knn/dbscan/cosimilarity based user clusters

[LucaNyckees]

Problem

We want to add a brick to the recommender system, which takes into account similarities between users.

ToDo

• create a matrix with a row for each user, and then for each column (=product), the interaction the user had with that product (no interaction, or purchase with some rating)
• with this matrix, each user is a vector of dimension (1, total_nb_of_products). Cluster those vectors with any algorithm passed as argument (can be kNN, DBSCAN, cosine similarity etc., as long as it does not suffer too much from high dimensionality)
• integrate this brick to the recommendation system by proposing products of similar users for the active user

Context

No response

Notes

No response