Recommendations for User-Role Pairing Models – Seeking Alternatives to ALS

Hi everyone…

I’ve been working with Matrix Factorization using ALS to develop a recommendation model for suggesting new roles to users during onboarding. Despite my efforts, I’ve only managed to achieve a 45-55% error rate based on the roles suggested versus the roles users actually have. Since we don’t have ratings for role recommendations, we’re using an implicit rating of 1.

I’m considering switching to a content-based recommendation model that factors in user attributes like job profile, seniority level, related projects, and other applications they access. I believe this approach could potentially perform better.

However, most online resources and discussions focus on collaborative ALS models, especially for movie recommendations. I also tested a kNN model, which achieved about 66% accuracy but takes hours to run for our user base.

TL;DR: I’m looking for recommendations for a recommendation model that leverages user attributes to suggest roles a user might need or want to request. Any insights or experiences with such models would be greatly appreciated!

Thanks in advance for your help.

1 Like

Have you explored Matrix Factorization with implicit feedback techniques?

1 Like

Most vector databases now include ANN, which should be faster than brute force KNN but has lesser accuracy.