In this session we'll shift our focus from set retrieval to ranked retrieval. This simplifies two things for the user: (1) the query formulation process becomes easier because Boolean logic is not required, and (2) using the results becomes easier because the user can be offered an almost unlimited number of results, sorted in a way that makes it easy for them to recognize what the system estimates are the best ones. We'll do this by computing a score for each document and then sorting on that score, so the art here is in how we compute that score. I'll briefly describe a few different ways of doing that, but then we'll focus on one, called BM25.

To prepare for this class, read 237 (from the start of section 7.1.3)-252 (stop at the start of section 7.3) of the Croft textbook.

We'll do two things in this session:

A lecture on ranked retrieval. The slides used in class are available as Powerpoint or PDF.
An in-class lab session in which we will use the Session 8 Google Colab notebook to do some ranked retrieval. All Collab notebooks can be viewed on the course's Google Drive site.