Prediction Algorithm of Hashtags for Image Posting Social Network Services

When posting images on a social networking service (SNS), many hashtags are often added to the posts. Since searching for hashtags by oneself is a difficult and time-consuming task, systems that automatically recommend hashtags have been suggested. Conventional systems use co-occurring hashtags obtained by searching for hashtags as keywords to make recommendations, which leads to the problem of recommending multiple hashtags that are not directly related to the post. To solve this problem, this study presents new indexes to evaluate the relevance of the posted images and the hashtags, such as a reverse co-occurrence count index, a reverse co-occurrence ranking value index, and a similarity between comments index. The relevance between actual posted images and the variables is derived from the results of a questionnaire survey conducted among actual Instagram users. The results show that accuracy depends on the number of the latest posts used for estimating indexes. Also, if the number of the latest posts is more than 80, the reverse co-occurrence count has the highest accuracy, but the reverse co-occurrence rank shows a stable and good accuracy when there are more than 50 posts.