Learning Attribute to Feature Mapping

[zerowgravity]

For the cold start use case, just curious if we can add an attribute to feature mapping step as an intermediate step that has the same performance gains of this library?

Reference: https://pdfs.semanticscholar.org/f471/df4f220b830657914106c88efd5132608288.pdf

[ibayer]

if we can add an attribute to feature mapping step as an intermediate step that has the same performance gains of this library?

Can you add some details?

[zerowgravity]

Referencing the paper above:

Generally, the runtime overhead of adding mapping functions to an existing factorization model is low. For each new entity, the factors need to be estimated once, and can be either be stored in the pre-existing factor matrices, or in special data structures. After that, the computation of a prediction takes the same time as with just the underlying model. Note that factorization models themselves are among the fastest state-of-the-art methods.

I was wondering if this library could have a module that does the attribute mapping as a precursor to the learning algorithm.

For example something that generates item/user factors from its attributes.

[ibayer]

The paper you reference [0] addresses the cold start problem for positive only feedback recommender systems. The cold start problem is address with mapping functions many of them can be expressed by factorization machines though feature engineering (see [1] for details).

The positive only feedback is address though the BPR framework [2], fastFM includes a experimental implementation. We have suggested implicit coordinate descent [4] as an alternative to BPR but this is not yet implemented in fastfFM.

HTH

[0] Gantner, Zeno, et al. "Learning attribute-to-feature mappings for cold-start recommendations." Data Mining (ICDM), 2010 IEEE 10th International Conference on. IEEE, 2010.

[1] Rendle, Steffen. "Factorization machines with libfm." ACM Transactions on Intelligent Systems and Technology (TIST) 3.3 (2012): 57.

[2] Rendle, Steffen, et al. "BPR: Bayesian personalized ranking from implicit feedback." Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. AUAI Press, 2009.

[3] Bayer, Immanuel, et al. "A generic coordinate descent framework for learning from implicit feedback." Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 2017.