Article Preview
Top1. Introduction
Twitter users broadcast short messages of 140 characters called tweets to users who follow their activities, in this form of communication users often use hashtags (the # symbol concatenated with a short character string) to categorize their posts in order to give meaning or to join communities around particular topics. Thus, users tend to use the appropriate hashtags in their tweets which is crucial for the popularity of their messages, in the measure that hashtags can be seen as a way to give some context to the tweet and they make tweets easily exploitable by other users in Twitter sphere. Moreover, they make tweets more accessible by hashtag-based search engines such as hashtags.org1. Since hashtags are neither registered nor controlled by any user or group, it will be hard for some users to find appropriate hashtags for their tweets (Kywe, Hoang, Lim, & Zhu, 2012). The problem of hashtags suggestions can be defined as follow: given a message entered by a user, retrieve the most reliable hashtags from the top-n similar messages. Indeed, some microbloggers try to create their personalized hashtags in a meaningful way to grant that these hashtags should be widely used for the topic but in many cases the truth may be the contrary, many users add the # symbol prefixed to every word in their tweets; wishing that one of these hashtags could be widely used by others microbloggers for the topic but that may hinders the quality of the tweet, and the users may be taken as tweet spammers, so it is difficult to users to find the appropriate hashtags for their tweets and they may spend a lot of time to retrieve the appropriate ones, so many efforts have been conducted to assist users in this process, some authors interpreted this problems as a recommender system based on collaborative filtering, other authors have proposed an Information retrieval approaches for hashtags suggestions in Twitter sphere by suggesting hashtags from the most similar tweets using term-based models such as TF-IDF (Zangerle, Gassler, & Specht, 2013) (Zangerle, Gassler, & Specht, 2011) (Mazzia & Juet, 2011). Semantic-based features namely the semantic overlap between the tweet and the query computed using DBpedia are shown to be highly effective for tweet search (Tao, Abel, Hauff, & Houben, 2012).
Previous studies in hashtag suggestions/recommendations use lexical matching between tweets to recommend hashtags to users (Zangerle, Gassler, & Specht, 2013) (Godin, Slavkovikj, De Neve, Schrauwen, & Van de Walle, 2013). Term based models are efficient in term of computation performances and the maturity of term weighting theories make this models very rampant. However, term based approaches often suffer from the problems of polysemy and synonymy and very sensitive to term use variation, especially in the context of micro-posts, due to the limited length of this form of communication, his noisy nature and his highly contextualization. Thus, in their approach the authors leverage semantic features to improve the recommended hashtags by harnessing contextual information which was not studied in previous works.
On the whole, the main contributions of this paper are:
- •
The authors present an approach for hashtag suggestions/recommendations based on semantic similarity.
- •
They use a method for named entity linking for the context of tweets by considering the nature of this form of communication.
- •
They study the impact of different hashtags ranking strategies on their system.
- •
They study the impact of combining ranking strategies on the hashtags recommendation process.
- •
To evaluate the effectiveness of their approach they use a real data set harvested from Twitter; the experiment results show that their approach improves respectively lexical and semantic based recommendation by more than 11% and 7% on recommending 5 hashtags.