In this demo, we took a news-wire corpus spanning two months and analyzed how the topics treated therein, change in time.
First, each document in the corpus is analyzed using PoolParty Semantic Suite to find which concepts from Eurovoc it contains.
Then, the corpus was split into weekly sub-corpora, each having over 1000 news wires.
In each of the weeks, we identified topics using Non-negative matrix factorization.
The topics are sets of concepts, and are shown in different colors in the circular visualization: one dot per concept it contains.
As weeks go by, the topics treated in the corpus change. These changes can be visualized by clicking Play, with the colors being shared across weeks by topics which are similar.
Thus, the green topic in the first week matches to the green topic in the second week.
Notice how some concepts "fly" from one topic to another.
In the second visualization, a Sankey diagram shows how concepts flow between topics. Sometimes two topics merge from one week to another, sometimes a topic splits into two or more.