| Popis: |
Images are an important part of collection items in any digital library. Mining information from social media networks, and especially the Instagram, for Image description has recently gained increased research interest. In the current study we extend previous work on the use of topic modelling for mining tags from Instagram hashtags for image content description. We examine whether the hashtags accompanying Instagram photos, collected via a common query hashtag (called ‘subject’ hereafter), vary in a statistically significant manner depending on the similarity of their visual content. In the experiment we use the topics mined from Instagram hashtags from a set of Instagram images corresponding to 26 different query hashtags and classified into two categories per subject, named as ‘relevant’ and ‘irrelevant’ depending on the similarity of their visual content. Two different set of users, namely trained students and generic crowd, assess the topics presented to them as word clouds. To invest whether there is significant difference between the word clouds of the images considered as visually relevant to the query subject compared to those considered visually irrelevant. At the same time we investigate whether the word cloud interpretations of trained students and generic crowd differ. The data collected through this empirical study are analyzed with use of independent samples t-test and Pearson rho. We conclude that the word clouds of the relevant Instagram images are much more easily interpretable by both the trained students and the crowd. The results also show some interesting variations across subjects which are analysed and discussed in detail throughout the paper. At the same time the interpretations of trained students and the generic crowd are highly correlated, denoting that no specific training is required to mine relevant tags from Instagram hashtags to describe the accompanied Instagram photos. |
