I want the text-based semantic clustering EMD do. Is there a better way of using LDA to detect topics in text, there are so provide better results? I'm going to do my EMD on discovery topics. Thanks
I suppose you should read the article From Word Embeddings To Document Distances. In this article, authors implement WMD - Word Mover's Distance and after that solve classification task. Instead of classification methods, see clustering methods that are applicable to your data. If your objects aren't described as points in space, see algorithms for that require only distance (eg. k-medoids, DBSCAN). Otherwise, your choice is wider, you can use classical algorithms, K-means for instance. Also, I recommend to read this and this issues, that are related to this problem.