Static contents defined in Web 1.0 were replaced with structured user generated contents by means of Web 2.0. Wikis, Blogs, Social Networks, and Social Bookmarking Systems are some of the examples where users can generate and publish contents. Generating contents by use More
Static contents defined in Web 1.0 were replaced with structured user generated contents by means of Web 2.0. Wikis, Blogs, Social Networks, and Social Bookmarking Systems are some of the examples where users can generate and publish contents. Generating contents by users leads to creation of heterogeneous data which makes computation and algorithms hard to be applied. Web 2.0 benefits hashtags (tags) in order to solve the heterogeneous problem of the contents in which users can label their contents with hashtags. This technique cannot help in microblogging systems such as Twitter because of number of characters in each tweet (140 characters per tweet) and leads the tags or words be truncated or be used in heterogeneous form. In the current paper, a novel method is introduced based on Latent Dirichlet Allocation which can be used for numericalization tweets in a vector namely topic vector (TV). Additionally, TV is used for modeling users’ taste which can improve hashtag recommendation. The proposed method has been tested on 8396744 real tweets in English. The top 1 to 5 hashtags are recommended for each tweet and results show precision more than 20% and recall more than 45%. The improvement applied by TV shows that the most precision is increased from 3% to 32%, and recall from 21% to 46% to the best method tested by the authors.
Manuscript profile
