Label Propagation for Topic Mining
For details refer to report.pdf
Microblogging has become a trend,thus many users are using such platforms to share their opin- ions.Mining these opinions can reveal valuable in- formation to companies, such as what features to develop in their products or marketing strate- gies.However there are major drawbacks when min- ing tweets, first their length do not provide enough word occurrences so traditional bag-of-words meth- ods have limitations, second they are extremely noisy and third annotated data is limited and costly to get.
In this paper we address the problem of classify- ing short product reviews by using semisupervised learning. A comparison among two semisupervised methods and a supervised method is given. Finally we propose different ways to improve the given re- sults by injecting knowledge into the semisuper- vised mode