Hierarchical Topics in Texts Generated by a Stream

We observe a stream of text messages, generated by Twitter or file and present tool which constructs dynamic list topics. Each tweet generates edges graph where the nodes are tags link author with in tweet. consider large clusters approximate Reservoir sampling. study giant components each component represents topic. The high degree their provide first layer topic, iteration over hierarchical decomposition. For standard text, we use Weighted sampling weight is similarity between words given Word2vec. overlapping windows topicalization on window. compare this approach Word2content LDA techniques case viewed as stream.