Natural Language Processing
Natural language processing is a hard task for traditional machine learning models, since they tend to ignore the relationships between the sentences, words, and text in documents. Because of this, they require careful feature construction and often make assumptions about conditional independences between all of the elements.
Even then, typically these features do not scale with large amounts of natural language data. Therefore, methods which can extract relations from text and exploit these relations during learning and inference will reduce the need for careful feature construction, and also scale better into larger data sets.
Models learned using relational machine learning methods can be used to create robust and interpretable predictions. In these projects, BoostSRL has been applied to extracting relations from clinical data, adverse drug event relations, and financial documents.
- Sriraam Natarjan, Ameet Soni, Anurag Wazalwar, Dileep Viswanathan, and Kristian Kersting. “Deep Distant Supervision: Learning Statistical Relational Models for Weak Supervision in Natural Language Processing.” Morik Festschrift, LNAI 9580 2016.
- Sriraam Natarajan, Vishal Bangera, Tushar Khot, Jose Picado, Anurag Wazalwar, Vitor Santos Costa, David Page, and Michael Caldwell. “Markov Logic Networks for Adverse Drug Event Extraction from Text.” Knowledge and Information Systems (KAIS), 2016.
- Ameet Soni, Dileep Viswanathan, Jude Shavlik, and Sriraam Natarajan. “Learning Relational Dependency Networks for Relation Extraction.” International Conference on Inductive Logic Programming (ILP), 2016.
- Phillip Odom, Vishal Bangera, Tushar Khot, David Page, and Sriraam Natarajan. “Extracting Adverse Drug Events from Text using Human Advice.” Artificial Intelligence in Medicine (AIME), 2015.
- Sriraam Natarajan, Jose Picado, Tushar Khot, Kristian Kersting, Christopher Re and Jude Shavlik. “Using Commonsense Knowledge to Automatically Create (Noisy) Training Examples from Text.” International Workshop on Statistical Relational AI, 2012.
- Tushar Khot, Siddharth Srivastava, Sriraam Natarajan, and Jude Shavlik. “Learning Relational Structure for Temporal Relation Extraction.” International Workshop on Statistical Relational AI, 2012.