Segmenting Hashtags using Automatically Created Training Data

1. Hashtags increasingly used to convey the actual message in tweets. Phrases and sentences turned into a hashtag. 2. Word with sentiment may trap inside a multi-word hashtag 3. Noisy and compact nature of language leads to hashtags very difficult to segment; sometimes depends on context. eg. #together; “to get her” or “together”? 4. Can we use carefully auto-segmented hashtags for training? RELATED WORK