5 Collaborative filtering

 

This chapter covers

  • Designing proper graph models for a collaborative filtering approach
  • Importing existing (nongraph) datasets into the graph models designed
  • Implementing working collaborative filtering recommendation engines

The content-based (also called content-filtering or cognitive) approach to recommendations described in chapter 4 creates profiles for users and items to characterize them. The profiles allow systems to match users with relevant items. The general principle of content-based methods is to identify the common characteristics of items that have received favorable feedback from a user (a positive rating, a purchase, a click) and then recommend to this user new items that share these characteristics. Content-based strategies require gathering information that might not be readily available, easy to collect, or directly relevant.

5.1 Collaborative filtering recommendations

5.2 Creating the bipartite graph for the User-Item dataset

5.3 Computing the nearest neighbor network

5.4 Providing recommendations

5.5 Dealing with the cold-start problem

5.6 Advantages of the graph approach

Summary

References