Hashtag Recommendation Using Word Sequences’ Embeddings

Abstract

Nowadays, billions of people use social networks such as Twitter. Twitter users create and use hashtags in their tweets to classify them corresponding to topic or theme. Hashtags have been progressed into a multifaceted instrument to tag and track content, emphasise a standpoint or galvanise communal support across published posts on social networks. Although, by dint of the free hashtag creation strategy, users are having a broad toughness to choose suitable hashtags for their posts. In this paper, we introduce an approach for hashtag recommendation in Twitter based on tweets embeddings. We first make use of multiple techniques to calculate embeddings of the tweets in the corpus. Next, we use the k-means clustering algorithm in order to divide the heterogeneous tweets into clusters of similar tweets. Afterwards, we compute the similarity between the entered tweet embeddings and the centroids embeddings of each obtained cluster to recommend the most appropriate hashtags to the user. Through miscellaneous experiments, we introduce an itemized study on how the techniques used for tweet embeddings influence on the final set of the recommended hashtags.