Most useful clustering algorithm for NER / document matrix

I have a matrix composed of documents in columns and named entities recognized in all the documents as rows.

K-means clustering has not offer me a meaningful set of clusters, and indeed one of the clusters covers more than half of the entire NER set.

I would like to obtain a set of cluster not so dissimilar and that I could make interpretative sense of it. I mean, I would like to cluster my set of documents in a meaningful way, or at least with partial clues involved.

Thanks in advance for any suggestions about the topic.