Sentiment Analysis Model Integration Using Combinatorial Fusion Analysis
Abstract
This thesis explores the efficacy of the Combinatorial Fusion Analysis (CFA), a method for calculating the diversity of model outputs based on rank-score diversity, also known as cognitive diversity, of machine learning models and which differs significantly from traditional measures of model diversity and independence by incorporating the Euclidean score space with the non-Euclidean rank space, in combining machine learning models for natural language processing. This is done primarily through integrating the results of sentiment analysis models, but with methods and practices that are general enough to be partially or fully applicable to other natural language processing tasks. By applying CFA on the IMDB sentiment analysis dataset, the thesis demonstrates how the approach performs in this dataset and shows a significant increase in sentiment classification accuracy compared to base models for combination. This result is competitive with other state of the art models and sets a new benchmark in sentiment analysis accuracy of 97.116%.
Subject Area
Computer science|Computer Engineering
Recommended Citation
Patten, Sean, "Sentiment Analysis Model Integration Using Combinatorial Fusion Analysis" (2024). ETD Collection for Fordham University. AAI31298306.
https://research.library.fordham.edu/dissertations/AAI31298306
