objective knowledges in a specific domain

Social science – the fields of academic scholarship exploring aspects of human society

7. Enabling the Social Web (Krishna Gummadi)

Discovering the web by word-of-mouth

Collaborative ranking based content search

• No links btw information
• So, they are ranked by users

Sybil attack

• Attacker creates many fake identities(Sybils) and use them to manipulate the system

Defense approaches

• New possible approach – using social networks to detect Sybils

Group attachment theory

• Explains how humans join and relate to groups
• Common-identity based groups (like music, politics)
  • less cohesive
  • usually large in terms of # of nodes
• Common-bond based groups (like family, alumni)
  • more cohesive
  • not very big in terms of # of nodes

Sybil defense

• Define Good / evil
• Sybil Tolerance?

Minor Tweaks on Bit hacks

I love when this kind of tweaks make my life much easier.

Well, not my life…my program.

Remembering all of these tweaks would be a burden, but this is what makes differences.

This guy is a real geek. http://catonmat.net/p/143

Yes, There is a Correlation – From Social Networks to Personal Behavior on the Web

1저자가 Parag Singla라는 사람인데 UW졸업하고 PostDoc으로 UT Austin에 있단다. 괜히 반갑네. ㅎ

MS, Redmond에 인턴으로 있을 때 쓴 논문같은데, MSN data를 이용해서 사용자들의 Demographic information과 Personal Behavior(여기에선 keyword searching을 다뤘다)의 관계를 살펴본 논문이다.

확률모델로 비교적 복잡하지 않은 실험 & 분석을 몇개 했는데 간략하게 요약만…

결과가 재미있다기보다는 이걸 base로 시작할 수 있는 further study가 더 재미있을 듯.

Summary

People who talk to each other on the messenger network are more likely to be similar than a random pair of users

Similarity is measured in terms of matching on attributes such as queries issued, query categories, age, zip and gender

The similarity increases with increasing talk time, decrease with increasing average time spent per message

Related Work & Future Work

• Homophily = people with similar characteristics tend to be connected
• [1] Review of work done on homophily in real-world networks
• (additional) how it affects the evolution of the social networks over time
• [2] How the participation in online communities might affect the every day lives and behavior of the people in the physical world
• [3, 4, 5] How important each node is in propagating certain ideas or innovations through the network
• [6] Understanding the dynamics of a viral marketing

1. M. McPherson, L. Smith-Lovin, and J. Cook. Birds of a feather: homophily in social networks. Annual Review of Sociology, 27:415–444, 2001.
2. L. Sproull and J. Patterson. Making information cities livable. Communications of the ACM: Special Issue on Information Cities, 47:2:33–37, 2004.
3. P. Domingos and M. Richardson. Mining the network value of customers. In 7th Intl. SIGKDD, pages 57–66, 2001.
4. M. Richardson and P. Domingos. Mining knowledge-sharing sites for viral marketing. In 8th Intl. SIGKDD, pages 61–70, 2002.
5. D. Kempe, J. Kleinberg, and E. Tardos. Maximizing the spread of influence in a social network. In 9th Intl. SIGKDD, pages 137–146, 2003.
6. J. Leskovec, L. Adamic, and B. Huberman. The dynamics of viral marketing. In ACM Conference on Electronic Commerce, pages 228–237, 2006.