last.fm API and collection of a database of hundreds of thousands of users with the frequency of their listening to the latest tracks in a year. What difficulties may arise?

Question

Last.fm

The idea is this, step by step:
1. If I'm not mistaken, this step cannot be implemented directly through the API, therefore, without the API, parse the listeners of your favorite performers. There is a Listeners section on the performers' pages and there are 270 listeners in total on 9 pages - the most active, as far as I understand.
2. Again, through the API, as far as I understand, it will not be possible to get the users' neighbors, so they can be parsed.
3. Get friends of these users via API. As a result, you will get a list of some tens or hundreds of thousands of users for further analysis of their profiles.
4. Using the API, get the statistics of listening to tracks by each user for the last year. Save the most frequently listened tracks of users with reference to their login.
5. Take your list for comparison.
6. Define formulas for calculations. For example, out of the top 100 tracks per year, the average number of listens to a track is 5 for a user, and 10 for me, i.e. 2 times more is to reduce to a common denominator and compare all the tracks from the top 100. In the case of convergence of tracks, equate each converged track with a value from 0.1 to 1 and multiply by the number of convergences from the top 100, thus obtaining a rating of the total intersection in the top 100 for the year between each user and himself.

I see sense in these sophistications in the context of the imperfection of the algorithm for selecting neighbors according to musical tastes, which selects neighbors by performers, and not by tracks. Each artist can have hundreds of tracks, and the couple of tracks that I listen to are most likely not listened to by those whom last fm defines as my neighbors, so when I go into the profiles of these "neighbors", I will never find new music tracks that are interesting to me and I will spit for a long time, and perhaps even suffer from a headache after listening to something that does not fit into my tastes at all. The musical recommendations of the service are as far from my preferences as the tastes of these "neighbors", so I'm trying to invent something of my own. The recommendations of other services are also not ideal. Yes, and I want to program something.

Question: what problems can I face when implementing my algorithm, besides the fact that user IDs will have to be parsed, since they cannot be obtained through the API? And how patient is last fm? I did something similar for vk and there were difficulties, as far as I remember, in that when I began to actively receive audio lists of hundreds of thousands of users, vk changed something and stopped issuing the entire audio list of each user, which could reach up to 10,000 songs, and began give the list in microdoses of 100, of course, requesting a captcha after each one... Until vk made changes, the scheme rarely worked, but there were users with an increased number of similarities, and among their audio it was possible to find new music for themselves, bypassing the blood from their ears .

P.S.
Or maybe something similar (search for users by converging tracks, not track artists) already exists?
To what extent it could be useful / in demand for others? I would like to hear opinions / advice / recommendations on my idea.