Suppose you have a binary documents-by-terms matrix with 50,000 documents and 500 terms. Management asks you to develop a search application for finding a small subset of documents most relevant to a subject or question of interest across the documents. Which unsupervised learning method(s) would you select as a foundation for such a search application? Consider methods such as cluster analysis, block clustering/biclustering, topic modeling, or, perhaps, discrete multivariate analysis. Explain why you are recommending your selected method.

To promote your newly developed search application, you want to show that it is superior to Google's PageRank or standard algorithms from Elasticsearch. What indices would you use to demonstrate the superiority of your application?