Side project...

[prosto_tak]

In my spare time, I've built what I call Star Finder (http://www.starfinderplugin.com/), a Chrome Plugin that lets you score every LinkedIn Profile you open in your Chrome browser. The score is based on a number of factors: the number and quality of their recommendations, connections,

Comments

[chocolana]

Looks great, will check it out!

[prosto_tak]

Cool! It obviously works best for those people who have stuff in their LI account than for those who don't :)

[inka]

Spare time!

[prosto_tak]

Yes, I occasionally have this luxury :)

[the_furman]

installed

[prosto_tak]

Is it working for you?

[sim0nsays]

Installed, tried marking a bunch of people I consider really good.

Looks like scraping data off of LinkedIn itself doesn't correlate that well - scores were all over the place. Doesn't seem to be about "richness" of LinkedIn profile, which I guess is a good thing.
If it turns out "starring" people is most important part of the system, then critical mass could become a problem.

Pretty cool though, even as time saver for browsing linkedin. Also, clever way of finding top people from your immediate network, lol.

[prosto_tak]

Yes, good points. There are many factors to the score, with different weights. "Starring" people is one factor, but certainly it's not the whole thing.

This toy computes something sort of similar to PageRank, but incrementally. It matters who gives you a recommendation or endorsement. The higher is their score - the better, but if that person not been scored yet, you don't get the full benefit. The person is only scored when somebody opens their profile.

Give me a couple of examples of people whose score does not match your expectations, I'll be curious to take a look.

[janya]

Looks weird - confidence value changed on subsequent reloads for the same profile, people with the same names are treated as the same person even if they are obviously different people. Not sure what the score means, either. What are the low and high values for the score?

[prosto_tak]

The score ranges from 4 (usually means no data) to 10 (perfect score). Generally, anything above 8 is very good.

The system learns on every profile you (or others) open, every time somebody is starred, etc. Thus confidence may change.

The same name problem today exists when looking up people on Github or Anglelist. It's fixable, but fixing it is a bit painful, thus I have not done it yet.

Is this making sense?

[janya]

> The score ranges from 4 (usually means no data) to 10 (perfect score). Generally, anything above 8 is very good.
This is really non-obvious, and the range is rather small.

> The system learns on every profile you (or others) open, every time somebody is starred, etc.
A single open should not make visible difference, and opening of my own profile should make zero difference.
In my case, the confidence went from low to medium.

> Is this making sense?
yes! in fact, I am surprised LinkedIn isn't providing this kind of scoring yet. There's totally a need for this.

[prosto_tak]

4-10 is a small range?

When there are more users, opening a single profile would not make a difference. Right now there are few users, so every open profile improves confidence and may push it from low to medium.