Snowball Relation Recognition Algorithm
(Redirected from Snowball Algorithm)
Jump to navigation
Jump to search
A Snowball Relation Recognition Algorithm is a Self-Training semi-supervised relation mention recognition algorithm that approximates the one proposed in (Agichtein & Gravano, 2000)
- Context:
- Input:
- 1) Text Documents that have been proceed by a Named Entity Recognizer;
- and 2) Positive Training Examples.
- Output:
- 1) A set of Relation Mentions;
- and 2) a Relation Mention Recognition Model.
- It can be implemented by a Snowball System to solve a Snowball Task.
- It uses a Five-Tuple Lexically-based Relation Recognition Classifier Relation Mention Recognition Model.
- It uses Clustering to generalize Patterns.
- It uses Bootstrapping to benefit from Unlabeled Data.
- It uses Pattern Precision and a minimum precision Threshold to stop Pattern Generation.
- It is optimized for One-to-One Relations and One-to-Many Relations.
- It can be found at http://snowball.cs.columbia.edu/
- Input:
- Example(s):
- It can detect the ORGANIZATION-HEADQUARTER_LOCATION relation based on a few examples.
- ...
- …
- Counter-Example(s):
- See: Snowball Internal Parameters, Snowball Algorithm Description, Snowball System, PPLRE Snowball,
References
2006
- (Xia, 2006) ⇒ L. Xia. (2006). “Adaptive Relationship Extraction by Machine Learning." Masters Thesis, Sheffield University.
2003
- (Yu & Agichtein, 2003) ⇒ H. Yu and Eugene Agichtein. (2003). “Extracting Synonymous Gene and Protein Terms from Biological Literature.” In: Proceedings of the 11th International Conference on Intelligent Systems for Molecular Biology (ISMB-2003). (paper.pdf)
2000
- (Agichtein & Gravano, 2000) ⇒ Eugene Agichtein and Luis Gravano. (2000). “Snowball: Extracting Relations from Large Plain-Text Collections.” In: Proceedings of the 5th ACM International Conference on Digital Libraries (DL-2000).