concept topic in category machine learning

This is an excerpt from Manning's book Mahout in Action.
Google News groups news articles by topic using clustering techniques, in order to present news grouped by logical story, rather than presenting a raw listing of all articles. Figure 1.3 illustrates this.
LDA is more powerful than standard clustering because it can both cluster words into topics and documents into mixtures of topics. Suppose there’s a document about the Olympics, which contains words like gold, medal, run, sprint, and there’s another document about the 100 m sprints in the Asian games that contains words like winner, gold, and sprint. LDA can infer a model where the first document is considered to be a mix of two topics—one about sports that includes words like winner, gold, and medal, and the other about the 100 m run that includes words like run and sprint. LDA can find the probability with which each of the topics generates the respective documents. The topics themselves are distributions of the probabilities of words, so the topic sports may have the word run with a lower probability than in the 100m sprint topic.
The number of topics to model, –k 10