Recommender systems automate the process of making real-time recommendations to customers.
A simple example: an online customer who is browsing a store for one item (e.g. a power drill), places the item in their shopping cart, and is then recommended to buy a complementary item (e.g., a set of drill bits). This example is trivial. Machine learning can go further, often uncovering unexpected buying patterns, based on unforeseen relationships between different customers and between different products.
A robust recommender system would take into account where on the site the customer had visited, their history of purchases at the site and even their social network history. It may be that the customer browsed for mortar on the last visit to the site. Perhaps the user also asked friends about selecting bathroom tiles on Facebook. In this case it might make sense to recommend a mortar mixing attachment – since it is clear the customer is doing a tiling project. For a machine learning algorithm, identifying non-explicit relationships like this is typical.
A machine learning recommender system improves with time. It learns from successful, and unsuccessful recommendations. A very related application of machine learning is that of placing online advertisements in response to customer behavior/searches (the Google Adwords problem).
The same underlying technology can be used to provide customers with many other kinds of personalized experiences, based on data of many kinds.
Examples of real-time recommendation systems:
- Product recommendations while a user is visiting an online store
- Focused brand message placement while a user is shopping/browsing online
- Dynamic, context sensitive advertising