Social microblogging platforms such as Twitter have become hugely popular forms of this latest sort of blogging. Twitter users make and use hashtags in their tweets to categorize them according to topic or theme, likewise to make them ascertainable to other bloggers through search. However, the liberated hashtag creation policy make a wide hardness for bloggers to find appropriates hashtags for their posts. Indeed, the task of recommending hashtags has many benefits to afford; notably it assists users to choose relevant hashtags for their posts in real time, which will save them from a supplementary stress. Actually, the achieve success of several models of neural networks for calculating word embeddings, has driven approaches for generating syntactic and semantic embeddings for long and noisy text, such as paragraphs, sentences and micro-blogs. On the parallel lines, our aim is to develop a hashtag recommender system to assist users to choose relevant hashtags for their posts in real time, based on using semantic embeddings representation of tweets, which we can subsequently use to capture semantic similarity or relatedness between tweets. In the current paper, we introduce an approach to hashtag recommendation in Twitter that is based on the following proceedings: Using a pre-trained word embeddings on a large corpus such as Google News applying one of the famous embeddings methods, Representing a given tweet by a weighted averaging value of its word embeddings, Then combining these features with the DBSCAN (density-based spatial clustering of applications with noise) clustering algorithm, to divide the heterogeneous collection of tweets into clusters that contain syntactically and semantically similar tweets. Afterwards, Recommending the top-K suitable hashtags to the user after computing the similarity between the entered tweet and the centroids of obtained clusters. Our system achieved promising results which demonstrate the effectiveness of our approach.
展开▼
