Paper: Factorized Personalized Markov Chains for Next Basket Recommendation (Rendle et al. 2010)
This repository contains my implementation of FPMC in R that predicts a user's next purchase based on the past purchase history. The algorithm combines a matrix factorization of user-item matrix to model user preferences and factorized (first order) Markov chains to consider sequential dynamics.
FPMC models both long-term user preference (matrix factorization) and short-term sequential dynamics (markov chains). It factorizes two matrices: the user-item matrix and the item-item transition matrix using a S-BPR (Sequential Bayesian Personalized Ranking) loss and sums up the similarity based on both. S-BPR uses a sigmoid function to characterize the probability that a true item is ranked higher than a false item given a user and the model parameters, assuming independence of users and time steps.
It has been shown using multiple datasets that FOSSIL (Factorized Sequential Prediction with Item Similarity Models) performs better than FPMC, but the performance could vary depending on the nature of data generating process and the task domain.
Run the preprocess.R script on movielens_trunc.csv file and then the fpmc.R on the training set produced. I'm going to upload the python version of the algorithm as well as functions for evaluation metrics (i.e. AUC, precision, recall) soon, so stay tuned.