Seasonal Fluctuations in Collective Mood Revealed by Wikipedia Searches and Twitter Posts

摘要

Understanding changes in the mood and mentalhealth of large populations is a challenge, with the need for largenumbers of samples to uncover any regular patterns within thedata. The use of data generated by online activities of healthyindividuals offers the opportunity to perform such observationson the large scales and for the long periods that are required. Various studies have previously examined circadian fluctuationsof mood in this way. In this study, we investigate seasonalfluctuations in mood and mental health by analyzing the accesslogs of Wikipedia pages and the content of Twitter in the UK overa period of four years. By using standard methods of NaturalLanguage Processing, we extract daily indicators of negativeaffect, anxiety, anger and sadness from Twitter and comparethis with the overall daily traffic to Wikipedia pages aboutmental health disorders. We show that both negative affect onTwitter and access to mental health pages on Wikipedia follow anannual cycle, both peaking during the winter months. Breakingthis down into specific moods and pages, we find that peakaccess to the Wikipedia page for Seasonal Affective Disordercoincides with the peak period for the sadness indicator inTwitter content, with both most over-expressed in Novemberand December. A period of heightened anger and anxiety onTwitter partly overlaps with increased information seeking aboutstress, panic and eating disorders on Wikipedia in the late winterand early spring. Finally, we compare Twitter mood indicatorswith various weather time series, finding that negative affectand anger can be partially explained in terms of the climatictemperature and photoperiod, sadness can be partially explainedby the photoperiod and the perceived change in the photoperiod, while anxiety is partially explained by the level of precipitation. Using these multiple sources of data allows us to have accessto inexpensive, although indirect, information about collectivevariations in mood over long periods of time, in turn helpingus to begin to separate out the various possible causes of these fluctuations.