This document discusses using locality sensitive hashing (LSH) to solve large-scale search problems by clustering similar data points together. It presents an example of using LSH to find Facebook friends with similar interests. The key steps are: (1) representing each user as a vector of interests and computing minhashes, (2) clustering users into buckets based on minhash similarity, and (3) comparing a candidate to others in their bucket to find nearest neighbors. The performance of LSH involves tuning parameters like the number of minhashes and bands to balance false positives and negatives. Implementing LSH on MapReduce can make it scalable to large datasets.