Defense Knowledge Sharing and Yahoo Answers Everyone Knows

Defense: Knowledge Sharing and Yahoo Answers: Everyone Knows Something L. A. Adamic, et al.

Why study Knowledge Sharing? § It is a rising method of communication (YA, Wikipedia) § It differs from the classic model of a few authoritative sources § Current implementations differ from the traditional forum format § Opportunity to study interaction between people (always useful!)

…but is it social networking? § Yes! § There are regular users § Often have reused identities § Can become familiar with each other § Form networks of those who ask and those who answer § Most importantly: § It is real people communicating, sharing ideas, etc.

Why Yahoo! Answers? § The largest English language questionanswer site § Broad audience § Avoids bias of more specific purpose sites § Covers many topics § Provides a non-standard social networking methodology § Does not allow traditional forum style

Cluster Analysis § Classify question categories by § Thread length § Asker/replier overlap § Content length

Cluster Analysis § Provides a clear breakdown of categories of questions § Directly acknowledges the issue of the content of the questions § A good starting point, raises further questions about the nature of YA § Who is asking? § Are they the same people asking in each category?

Degree Distributions § Form graph by connecting users who ask to users who answer

Ego Networks § Shows connectedness of neighbors § In general, discussion topics are more connected § Larger strongly connected components § Factual answer topics are less connected § People who ask, aren’t answering § In fact, are less connected than a random graph of this size

Motif Analysis § Provides a further sense of “who responds to who” § Further indicates differences between categories

Network Structure Analysis § Provides insight into the structure of the network § How information flows in different categories under the same umbrella format § Could tailor the design of similar sites to the nature of the topics § Optimize sharing information of certain types

Network Structure Analysis § Provides strong indicators of the nature of user interaction § How is a given user going to use this service? § Ex: Discussion based have most mutual edges, perhaps better suited towards different model

Expertise § Analyzed 100 random selections from programming section § Rated 1 -5 (1 basic, 5 “pro”) § Found that only 1% was above a level 3

Expertise § Admittedly not a definitive sample! § Still an important hint about nature of YA § The questions seem shallow because they likely are § The fact that much of the information may be trivial is unimportant § Still an important topic to look into

User Entropy § Analyzed the distribution of answers from users across categories § Generally well spread out § Some users are very focused, very low entropy § But a high number have very high entropy, ask and answer many questions

Entropy Analysis § Emphasizes the breadth of YA § Many people are interested in many topics § Suggests the strengths of having a general site instead of something more tightly focused § As pointed out, this may be at the sacrifice of depth § Again, helps to build a user profile § Who is using this service? And for what?

Best Answers § YA allows the asker to pick a best answer, or other users to vote § Found: § People seem to like longer answers § No correlation between entropy across all categories and overall percentage of best answers (but some for top-level categories looking for expertise) § Reply length, track record good indicators

Best Answers § Provides a sense of what other users take to be most insightful § An important issue in Knowledge Sharing § In categories where applicable, raises question of correlation with the actual best answer § Another element not found in traditional forum

Best Answers § Provides an analysis of the Best Answer in terms of the category type § Compare to E. Agichtein, et al. Finding High. Quality Content in Social Media, WDSM’ 08, 2008 which focuses on the answer quality

So why was all this good? § Provides a first look at diverse, large scale question-answer network § Not attempting to delve into the intricacies of Yahoo! Answers § Acknowledges a number of issues that arise in such a model (Breadth vs. Depth) § High school homework problems § Provides a more well defined picture of structures and user interactions