Date of Award

May 2020

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Information Studies

First Advisor

Xiangming Simon Mu

Committee Members

Wolfram Dietmar, Jin Zhang, Soohyung Joo, Kun Lu

Keywords

ANN Classifier, Ensemble Query Expansion, Health Information Retrieval, Information Retrieval, LDA Topic Model, MeSH-term Query Expansion

Abstract

Information retrieval in the health field has several challenges. Health information terminology is difficult for consumers (laypeople) to understand. Formulating a query with professional terms is not easy for consumers because health-related terms are more familiar to health professionals. If health terms related to a query are automatically added, it would help consumers to find relevant information. The proposed query expansion (QE) models show how to expand a query using MeSH (Medical Subject Headings) terms. The documents were represented by MeSH terms (i.e. Bag-of-MeSH), which were included in the full-text articles. And then the MeSH terms were used to generate LDA (Latent Dirichlet Analysis) topic models. A query and the top k retrieved documents were used to find MeSH terms as topic words related to the query.

LDA topic words were filtered by 1) threshold values of topic probability (TP) and word probability (WP) or 2) an ANN (Artificial Neural Network) classifier. Threshold values were effective in an LDA model with a specific number of topics to increase IR performance in terms of infAP (inferred Average Precision) and infNDCG (inferred Normalized Discounted Cumulative Gain), which are common IR metrics for large data collections with incomplete judgments. The top k words were chosen by the word score based on (TP *WP) and retrieved document ranking in an LDA model with specific thresholds. The QE model with specific thresholds for TP and WP showed improved mean infAP and infNDCG scores in an LDA model, comparing with the baseline result. However, the threshold values optimized for a particular LDA model did not perform well in other LDA models with different numbers of topics.

An ANN classifier was employed to overcome the weakness of the QE model depending on LDA thresholds by automatically categorizing MeSH terms (positive/negative/neutral) for QE. ANN classifiers were trained on word features related to the LDA model and collection. Two types of QE models (WSW & PWS) using an LDA model and an ANN classifier were proposed: 1) Word Score Weighting (WSW) where the probability of being a positive/negative/neutral word was used to weight the original word score, and 2) Positive Word Selection (PWS) where positive words were identified by the ANN classifier. Forty WSW models showed better average mean infAP and infNDCG scores than the PWS models when the top 7 words were selected for QE. Both approaches based on a binary ANN classifier were effective in increasing infAP and infNDCG, statistically, significantly, compared with the scores of the baseline run. A 3-class classifier performed worse than the binary classifier.

The proposed ensemble QE models integrated multiple ANN classifiers with multiple LDA models. Ensemble QE models combined multiple WSW/PWS models and one or multiple classifiers. Multiple classifiers were more effective in selecting relevant words for QE than one classifier. In ensemble QE (WSW/PWS) models, the top k words added to the original queries were effective to increase infAP and infNDCG scores. The ensemble QE model (WSW) using three classifiers showed statistically significant improvements for infAP and infNDCG in the mean scores for 30 queries when the top 3 words were added. The ensemble QE model (PWS) using four classifiers showed statistically significant improvements for 30 queries in the mean infAP and infNDCG scores.

Share

COinS