College Library Personalized Recommendation System

Created: 2022-06-03 13:14
#paper

Main idea

In this paper an hybrid algorithm (Collaborative filtering + Content-based filtering) to recommend books is proposed.

In deep

Steps:

  • Collaborative filtering: given the matrix R=UxI where U is the set of users and I is the set of categories, $R_{i,j}$ is the number of books borrowed by user i in category j. The similarity can be computed using Person correlation or cosine similarity. From the set of k most similar users we get the books (after deleting the books the target has already read). Since the matrix is very sparse (99.9% is empty) use k-means before calculating similarity;
  • Content-based: build books' features and users' profiles -> for each user create a list of borrowed books and importance feature (?)
  • Hybrid approach -> obtain lists of recommendations from the previous apporaches and mix them -> how? It is not explained

References

  1. Paper

Code