To See the Invisible: An Empirical Comparison of Methods for Text-Based Sentiment Analysis of Online Contents From People With Autism Spectrum Condition
Journal
International Journal of Market Research
Journal Volume
65
Journal Issue
4
Start Page
402
End Page
422
ISSN
1470-7853
2515-2173
Date Issued
2023-02-21
Author(s)
Abstract
With the growing awareness of autism, more and more organizations tried to adapt their service offerings to the special needs of autistic groups. However, the extant relevant research in marketing is still limited, and lacks empirical evidence from the field as well as affective measures. Thus, we explore a secondary research method that is able to capture affective states of autistic people. This research aims to solve two critical assumption questions relevant to the feasibility to utilize online secondary data here: (1) How to efficiently discriminate autistic groups from neurotypical people among a huge amount of unstructured online data, given that the autistic identity is usually unlabeled and invisible? (2) Do the online contents by autistic groups provide good diagnosticity for sentiment analysis, given that many studies in this area doubted the emotional processing ability of autistic people? In our case study, we focused on people who were diagnosed with autism and successfully identified 664 autistic individuals, and then collected their user-generated content on Reddit. In the end, we collected 9305 sentences in 842 posts for sentiment analysis. Based on the results of three content analysis methods in our case study, we recommend a hybrid method that involves a collaboration between human coders and AI-enabled analysis. We provide a detailed procedure about how to implement this method, and also metrics that help to evaluate the analysis results. This method can significantly improve the efficiency of the coding process with an acceptable loss of data points, which solves the first assumption question. This paper further shows that autistic groups would provide emotional information in an online environment but in a different behavioral pattern at the aggregate level, which solves the second assumption question and is initial evidence to encourage future empirical research on autistic consumers to include emotional factors.
Subjects
AI-enabled coding method
autism spectrum condition
machine learning
sentiment analysis
user-generated data
SDGs
Publisher
SAGE Publications
Type
journal article