![Computer Science ETDs](../../assets/md5images/a8c896d0fc7ed4cb27fb019ff52dbb1d.jpg)
Computer Science ETDs
Publication Date
7-1-2013
Abstract
The Cognitive Paradigm Ontology (CogPO) defines an ontological relationship between academic terms and experiments in the field of neuroscience. BrainMap (www.brainmap.org) is a database of literature describing these experiments, which are annotated by human experts based on the ontological framework defined in CogPO. We present a stochastic approach to automate this process. We begin with a gold standard corpus of abstracts annotated by experts, and model the annotations with a group of naive Bayes classifiers, then explore the inherent relationship among different components defined by the ontology using a probabilistic decision tree model. Our solution outperforms conventional text mining approaches by taking advantage of an ontology. We consider five essential ontological components (Stimulus Modality, Stimulus Type, Response Modality, Response Type, and Instructions) in CogPO, evaluate the probability of successfully categorizing a research paper on each component by training a basic multi-label naive Bayes classifier with a set of examples taken from the BrainMap database which are already manually annotated by human experts. According to the performance of the classifiers we create a decision tree to label the components sequentially on different levels. Each node of the decision tree is associated with a naive Bayes classifier built in different subspaces of the input universe. We first make decisions on those components whose labels are comparatively easy to predict, and then use these predetermined conditions to narrow down the input space along all tree paths, therefore boosting the performance of the naive Bayes classification upon components whose labels are difficult to predict. For annotating a new instance, we use the classifiers associated with the nodes to find labels for each component, starting from the root and then tracking down the tree perhaps on multiple paths. The annotation is completed when the bottom level is reached, where all labels produced along the paths are collected.
Language
English
Keywords
Annotation, Ontology, Machine Learning, Naive Bayes, Decision Tree
Document Type
Thesis
Degree Name
Computer Science
Level of Degree
Masters
Department Name
Department of Computer Science
First Committee Member (Chair)
Luger, George Jr
Second Committee Member
Turner, Jessica Jr
Third Committee Member
Williams, Lance Jr
Recommended Citation
Xu, Jiawei Jr. "Ontology-based annotation using naive Bayes and decision trees." (2013). https://digitalrepository.unm.edu/cs_etds/62