covNMF: Covariate-Based Recommender Systems
In Pooja-Rajkumar/rectools: Advanced Recommender System

Description Usage Arguments Details Value Author(s)

Tools to incorporate user and item information into recommender system methodology.

1
2
3

covNMF(narrowrat,k)
getPreds(narrowrat)
getNMF(narrowrat,k)

`narrowrat`	Three (or move)-column data frame, with the format within each row being (userID, itemID, rating.
`k`	Desired NMF rank.

The covNMF function inputs a user ratings matrix, and if covariate information is present, i.e. there are more than three columns, regresses the ratings against the covariates. This is used to compute predicted values of the ratings, given the covariates. The predicted values are subtracted from the actual ratings, truncating below at 0 if necessary. (If no covariate information is present, then the ratings do not change.)

Then the nonnegative matrix factorization is performed. First, the full matrix is constructed from (the first three columns of) narrowrat. NMF is applied to the result, by calling getNMF.

If several values of k will be tried, one should form the predicted values separately, by calling getPreds, then adjusting the input matrix, then calling getNMF.

The functions covNMF and getNMF return the matrix approximating the one built by buildMatrix, thus providing estimated values for the missing ratings.

Norm Matloff

Pooja-Rajkumar/rectools documentation built on May 8, 2019, 2:56 p.m.