Improving Classification of Age From Text With Semi-Supervised Conditional Random Fields
Year: 2016
Type: dissertation
Abstract: Forum text data represents a rich source of unstructured information that can be made more interpretable if given some structure. To this end, we create a large data set by collecting forum posts from Autism discussion sites and applying rigorous regular expression pattern matching to label possible mentions of age. We apply a variety of text classifiers to learn how to predict age from text. In addition, we employ shallow neural networks to train word embeddings on the forum corpora. Lastly, by taking advantage of metadata from the forum posts, we transform the classification problem in one of structured learning. By using Conditional Random Fields to take advantage of sequenced data, we show a marked improvement of classification performance over non-sequenced based classifiers for this task.
Author: Vincent D. Nguyen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2946307556