If a phrase has many words with low idf weighting, then its overall score should be lower compared to a phrase with more significant words – this is the intuition behind our tf-idf scoring strategy. As an example, assuming that the normalized tf of each word above is 0.5, the average tf-idf score for “machine-learning-application” would be 3.21 and the average tf-idf score for “machine-learning-as
![Topic Suggestions for Millions of Repositories](https://cdn-ak-scissors.b.st-hatena.com/image/square/a195a2957cd2636e21cbbac6efd3452a8960fd55/height=288;version=1;width=512/https%3A%2F%2Fcloud.githubusercontent.com%2Fassets%2F7935808%2F22560946%2F9de115ca-e933-11e6-96ff-87f15e4cf989.png)