I've adopted a very simple approach: 80 words per "paragraph". I am now experimenting with computing the embeddings of each sentence and try to detect topic seg

I've adopted a very simple approach: 80 words per "paragraph". I am now experimenting with computing the embeddings of each sentence and try to detect topic segments. But the simple approach yields pleasant segments AFAIK.