juhuitseng / fpmc

Python implementation of "Factorizing Personalized Markov Chains for Next-Basket Recommendation"

FPMC

FPMC[1] implementation for python3 with Numba.

Dependencies

• Python3
• Numpy
• Numba >= 25.0

How to run

Just type

python3 run.py data/

If Numba is not installed, implementation in generic python will be used. Numba version is 10x faster than generic version.

Notes

This implemtation is the same as original paper except:

• Number of negative sample: default is 10
• Use one basket to predict one item. That is, size of "next basket - i" is 1.

Data format

Please refer to data/idxseq.txt.

The format is:

[user index] [item index] ... [item index]

The last one item is regarded as next item (next basket), and is what our FPMC will predict.

Reference

• [1] Rendle, S., Freudenthaler, C., & Schmidt-Thieme, L. (2010). Factorizing personalized Markov chains for next-basket recommendation. Proceedings of the 19th International Conference on World Wide Web - WWW ’10, 811. http://doi.org/10.1145/1772690.1772773