Fig 1.
Instagram’s wildlife selfie pop-up alert system showing the warning presented when users search for wild animal selfie hashtags (e.g. #elephantselfie) (left), the follow-on options given (middle), and an extract from the information provided (right) when users select ‘Learn More’.
Fig 2.
Example Instagram screenshots showing ‘elephant selfie’ images using the hashtags #elephantselfie (a, b), and #elephantride (c).
Table 1.
Categories used to describe posts.
‘Situation’ refers to whether the elephant is captive (C) or wild (W); the presence of people in the image is denoted by Y (yes) or N (no), and the role of the people in the image as either ‘in close proximity to, or touching the elephant’ (close/touching) or ‘observing from a distance’ (observing).
Table 2.
Hashtags associated with elephant selfie posts on Instagram.
Data shown are the post volume (the number of posts in which the hashtag was used), whether Instagram’s pop-up alert was triggered (Y = yes, N = no), and whether any of the posts had been posted (i.e. the hashtag had been used) since the warning was introduced in 2017. Not all posts conformed to our definition of an elephant selfie; hashtags were included if at least one post using it was relevant (see footnote). Only hashtags with a post volume > 500 are shown here, the full set of identified relevant hashtags is provided in S1 Appendix. Neither the list of hashtags here, or the full list of hashtags given in S1 Appendix, is intended to be exhaustive. Bold text indicates hashtags used for further analysis.
Fig 3.
Types of post using each of six hashtags, where the first of each ‘pair’ (#elephantselfie-#elephantselfies, #elephantride-#elephantrides, #elephanthugs-#elephantcuddles) triggers Instagram’s warning (marked by an!) and the second does not.
Relative proportions of post types were statistically significantly different between pairs of terms in all cases; n = 468, 225, 871, 487, 193, 268, respectively. Content categories defined in Methods and Table 1.
Fig 4.
Violin plots showing popularity (as measured by number of likes) of elephant selfie posts using each of six hashtags, where the first of each pair of terms—a). #elephantselfie and #elephantselfies, b). #elephantride and #elephantrides, c). #elephanthugs and #elephantcuddles—triggers Instagram’s warning (marked by an!) and the second does not. Circles depict the median (RCVQ = 0.64, 0.88, 0.86, 1.94, 0.91, 0.97, respectively); maximum value for #elephantselfie not shown to improve visual comparison.
Fig 5.
Violin plots showing the distribution and range of sentiment scores for comments on elephant selfie posts using each of six hashtags, where the first of each ‘pair’ of terms—a). #elephantselfie and #elephantselfies, b). #elephantride and #elephantrides, c). #elephanthugs and #elephantcuddles—triggers Instagram’s warning (marked by an!) and the second does not. Circles depict the median, the bars depict the inter-quartile range.
Table 3.
Sentiment scores for comments to posts using each of six hashtags, where the first of each ‘pair’ of terms triggers Instagram’s warning and the second does not.
Sentiment scores (assigned, based on text and emojis combined, using the “sentimentr” package (Rinker 2021) in R), reported as mean, SE, and median, and percentage negative, neutral, and positive; sample size (n) is given as the number of posts, the total number of comments for all posts using each of the hashtags combined, and the number of words used in the analysis.
Table 4.
Most frequently occurring words in comments on elephant selfie posts using each of six hashtags, where the first of each ‘pair’ of terms—a). #elephantselfie and #elephantselfies, b). #elephantride and #elephantrides, c). #elephanthugs and #elephantcuddles—triggers Instagram’s warning (marked by an!) and the second does not. Data are the frequency of occurrence of words and/or emojis appearing the comments, including only those with a frequency of occurrence of 10 or more. Stopwords were not included, except for the word ‘not’ which may be important as a valence shifter. Words and emojis with negative sentiment scores are shaded in the table.