restful api endpoint about github-recommendation-engine HOT 7 OPEN

Hi @x0rzkov .

Thank you for your feedback

Well, the original idea was collecting data from different data sources, like Github and Twitter, to categorize the user and then suggest recomendations about github repos and computer science books, based on that data.

The problem I found is the resource constraint. Heroku free tier gives 1 CPU and some hundreds of MB for the project. And my home-lab kubernetes cluster does not have a good SLA ;-)

A feasible alternative could be to deliver it as a pip package, refactor the code to use flask-api or similar, and let the user host the api endpoint on his own infra.

But to be honest, I don't know if enough people are interested in the project to justify the effort.

Kind regards

from github-recommendation-engine.

@jaimevalero I would integrate it to https://paper2code.com and that would be awesome.

from github-recommendation-engine.

I am looking at paper2code.com (I discover it today) and it is very impressive. Congrats.
I think I would need to know more about the stack it uses, to know how to integrate it.

from github-recommendation-engine.

@jaimevalero thanks for your reply and compliment :-)

Paper2code tries to link publications/research papers with code repositories on github and other vcs. One of our feature is to recommend similar papers based on the simhash of their abstracts.

For example, a paper page, tensorflow-large-scale-machine-learning-on-heterogeneous-distributed-systems, click on the similar tab. You will see a list of papers we have vectorized with
torch2vec.

So the idea would be to recommend repositories based on topics or hashtags.

For instance, if you check the page Language-agnostic_BERT_COVID19_Twitter, which is code page, I would like to create a tab for suggesting similar or recommend other github repositories but I need a restful service for that as we wrote the whole website in golang.

Cheers,
X

from github-recommendation-engine.

@x0rzkov , It is an interesting idea. One of the issue I see it that we would need a big dataset of github repositories, in order to make meaningful recommendations.

Currently I use this dataset of about 17k users and 1.4k different labels, but I need to update and refactor it - and deliver the api within a docker-compose file.

Ok, as I need several slots of my time to refactor this, let me find some spare time to write an api, and see where this goes. :-)

from github-recommendation-engine.

I have a dataset of 3.7M READMEs in total and 261k READMEs linked to a research papers.
So, I can make a dump for you if you want.

Ps. I can help you for the docker part ! Just tell me what to do ^^

Do you have a twitter or telegram account for faster DM ?
My telegram account is deepocrates

Cheers,
X

from github-recommendation-engine.

@x0rzkov I have some problems creating telegram account, I will try later.

Meanwhile, work in progress ... https://github.com/jaimevalero/github-recommendation-engine/tree/feature-api

from github-recommendation-engine.