Research Day
SENTIMENT ANALYSIS OF INITIATING MESSAGES POSTED AROUND A BRAIN CANCER DIAGNOSIS IN A MODERATED CANCER ONLINE COMMUNITY
Document Type
Abstract
Date
2021
Abstract
Background: Social media has become a haven for patients and their loved ones where they can exchange information and find support. Analysis of posted messages can provide valuable knowledge about disease experiences.
Objective: To perform sentiment analysis of anonymous messages about individuals recentlydiagnosed with brain cancer.
Methods: Ninety-seven free-text messages posted in a moderated cancer forum were analyzed with text mining tools. Retrieval of messages' polarity and emotions (Plutchick's wheel) was conducted using Bing and NRC emotion lexicons. Overall estimates were meta-analyzed. Potential differences of items between patients' age group, sex, tumor severity, time since diagnosis, and messenger's relationship to patient were assessed. Principal component (PCA) and multiple correspondence (MCA) analyses on emotions were performed to assess their co-occurrence in messages and lexicon respectively. (IRB: WMed-2019-0487)
Results: Messages display on average 9.7% (95% CI 8.9, 10.6) positive, and 11.0% (95% CI 10.1, 12.0) negative affect. Polarity is lexicon-dependent. Fear was the most prevalent (10.7%, 95% CI 9.8, 11.7) emotion, present in all messages. The eight basic emotions loaded into three components (73% variance explained), corresponding to positive emotions, negative emotions, and their combination. Messages with higher or unspecified severity showed slightly increased negative affect (p-v=0.019).
Discussion: Initiating messages around a brain cancer diagnosis portray a neutral affect by far, and certain amount of fear. Affect differs by disease severity. Future endeavors include addition of the other thread's messages, their content analysis, assessment of threads from other cancer sites and care phases to characterize them and compare them.