14:35 | 15:15
Keywords defining the session:
Takeaway points of the session:
- Small tweaks can lead to big improvements in the Learning to Rank space.
- You don't need to be a data science genius to make a start!
I will begin by motivating the importance of Learning to Rank using a real search engine which is currently implemented in production. I will discuss their existing methodology for ranking results, and show the downfalls of their methodology.
I will then go on to discuss the basics of Learning to Rank. I will explain normalised discounted cumulative gain (nDCG) which is the main metric used to determine how good the results returned for a specific search query are.
I will show that standard linear regression can be used to rank search results, and will show how successful such a method can be. I will then go on to introduce the topic of decision trees, and show how they can be used in Learning to Rank. We will compare the methods, focusing on the nDCG and the complexity of each of the algorithms, and I will point out standard python libraries which contain off-the-shelf versions of these methods.