Abstract
Unlike other media, research on the credibility of information present on social media is limited. This limitation is even more pronounced in the case of healthcare, including dementia-related information. The purpose of this study was to identify user groups that show high bot-like behavior and profile features that deviation from typical human behavior. We collected 16,691 tweets about dementia posted over the course of a month by 8400 users. We applied inductive coding to categorize users. The BotOrNot? API was used to compute a bot score. This work provides insight into relations between user features and a bot score. We performed analysis techniques such as Kruskal-Wallis, stepwise multiple variable regression, user tweet frequency analysis and content analysis on the data. These were further evaluated for the most frequently referenced URLs in the tweets and most active users in terms of tweet frequency. Initial results indicated that the majority of users are regular users and not bots. Regression analysis revealed a clear relationship between different features. Independent variables in the user profiles such as geo data and favourites count, correlated with the final bot score. Similarly, content analysis of the tweets showed that the word features of bot profiles have an overall smaller percentage of words compared to regular profiles. Although this analysis is promising, it needs further enhancements.