karlhigley / lexrank-summarizer Goto Github PK

It would be nice to know how many or what proportion of sentences are removed by the boilerplate filtering. Spark's accumulator variables would provide a good way to track that.

Using the Dataframes API instead of using RDDs directly may provide a speed improvement through the use of the Catalyst optimizer.

Since the summarizer is already computing term and document frequencies, it would be nice to be able to specify that the top X terms by one of these metrics should be treated as stopwords.

It's likely more efficient than the current flatMap based approach.

It would be nice to sort the excerpted sentences according to the order that they appear in the original documents, rather than their PageRank scores. If this were possible, the excerpted sentences could be presented back to back in paragraph form, rather than as a list of bullet points.