Please provide python code and explanations for the following: The current approach to finding topics relied on tf$-$idf, UMAP, DBSCAN, and

a custom tf$-$idf approach to explain each topic. In this section, you need to

compare the original method with alternative methods by making a maximum

of one change. In other words, you will always modify the original model.

Question $6\mathrm{.}1[6]$

Replace with a pretrained sentence embedding model. Discuss

whether the sentence embedding model provides better embeddings by com$-$

paring the similarity of documents using cosine distance.

Question $6\mathrm{.}2[6]$

Replace umap with PCA. Discuss the impact of using PCA on the quality

of clusters and the overall topic modeling results. In your answer specifically

discuss what properties each technique tries to observe and whether it is relevant

for the subsequent steps.

Question $6\mathrm{.}3[6]$

Use K$-$Means instead of DBSCAN for clustering. Compare the clustering results

and specifically discuss the appropriateness of K$-$Means for this task.

Question $6\mathrm{.}4[10]$

Train then use a decision tree $($surrogate model$)$ instead of a custom tf$-$idf to

globally explain why instances are assigned to the largest cluster obtained. Your

decision tree should be interpretable. In your answer clearly describe the steps

you have taken to train the decision tree and motivate how the trained decision

tree is interpretable.