Comments on: IDEA #45 – Social Connections Search Engine

By: Graydon

Graydon — Thu, 22 Mar 2007 03:03:09 +0000

I think the visual element would be important when dealing with some individuals to really decipher the “sphere of influence” as Tom says…

I think something like liveplasma would be good…

What might be a good launching point is to start at the site to site interaction level (re: http://ideas.marlincreek.com/content/view/121/1/ )
and then “learn” more of the actual person to person items and incorporate and expand.

By: Tom C.

Tom C. — Mon, 19 Mar 2007 04:52:23 +0000

There are certain bloggers whose ideas/opinions/sentiment/notions I respect and most often I find myself trying to figure out their sphere of influence. This search engine will definitely give me a comprehensive view of their sphere of influence. Since the quality of their relationships is not really important in this case (to begin with), I think a simple “degree centrality” algorithm that runs over the crawled links should be sufficient to start with. However, the degree centrality algorithm would have to consider not just links (as the graph edges connecting the graph nodes) but 1) links, 2) mentions and 3) inter-blogger participation (e.g. the way David Cohen is participating on this Techquila response thread) to measure the degrees. Once the social network graph is established it can he mined for other data of data such as identifying bloggers who are viewed as experts on particular topics and so on. An enhancement that I can also think of, for maybe version 2.0, is a simple sentiment analyzer that can be used to score “the love” between bloggers and address “quality of their relationships” that I mentioned above.
Regarding, problems with identifying web presence instances; I think that’s a problem that can be solved iteratively; its not showstopper.

I like this idea.

By: David Cohen

David Cohen — Sun, 18 Mar 2007 17:15:13 +0000

Chris – that would be part of the magic, you’re right. I think if someone gets into this, they’ll find that there are some “very high confidence” correlations (such as when somebody refernces their linkedin profile on their blog) and then some possible matches. I think for the system to be successful, it would need to only count on the high confidence stuff, and be able to expand the result to include less confidently held information upon request.

By: Chris Saad

Chris Saad — Sun, 18 Mar 2007 00:07:26 +0000

The problem with this idea is mapping different instances of people to the same ‘person’.

I.e. A blog + Myspace Account + Email Address + hCard all might represent 1 person – but how do you automatically correlate them.

By: David Cohen

David Cohen — Sat, 17 Mar 2007 23:33:35 +0000

Captain Kirk – we should talk!

Maleksh – It’s not a popularity “list” that would be valuable, but rather the ability to search on a person (or between people) to better understand the social or professional ties that exist between them.

By: alex

alex — Sat, 17 Mar 2007 22:54:20 +0000

How do you determine an identity? If you link to TechCrunch, are you in fact linking to Mike Arrington, or to his blog (which consists of a number of contributors)?

By: James D Kirk

James D Kirk — Sat, 17 Mar 2007 21:04:00 +0000

How quickly we forget! I posted this back in February, and pinged this site in the article!

Maybe I should use it as an idea for TechStars!

By: maleksh da3wa

maleksh da3wa — Sat, 17 Mar 2007 18:16:56 +0000

ok… the idea is great, but also you might be in business with Mike Arrington and keep linking to each others, then you are on the top of the list?

By: maleksh da3wa

maleksh da3wa — Sat, 17 Mar 2007 18:06:23 +0000

this one reserves a dig