Block-LDA: Jointly modeling entity-annotated text and entity-entity links

We present a model that improves entity entity link modeling in a mixed membership stochastic block model, by jointly modeling links with text about the entities that are linked in the relational data. The model also correspondingly improves the modeling of text annotated with entities using externally supplied entity-entity relations. We apply the model to a protein-protein interaction (PPI) dataset supplemented by a corpus of s of scientific publications annotated with the proteins in the PPI dataset. Evaluation of the model using functional category prediction of proteins and perplexity shows improvements when joint modeling is used over baselines that uses only link or text information.

• Type: Journal
• Year: 2011
• Where: SDM
• Authors: Ramnath Balasubramanyan, William W. Cohen
• Keywords: Block Model, Data Mining, Entity Relations, Protein Interaction, SDM 2011