Medical Relevancy of Cancer-Related Tweets and Its Relation to Misinformation

Abstract

Social media is one of the most dominant ways of spreading information. Still, unfortunately, these open platforms provide ways to spreading misinformation which can be extremely dangerous, especially when relevant to sensitive issues such as health-related information. Hence such platforms require an effective autonomous misinformation detection mechanism. Understanding the data is one of the necessary artifacts for building such a mechanism. In this work, we attempted to determine the medical relevancy of cancer-related tweets and explore whether they contain misinformation. We created a dataset of roughly 500 tweets and labeled them according to their medical relevance: medically relevant, not medically relevant, or unrelated to cancer. We ran logistic regression and support vector machine models on them. The highest proportion of correctly identified “medically relevant” tweets, i.e., accuracy, was 0.795. Our analysis hints at some features and factors that can automatically improve cancer-relevant and non-relevant tweet detection.