Download presentation
Presentation is loading. Please wait.
1
Shock Group NER Progress
Laura Christiansen
2
Overview System needs to identify NE in shock abstracts
Metamap is an untrainable option and sometimes wrong Find an alternative to Metamap for NER Modified version of ABNER Use Metamap initially
3
Data Prep Metamap-processed abstracts with IOB notation used to train model Needs minor modification Distribution of 5 named entities calculated across entire corpus Modified k-means algorithm used to create n folds for stratified cross-validation Each fold roughly the same size Each fold has roughly the same distribution as the corpus
4
Stratified Cross Validation
Using folds selected in data prep: Train ABNER model based on Metamap IOB output Use model to identify NE in test set(s) Compute confusion matrix results Output selected test cases for manual review
5
Results Program currently running on smaller dataset
Testing program and reading through output Minor problems with the larger dataset need to be addressed Will run full dataset of 281 abstracts later in the week
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.