Design following recommendation mechanism

[yoavartzi]

We need a mechanism to recommend users to follow.

Some options:

1. Given semantic scholar link, try to get the most related users from our exiting user base.
2. If we can compute a user embedding based on their past research, we can return the nearest neighbors.

Need to study the Semantic Scholar API. They compute citing and cited authors. If we have can get these, we can intersect them with our user pool, and identify users to recommend.

[yoavartzi]

After the design, we should create issues to implement it under the current milestone.

[yoavartzi]

Simple process:

1. Use all available sources to retrieve all papers for the user, and retain the 10 most recent ones.
2. Embed each one using something like fasttext.
3. Put all papers from all users in the same space, and somehow compute a recommendation top-k/distribution for each user. The ranking of recommended users is drawn from this set.