Multi-Level Summarization in Instapaper

Today Instapaper launched Summaries, which is a feature I’ve wanted to build for a long time. Summaries help readers both understand an article before reading it, and help them recall the details of previously read articles.

Building summarization features has become a whole lot easier with new tools and APIs, and this post will outline the technical details on how Instapaper generates its summaries.

For instance, Instapaper Summaries use TextRank for Sentence Extraction, which is based on the premise that the most relevant sentences have the most similarity to other sentences in the document. The algorithm identifies the most relevant sentences by constructing a graph where every node is a sentence, and sentences are connected by similarity based on overlapping words.

Instapaper uses an open source library to build the graph using the TextRank algorithm, grabs the top 15 sentences (or fewer when fewer are returned), and then sorts them in the order they appear in the text. In the application code:

Leave a Comment