FANDOM


From Leeuwenberg and Moens (2016)[1]: "(Kordjamshidi et al., 2015)[2] used an efficient combination of a structured perceptron or structured support vector machine with integer linear programming. In their experiments, they compare a local learning model (LO), a local learning model with global inference at prediction time (L+I), and a structured learning model with and without inference during training (IBT+I, and IBT-I respectively). In their experiments L+I gave better results than LO, but a more significant improvement was made when using structured learning in contrast to local learning."

From Kordjamshidi et al., (2015)[2]: "The task of finding the location of biological entities is a kind of localization in the biomedical domain, so we aim to place it in the context of general domain-independent spatial language understanding, formulated in our previous research on the Spatial Role Labeling (SpRL) task [2-4]. SpRL considers generic location information expressed in free text about arbitrary entities– for example, finding the location of a book, when it is described in a sentence by referring to a table in a room. Here, we show the analogy between generic SpRL and the extraction of domain-specific localization relations in the biomedical literature. This analogy illustrates the challenges of applying generic NLP semantic extraction models to information extraction from the biomedical domain. The main contributions of this paper are as follows:"

References Edit

  1. Leeuwenberg, Tuur, and Marie-Francine Moens. "Structured learning for temporal relation extraction from clinical records." In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics. 2017.
  2. 2.0 2.1 Parisa Kordjamshidi, Dan Roth, and Marie-Francine Moens. 2015. Structured learning for spatial information extraction from biomedical text: bacteria biotopes. BMC Bioinformatics, 16(1):1