Unsupervised Keyword Extraction for Hashtag Recommendation in Social Media

Abstract

Hashtag recommendation aims to suggest hashtags to users to annotate and describe the key information in the text, or categorize their posts. In recent years, several hashtag recommendation methods are proposed and developed to speed up processing of the texts and quickly find out the critical phrases. The methods use different approaches and techniques to obtain critical information from a large amount of data. This paper investigates the efficiency of unsupervised keyword extraction methods for hashtag recommendation. To do so, well-known unsupervised keyword extraction methods are applied to three real-world datasets including a new dataset containing texts of user-generated posts on a social learning platform. Experimental evaluations demonstrate that statistical methods performs newer methods including graph-based and embedding-based approaches in generating hashtags for long text, whereas the embedding-based approaches works better on generating hashtags for short texts. As a consequence, it can be concluded that unsupervised keyword extraction models can be adapted for hashtag recommendation when the social platform is new or there is no existing data to develop dedicated supervised learning models.

An Improved TextRank Keyword Extraction Method Based on the Watts-Strogatz Model

“…Since the inception of a keyword extraction method based on word frequency statistical rules by Luhn [3] in 1957, the development of automatic keyword extraction technologies has evolved over more than seven decades, spawning a multitude of methods. These methods are principally categorized into supervised and unsupervised extraction methods [4]. Although supervised keyword extraction methods can achieve higher extraction accuracy, they require a large volume of annotated data, incurring significant manual labor costs.…”