Q&A SYSTEMS - CHI 2009
The first presentation was very interesting, insightful and well presented. I am definitely going to follow up with this paper and visit his websites to see what is going on. Several ideas as to how I might be able to investigate this further. Would like to see what could be done to develop some niche q&a sites and also investigate how this could be overlaid onto developments in the online (distance learning) education environment.
Facts or Friends? Distinguishing Informational and Conversational Questions in Social Q&A Sites
F. Maxwell Harper (presenter), Daniel Moy, Joseph A. Konstan, University of Minnesota, USA
This research examines the differences between “conversational” and “informational” question-asking behavior in several online question and answer sites, learning in the process about linguistic and social indicators of information quality.
Started out by talking about the importance of his chair. What would happen if the hydraulic lift in his chair broke? How would he find out how to fix it? Google Search does not come up with the answer. What about a Q&A site like Ask MetaFilter. Users get answers and can mark the ones they find most useful. Yahoo Answers - 90,000 questions per day = 2.7 mill web pages per month - equivalent to all of wikipedia. Turns into to a type of social conversational site where conversations are initiated in the form of a question like "what should you be doing right now?"
two types of questions in the world
informational - asked with intent of getting information of fact or informational nature.
conversational - intent of stimulating discussion
Do they differ in terms of archival data?
developed analysis tool that asked people to evaluate questions that were gathered from numerous q&a sites. 95% of questions had consensus of what type of question was being asked. Identified archival value of questions - informational had higher overall archival value.
Structural differences - words - spam - list word and biwords (combination of two words) - patterns emerge that help to determine category of question. Ex. you more frequent in conversational, I more frequent in informational
mapping of q&a users can be used to develop a social network metric and identify quantity of relationships in conversational vs. informational
clustering coefficient - didn't get details of this
Machine Learning Approach
Ensemble Architecture - start with question and metadata - extract features - read features into specialized classifiers - send output of specialized classifiers - meta classifier - final prediction
classifying questions as conversational or informational is intuitive and useful.
It is possible to predict intent of asker with accuracy - could lead to a type of automated tagging
broadly applicable to online social media sites - ex. feature set (category, text, social network)
grouplens.org
communitylab.org
NSF grants
mimir: A Market-Based Real-Time Question and Answer Service
Gary Hsieh (presenter), Carnegie Mellon University, USA
Scott Counts, Microsoft Research, USA
Presents a market-based real-time question and answer (Q&A) service. A controlled field experiment shows that market can improve Q&A quality but can reduce social exchanges important to a community.
high number of "frivolous" questions - database was Microsoft's Q&A system
support is needed to help differentiate questions - serious and important questions can end up being overlooked
used economic markets as model for developing q&a market
possible options - enable signaling by allowing askers to offer something of value to incentivize answerers
existing synchronous q&a services - aardvark, twitter answers
desire to leverage real-time and market q&a
how will markets impact q&a?
developed control study to measure this approach - mimir
im type application, filtering of incoming questions, created virtual currency value for exchange
design problems - preventing misuse, askers do not get answers (refund option)
impact on q&a
created two groups - non-market and market
4 hypotheses of market system
1. questions in market are more serious and more important - small but distinct difference
2. answers are better, higher quality - yes
3. fewer answers per question in market - 3.2 vs. 2.1 - yes
4. will paying diminish sense of community - yes 3.04 vs 2.49
payment environment reduces social element and invites more serious answers
takeaways - need to offer both within a community to allow for development of social element
analyzing mahalo answers - allows both free and pay questions on same site
Questions in, Knowledge iN? A Study of Naver’s Question Answering Community
Kevin Kyung Nam (presenter), Mark S. Ackerman, Lada A. Adamic, University of Michigan, USA
800% increase in online q&a communities from 2006 to 2008
Prior work
participation patterns for online communities
motivations and incentives
q&a communities
only 5.4% both ask and answer
users tend to specialize
little competition within a question - 1.3 to 2.5 answers per question
motivation - altruism, sense of ownership, disappointed with existing information, frustration, wants to help others find good info, business promotion, wants to become an expert, hobby
This is Jay Steele at CHI 2009 in Boston.
Facts or Friends? Distinguishing Informational and Conversational Questions in Social Q&A Sites
F. Maxwell Harper (presenter), Daniel Moy, Joseph A. Konstan, University of Minnesota, USA
This research examines the differences between “conversational” and “informational” question-asking behavior in several online question and answer sites, learning in the process about linguistic and social indicators of information quality.
Started out by talking about the importance of his chair. What would happen if the hydraulic lift in his chair broke? How would he find out how to fix it? Google Search does not come up with the answer. What about a Q&A site like Ask MetaFilter. Users get answers and can mark the ones they find most useful. Yahoo Answers - 90,000 questions per day = 2.7 mill web pages per month - equivalent to all of wikipedia. Turns into to a type of social conversational site where conversations are initiated in the form of a question like "what should you be doing right now?"
two types of questions in the world
informational - asked with intent of getting information of fact or informational nature.
conversational - intent of stimulating discussion
Do they differ in terms of archival data?
developed analysis tool that asked people to evaluate questions that were gathered from numerous q&a sites. 95% of questions had consensus of what type of question was being asked. Identified archival value of questions - informational had higher overall archival value.
Structural differences - words - spam - list word and biwords (combination of two words) - patterns emerge that help to determine category of question. Ex. you more frequent in conversational, I more frequent in informational
mapping of q&a users can be used to develop a social network metric and identify quantity of relationships in conversational vs. informational
clustering coefficient - didn't get details of this
Machine Learning Approach
Ensemble Architecture - start with question and metadata - extract features - read features into specialized classifiers - send output of specialized classifiers - meta classifier - final prediction
classifying questions as conversational or informational is intuitive and useful.
It is possible to predict intent of asker with accuracy - could lead to a type of automated tagging
broadly applicable to online social media sites - ex. feature set (category, text, social network)
grouplens.org
communitylab.org
NSF grants
mimir: A Market-Based Real-Time Question and Answer Service
Gary Hsieh (presenter), Carnegie Mellon University, USA
Scott Counts, Microsoft Research, USA
Presents a market-based real-time question and answer (Q&A) service. A controlled field experiment shows that market can improve Q&A quality but can reduce social exchanges important to a community.
high number of "frivolous" questions - database was Microsoft's Q&A system
support is needed to help differentiate questions - serious and important questions can end up being overlooked
used economic markets as model for developing q&a market
possible options - enable signaling by allowing askers to offer something of value to incentivize answerers
existing synchronous q&a services - aardvark, twitter answers
desire to leverage real-time and market q&a
how will markets impact q&a?
developed control study to measure this approach - mimir
im type application, filtering of incoming questions, created virtual currency value for exchange
design problems - preventing misuse, askers do not get answers (refund option)
impact on q&a
created two groups - non-market and market
4 hypotheses of market system
1. questions in market are more serious and more important - small but distinct difference
2. answers are better, higher quality - yes
3. fewer answers per question in market - 3.2 vs. 2.1 - yes
4. will paying diminish sense of community - yes 3.04 vs 2.49
payment environment reduces social element and invites more serious answers
takeaways - need to offer both within a community to allow for development of social element
analyzing mahalo answers - allows both free and pay questions on same site
Questions in, Knowledge iN? A Study of Naver’s Question Answering Community
Kevin Kyung Nam (presenter), Mark S. Ackerman, Lada A. Adamic, University of Michigan, USA
800% increase in online q&a communities from 2006 to 2008
Prior work
participation patterns for online communities
motivations and incentives
q&a communities
only 5.4% both ask and answer
users tend to specialize
little competition within a question - 1.3 to 2.5 answers per question
motivation - altruism, sense of ownership, disappointed with existing information, frustration, wants to help others find good info, business promotion, wants to become an expert, hobby
This is Jay Steele at CHI 2009 in Boston.
Comments (0)
Add a Comment