More on measuring influence in the Blogosphere

December 20th, 2006

As we’ve blogged before, we’re trying to come up with models of influence in the Blogosphere. We started with using Political Blogs as a starting point, with datasets courtsey Buzzmetrics and Lada Adamic. One of the tools we use is the polarity (sentiment) of the post to post links to infer the trust/influence/sentiment that should be reflected in a blog-blog link. Here is what our technique came up with for some “A list” political blogs .

From–To

Num

links

Polarity before

trust propagation

Polarity after

trust propagation

MM-MM

0

N/A

+1.007

MM –DK

0

N/A

-9.290

MM–IP

10

+1.000

+1.370

MM–AT

0

N/A

+3.530

DK–MM

0

N/A

-9.290

DK–DK

0

N/A

+8.570

DK–IP

0

N/A

+9.570

DK–AT

20

-0.084

+3.260

IP–MM

8

+1.000

+1.030

IP–DK

6

+1.000

+9.570

IP–IP

0

N/A

+1.060

IP–AT

0

N/A

-3.640

AT–MM

0

N/A

+3.530

AT–DK

5

0.342

+3.260

AT–IP

0

N/A

-3.640

AT–AT

0

N/A

+1.241

MM is Michelle Malkin, AT is Atrios (Eschaton), DK is DailyKos and IP is Instapundit. So looks like we got the sentiment polarity mostly right, even if in our original dataset (rather limited, admittedly) there were no direct links between two blogs. Of course the errors are obvious as well — Atrios trusts MichelleMalkin — that will be the day :-) We’ve figured out why and are now tweaking the technique to fix such errors. Stay tuned for the detailed tech report ….