An attempt of using clusering algorithms to explore TED corpus. The data is from https://www.kaggle.com/rounakbanik/ted-talks.
Used features: TF, TF-IDF, LSA
Clustering algorithms: K-means, MiniBatchK-means, hierarchical clustering, DBSCAN, iforest (abnormality detection)
| algorithm | entropy |
|---|---|
| MiniBatchKMeans | 5.06 |
| KMeans | 4.82 |
| hierarchical clustering average link | 5.28 |
| hierarchical clustering complete link | 5.17 |
| hierarchical clustering ward link | 4.85 |
| score | tile |
|---|---|
| -0.039 | An 8-dimensional model of the universe |
| -0.038 | Debate: Does the world need nuclear energy? |
| -0.025 | Does democracy stifle economic growth? |
| -0.021 | Why bees are disappearing |
| -0.018 | How we're growing baby corals to rebuild reefs |
| -0.015 | Our refugee system is failing. Here's how we can fix it |
| -0.012 | The laws that sex workers really want |
| -0.009 | How fear of nuclear power is hurting the environment |
| -0.008 | The refugee crisis is a test of our character |
| -0.007 | Why I still have hope for coral reefs |