Date of Award

2-12-2020

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Xiaowei Xu

Abstract

This research demonstrates an effective algorithm and provides a tool for efficient and accurate disambiguation of domain specific texts. The goals outlined in the research are: 1) to demonstrate that a simple knowledge base can be used to tag texts, 2) to demonstrate that a tensor can be used to efficiently and accurately tag entities in raw text documents, and, 3) create and provide a machine learning digital humanities tool for scholars interested in early Christian documents in English and Greek from late-antiquity. The research shows that a simple knowledgebase using a comma delimited file along with a rank-3 tensor using a Bayesian method for probability can result in a high level of recall and precision on domain specific projects. While this specific tool is for early Christian writings (English translations and Greek) from late antiquity, it can easily be modified to fit any corpora of texts. The end user need only create a simple comma delimited file with entity names, surnames, aliases, and connections to tag the texts in TEI standard. There are numerous named entity resolution/disambiguation tools available to scholars. However, each of them are suitable for “big data” and do not provide the accuracy needed in domain specific projects, especially projects dealing with non-contemporary entities. Therefore, this research demonstrates a novel approach and a novel tool for tagging raw text in large corpora.

Download

Share

COinS