Measuring Possible Future Selves: Using Natural Language Processing for Automated Analysis of Posts about Life Concerns
Keywords:Possible Future Selves
Proof of Concept for Computational Methods
Natural Language Processing
Machine Learning in Communication
MetadataShow full item record
Publisher:The Ohio State University
Series/Report no.:The Ohio State University. School of Communication Honors Theses; 2020
Individuals have specific perceptions regarding their lives pertaining to how well they are doing in particular life domains, what their ideas are, and what to pursue in the future. These concepts are called possible future selves (PFS), a schema that contains the ideas of people, who they currently are, and who they wish to be in the future. The goal of this research project is to create a program to capture PFS using natural language processing. This program will allow automated analysis to measure people's perceptions and goals in a particular life domain and assess their view of the importance regarding their thoughts on each part of their PFS. The data used in this study were adopted from Kennard, Willis, Robinson, and Knobloch-Westerwick (2015) in which 214 women, aged between 21-35 years, viewed magazine portrayals of women in gender-congruent and gender-incongruent roles. The participants were prompted to write about their PFS with the questions: "Over the past 7 days, how much have you thought about your current life situation and your future? What were your thoughts? How much have you thought about your goals in life and your relationships? What were your thoughts?" The text PFS responses were then coded for mentions of different life domains and the emotions explicitly expressed from the text-data by human coders. Combinations of machine learning techniques were utilized to show the robustness of machine learning in predicting PFS. Long Short-Term Memory networks (LSTM), Convolutional Neural Networks (CNN), and decision trees were used in the ensemble learning of the machine learning model. Two different training and evaluation methods were used to find the most optimal machine learning approach in analyzing PFS. The machine learning approach was found successful in predicting PFS with high accuracy, labeling a person's concerns over PFS the same as human coders have done in The Allure of Aphrodite. While the models were inaccurate in spotting some measures, for example labeling a person's career concern in the present with around 60% accuracy, it was accurate finding a concern in a person's past romantic life with above 95% accuracy. Overall, the accuracy was found to be around 83% for life-domain concerns.
Academic Major: Computer Science and Engineering
Undergraduate Research Scholarship by the College of Engineering
Items in Knowledge Bank are protected by copyright, with all rights reserved, unless otherwise indicated.