An extended spreading activation technique for hashtag recommendation in microblogging platforms

Abstract

Microblogging platforms such as Twitter have become immensely popular. Users of these platforms use hashtags to gather similar posts under one umbrella. However, the percentage of tweets containing hashtags is still small, and hashtag use is very heterogeneous because users are free to create their own hashtags. Therefore, users might spend considerable time searching for appropriate hashtags for their posts in real time. Thus, there is a need for a powerful and effective hashtag recommendation system to assist users. Recent publications have described methods to address this need from different viewpoints, but these methods are based on traditional approaches to classification and tag recommendation that are successful for long texts but not for short texts, such as tweets. The existing methods use a syntactic mechanism, which has limitations; for example, they do not consider blogger preferences in the choice of hashtags. In this paper, we introduce an approach to hashtag recommendation in microblogging platforms that is based on our extended version of the spreading activation technique and uses several semantic knowledge bases to fill the gap between the semantic context of tweets and the semantic meaning of hashtags. The proposed approach semantically annotates the current tweet and the most popular hashtags in a recent period, after which it constructs a semantic network using DBpedia. We propose the use of an extended version of the spreading activation technique to calculate the semantic similarities of tweets and hashtags and then recommend the top k suitable hashtags to the blogger. The experimental results obtained from two Twitter datasets collected in 2012 and 2016 demonstrate the effectiveness of our approach.