6 Session-based recommendations

 

This chapter covers

  • Implementing recommendation systems by using session data
  • Designing graph models for session-based recommendation engines
  • Importing existing datasets into the graph models

Chapters 4 and 5 introduced two of the most common approaches to implementing recommendation engines: content-based and collaborative filtering. The advantages of each approach were highlighted, but several drawbacks also emerged during the discussion. Notably, these techniques require information about users that is not always available. This chapter covers another approach to recommendations that is useful when it is difficult or impossible to get access to user interaction history or other details about the users. In such cases, applying the classic approaches would not produce good results.

6.1 The session-based approach

Suppose that you would like to build a recommendation engine for an online travel site. The site offers lodging reservations but doesn’t require login or registration in the early stages of the process. Using a session-based recommendation engine, it is possible to deliver recommendations even in cases like this one, in which little about the user is known.

6.2 The events chain and the session graph

6.3 Providing recommendations

6.3.1 Item-based k-NN

6.3.2 Session-based k-NN

6.4 Advantages of the graph approach

Summary

References