I would say that it highly depend on feature set you are doing to have from your sentimental analysis as a result. May be the best solution would be performing multiple clustering methods and compare results. Generic overview (visual) of clustering methods can be found here - http://scikit-learn.org/stable/auto_examples/cluster/plot_cluster_comparison.html

Good luck!

Hi Surya,

It's difficult to compare different cluster methods as there is no standard way to do that, unless you have a class attribute. If so, then you can try several cluster algorithms and use each clusterer as a basic method for classification algorithm. In this case, you can use evaluation metrics of classification process to evaluate the performance of cluster algorithms.

HTH.

Samer 

You may try widely used Support Vector Machines (SVM)

k-means is preferable.