Trend Detection On Twitter

In this paper, we focus on the problem of trend detection on microblogging website. Learning from the idea of k-Nearest Neighbor algorithm, we propose a latent signal source model which treats the trainning data as a proxy for the unknown latent signal source. We can do the online trend detection with our algorithm. To a certain extent,we solve the trend or not-trend classification problem of high high-frequency words.With a small sample of Tweets,trends can be detected earlier than the Twitter by our method。We can detected trend about20minutes earlier, while maintaining a relatively low error rate:true positive rate of false positive rate. And by analyzing the impact of the algorithm parameters, we show that the algorithm has strong flexibility.