LinkedIn

https://www.linkedin.com/feed/

1) Perform a Scenario Analysis

a) Who are the target users?

Recruiters, employers, and job-seekers.

b) What are their key goals?

To find suitable candidate/job.

c) How can you help them to accomplish their goals?

To widen their potential pools with quality candidates/jobs. The key word would be quality over quantity.

2) Recommendation system structure

The site offers 2 recommender systems

  1. New additions to network
  2. Potential jobs for job seekers (possibly good candidates to recruiters/employers)

Recommender System 1. New addition to network.

New additions to network must come from either job experience (potential coworkers) or connections to one’s current network. Let’s examine my today’s recommendations to try to understand Linkedin algorithm.

  1. Out of first 10 recommendations I do not know a single person.
  2. Out of 10 people all of them are connect to people in my network
  3. Number of common connections of the 10 people ranges from 2 to 16
  4. Most of 10 work in Healthcare field, the same as me
  5. Interesting thing that every time I refresh the page I get completely different recommendations. So, it seems the recommendations are not sorted in ranked order, but they appear to be sorted randomly. I do not agree with the approach
  6. Overall the system is not very useful - it is very broad, but by being broad it has its own downsides

Recommender Sytem 2. Job Search.

Potentially interesting job options.

  1. I personally like this system better than potential network additions. The system explains why the jobs are displayed. It tells me that based on my prior job search it found these options.
  2. The jobs displayed look reasonably close to my ideal job.
  3. Let’s examine the jobs
  • Locations are all over the country, which is discouraging to me as I am not looking to move. Not sure why the jobs are not more clustered around NYC metro.
  • Number of jobs is reasonable, while there are way too many proposed network additions.
  • All jobs are in analytical/data field. Not all of them in healthcare which is fine, I would prefer to see jobs outside of my field.

LinkedIn Source

According to https://www.kdnuggets.com/2017/10/linkedin-personalized-recommendations-photon-ml.html. Linkedin uses machine learning package Photon-MLA (scalable machine learning library on Apache Spark). The article states that “. we make personalized job recommendations using a Generalized Additive Mixed-Effect (GAME) model, which generated 20% to 40% more job applications for job seekers in our online A/B experiments.”. 20%-40% sounds substantial.

3) Possible improvements

Recommendation 1. Too many options could be downside.

Number of potential new additions to network appears to be just way too big and they appear to be displayed randomly. I would try to do more research into improving ranking of the new additions. Can we offer fewer but better matches? We know what is displayed and what one chooses as potential addition. Is our algorithm optimized? Did Linkedin invest as much effort in network recommendations as it did in job recommendations? If it did, why are the reason for such different outcome/interface design?

Recommendation 2. Add a hint on why the recommendation was proposed.

It would make sense to offer explanation on why a person was displayed in network recommendations the same way as it is done for the job search.