Collective Factorization for Relational Data
Project Contributors
- Nitish Gupta
- Sameer Singh
Collective Factorization for Relational Data
In this project we present a collective factorization framework for relational data to predict all relations in the database by collectively modelling all relations in the database, even though they maybe inaccurate and partially-observed.
The model is able to incorporate observed information from all the relations while also predicting relations of interest by learning a set of entity factors that are shared across all relations. Collective modelling relations is also especially useful for cold-start estimation for entities with no observed data for the relation being predicted.
As our model embeds all the entities in the same latent-space, it is able to compute similarity between entities that do not participate in the same relation in the database schema.
We use the probabilistic interpretation of matrix factorization and learn model parameters by maximizing the log-likelihood of the observed data.
We demonstrate the efficacy of our collective factorization model on relation prediction, cold-start estimation and entity similarity estimation by evaluating on the Yelp Datasets.