Graph Algorithms - Strongly Connected Components in Spark 2
Ever since I generated doc2vec (Word Embeddings) for our documents, we found interesting things by doing computations and comparisons to these vectors. For example, we try to find similar documents using cosine similarity and other similarity measures. These representations of the document give us the flexibility of doing a lot of stuff. We tried using the vectors in an ANNOY index to find near neighbors for a document quickly. Now I am exploring these same vectors in finding documents that are repeatedly written and discusses the same topic. If I want to find these documents I figured that these documents would be closely similar. Documents that address the same topic will probably have a set of standard vocabulary. What if I want to find the most influential document among these related documents? To do that we need to define the connections between these documents. When we say "connections," I can't help but think of a Graph (or Network). Another approach is to us