Microblogging systems, such as the popular service Twitter, are an important real-time source of information however due to the amount of new information constantly appearing on such services, it is difficult for users to organise, search and re-find posts. Hashtags, short keywords prefixed by a # symbol, can assist users in performing these tasks, however despite their utility, they are quite infrequently used. This work considers the problem of hashtag recommendation where we wish to suggest appropriate tags which the user could assign to a new post. By identifying temporal patterns in the use of hashtags and employing personalisation techniques we construct novel prediction models which build on the best features of existing methods. Using a large sample of data from the Twitter API we test our novel approaches against a number of competitive baselines and are able to demonstrate significant performance improvements, particularly for hashtags that have large amounts of historical data available.
|Publication status||Published - Mar 2015|
|Event||ECIR 2015 - 37th European Conference on IR Research - Vienna, Austria|
Duration: 1 Mar 2015 → …
|Conference||ECIR 2015 - 37th European Conference on IR Research|
|Period||1/03/15 → …|