Expertise Based Question Routing

A core theme for me in my study of collaboration technologies has been transferring tacit human knowledge from the siloed "wetware" of our minds into sharable digital media. One of the best interaction models we've found as a species to get knowledge out of someone is the question. Pose a question to the right person and you're likely to get the information you need.

Often, though, the hard part is finding the right person to ask. Here are some approaches used today:

1. No targeting. The idea here is to ask as many people as possible in the hopes that one will have at least some knowledge about the question. An example here might be asking a question to your Twitter followers or Facebook network. The idea is to make your information need explicit and available to as many people as possible without time spent focusing on a limited more suitable set.

2. Targeting based on coordinating structures (e.g. ontologies). Systems can offer the ability to focus targeting by providing a shared structure for those with questions and those with answers to leverage. An example of this is Yahoo! Answers as well as a site that I was involved with called GradShare. The ontologies in these systems enables those who would answer a question to navigate to the questions that are most likely to draw on their expertise.

3. Targeting based on personal knowledge. Characterized by reliance on personal knowledge of an individual and their areas of expertise, this approach is the most common and is based on an ability to both match a question with the expertise of a personal network as well as the ability to successfully engage the targets once found, including management of any social dilemmas such as trust.

Finally, a new breed of system is emerging that is perhaps best exemplified by a small firm called Aardvark. The approach that Aardvark and similar services takes is to use coordinating structures such as tags but also to use technology to classify the match between the question and the potential respondents and the to route the question only to those people who get high relevance scores. It doesn't always work but I've been using Aardvark for some time now and it's regularly asking me questions about where I live - which is easy enough - and what I know about (Startups) - which is not so easy. I've never filled out any questionnaire from Aardvark about my expertise so somehow they are able to determine that I have expertise in Startups and then send me related questions from strangers.

I think that an algorithmic approach to expertise finding and matching to improve question routing is an excellent idea and will be a major trend in the future of peer to peer information systems. Good matching increases answer quality, likelihood of particpation (I'd imagine), and future participation (both by the asker and the answerer). The impact within corporate environments or within professional communities (e.g. imagine a doctor being able to ask a question and getting an answer from a colleague they do not know that is dead on) could be substantial.