LAUSR

Abstract Images and textual metadata from social media sites such as Flickr have been used to understand the drivers and distributions of cultural ecosystem services (CES). However, using all available data from social media sites may not provide an accurate representation of individual services. For example, an image of nature might be described negatively in the image’s description. Here, we present a novel approach to refining social media data to represent CES better, including filtering by keywords, photograph content and enriching the data by including a measure of the sentiment expressed in the textual metadata. We demonstrate that the distribution of an enriched dataset of Flickr images representing hiking in the USA can contribute to different results and conclusions than the full dataset. Furthermore, we classified the contents of these hiking images and, using latent semantic analysis, clustered the images into ten groups based on the similarity of their content. The groups provide rich information, such as the importance of geodiversity and biodiversity in supporting a positive hiking experience. The application of this method can help to enrich social media data for CES studies, allowing researchers to further untangle the complex socio-ecological interactions that drive CES distributions, benefits and values.