Skip to content

kaHFM relies on Factorization Machines and it extends them in different key aspects making use of the semantic information encoded in a knowledge graph

Notifications You must be signed in to change notification settings

vitowalteranelli/HybridFactorizationMachines

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 

Repository files navigation

HybridFactorizationMachines

Semantically Interpretable Factorization Machines

In the last decade, model-based recommender systems have shown their highly competitive performance in different domains and settings. In fact, they usually rely on the computation of latent factors to recommend items with a very high level of accuracy. Unfortunately, when moving to a latent space it is hard to keep references to the actual semantics of the recommended item, thus making the predictive model a black-box oracle. In this work, we show how to exploit semantic features coming from knowledge graphs to properly initialize latent factors in Factorization Machines, thus training an interpretable model. In the presented approach, semantic features are injected into the learning process to retain the original informativeness of the items available in the catalog. An experimental evaluation on three different datasets shows the effectiveness of the obtained interpretable model in terms both of accuracy and diversity for recommendation results.

In this work, we propose a knwowledge-aware Hybrid Factorization Machine (kaHFM) to train interpretable models in recommendation scenarios. kaHFM relies on Factorization Machines and it extends them in different key aspects making use of the semantic information encoded in a knowledge graph.

With kaHFM we build a model in which the meaning of each latent factor is bound to an explicit content-based feature extracted from a knowledge graph. Doing this, after the model has been trained, we still have an explicit reference to the original semantics of the features describing the items, thus making possible the interpretation of the final results.

Reference

If you publish research that uses kaHFM please use:

This work is currently under review

The full paper describing the overall approach WILL BE available here PDF

Credits

This algorithm has been developed by Vito Walter Anelli and Joseph Trotta while working at SisInf Lab under the supervision of Tommaso Di Noia.

Contacts

Tommaso Di Noia, tommaso [dot] dinoia [at] poliba [dot] it

Vito Walter Anelli, vitowalter [dot] anelli [at] poliba [dot] it

Joseph Trotta, joseph [dot] trotta [at] poliba [dot] it

About

kaHFM relies on Factorization Machines and it extends them in different key aspects making use of the semantic information encoded in a knowledge graph

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Java 75.5%
  • Kotlin 24.5%