Empirical study of topic modeling in Twitter

摘要: Social networks such as Facebook, LinkedIn, and Twitter have been a crucial source of information for wide spectrum users. In Twitter, popular that is deemed important by the community propagates through network. Studying characteristics content in messages becomes number tasks, breaking news detection, personalized message recommendation, friends sentiment analysis others. While many researchers wish to use standard text mining tools understand on restricted length those prevents them from being employed their full potential.We address problem using topic models micro-blogging environments studying how can be trained dataset. We propose several schemes train model compare quality effectiveness set carefully designed experiments both qualitative quantitative perspectives. show training aggregated we obtain higher learned which results significantly better performance two real-world classification problems. also discuss state-of-the-art Author-Topic fails hierarchical relationships between entities Media.