Wednesday 16 January 2013

LinkedIn's Big Data Endorsement Game - Why You Might Want to Play

LinkedIn has received mixed reviews of its Endorsements functionality since it was introduced last year. Many don't like the influx of emails (though these can be turned off); or the fact that they can be endorsed for skills by those in their network who have no direct knowledge of their ability in that area (though these inappropriate endorsements can be hidden albeit with some effort); or that it's too easy to endorse someone which increases the problem. How many endorsements are legitimate? How can LinkedIn judge which ones are "weak" and which are "strong"?

Note that I have no connection with LinkedIn, and no knowledge of how LinkedIn actually processes our data. These are some speculative ideas about what LinkedIn might do, and why.

Weak and Strong

MarginHound's paper last year about the way in which LinkedIn might use the data acquired from members endorsements [1] got me thinking about the ways in which LinkedIn might start to treat endorsement data like Google treats page rank data in searches.

Consider what might happen when you search LinkedIn for a person with a particular skill set. Your search will return a number of people, and LinkedIn can now order these by the number of endorsements, which means that the ones at the top stand a good chance of being the most relevant, and most highly skilled.

The Problem

The "weak endorsements" problem, which has been widely reported [2][3], seems to be in danger of breaking this process.

However, help may be at hand, from endorsements themselves. First of all, you need to be familiar with LinkedIn endorsements and how they work [4]. You also need to know that you can hide endorsements [5] from your profile if you feel they are inappropriate, perhaps you've only met that person once at a networking event so they can't possibly endorse you for widget repairs.

Endorsement hiding allows LinkedIn to do two things:

1. Detect "endorsement managers" who take time to hide inappropriate endorsements.
2. Detect "compulsive endorsers" who endorse others, only for those endorsements to be hidden.

Using in Search

How can this be used during a LinkedIn search to improve the quality of search results? Well, having ranked search results in order of endorsements received, we could look at the endorsers and see how many of them are "compulsive endorsers" - and if they are, remove their endorsements from the ranking calculation.

In addition, we could also look at the endorsers and see which are "endorsement managers". We could make their endorsements "count double" (or use some other weighting) as we know they understand endorsements because they actively take time to manage the ones they receive.

Suppose Alice and Bob are "compulsive endorsers", whilst Andrew and Beverley are "endorsement managers". Claire and Charles both have "Social Media" in their skills on LinkedIn, and have received two endorsements for it as follows:

Claire from Alice and Bob
Charles from Andrew and Beverley

Since LinkedIn knows about the endorsers and their approach to endorsements, Claire's endorsements may be viewed less favourably by LinkedIn because it knows that Alice and Bob endorse pretty much anyone they can. Charles, however, may have his endorsements weighted more heavily because LinkedIn knows Andrew and Beverley manage their own endorsements, and additionally have very few of the endorsements they give out hidden by the recipients.

So, in this example, Charles will rank higher for endorsements for the "Social Media" skill than Claire.

Let's now suppose that Charles receives an endorsement from Diana. Diana herself has 150 endorsements for "Social Media" and LinkedIn analyses those and determines that 90% of them are "strong" endorsements. This allows LinkedIn to add extra weight to the new endorsement from Diana, because she has so many "strong" endorsements herself, indicating she is likely to be an expert in the field.

Remember, LinkedIn "Knows"

LinkedIn knows a lot about your network. All the time you've been on LinkedIn, it's been accumulating information about you. You have access to a lot of it, of course, but there's more that LinkedIn knows. A good example is how long you've been connected to someone on LinkedIn. You can find this out for yourself if you've kept the invitation and connection emails, but it would take a while to piece it together for your entire network.

LinkedIn on the other hand, just knows. It knows when the connection records were created. So one thing LinkedIn can do for assessing "strong" endorsements is look how long you've been connected to your endorser. Length of time isn't enough on it's own of course, you might get a "weak" endorsement from someone you've known for a long time, and a "strong" one from someone you've done some work for last week. It's just one parameter that can be used.

LinkedIn also knows who in your network is a former or current colleague, and what employers you had in common. This can also be used as a guide in determining whether an endorsement is "weak" or "strong".

Summary

As stated, I have no relationship with LinkedIn, and so don't know whether this is what is already going on, or if it's in the pipeline, however this is what I do know:
  • Criticism of Endorsements has been widespread and LinkedIn will have seen it
  • There are some very clever people working at LinkedIn
  • LinkedIn won't reveal the secrets of Endorsements in the same way Google don't reveal the secrets of search
  • Such a "socially determined" level of expertise could be used in areas other than search


 Further Reading


No comments:

Post a Comment