Explorers and Exploiters

In broad terms, a recommender system:

1. Has a set of options to suggest
2. Ranks those options according to some criteria
3. Returns those options in ranked order

But this process is not exclusive to recommenders. Humans make decisions in a similar way.

For example, when choosing a career:

1. We have a set of jobs available to us.
2. We evaluate those jobs based on our preferences and their market value.
3. Choose the jobs maximize our value(enjoyment, salary) function.

The same can be applied to our leisure time, replacing “jobs” with “hobbies” and “salary” with “prices.”

So how can we use what we’ve gained by developing recommenders to help us make better decicions in our own lives?

Exploring and Exploiting

One of the fundamental tradeoffs in recommender system design is the balance between exploring and exploiting user preferences.

A recommender system explores user preferences by recommending things that it’s unsure if they’ll select. It exploits that knowledge by recommending options the user is most likely to select.

Neither extreme is ideal, and effective recommenders find a pleasing balance between the two. For example, a recommender of books which is too exploitative might only suggest books by an author the user has already read. But a recommender which is too exploratory would suggest such random books that the user would hardly call them “recommendations” at all. Ultimately, users grow bored with recommenders that do too much exploring or exploiting.

Explorers and Exploiters

Just like the recommenders, we are all explorers and exploiters. We have a life to fill, and all the possible jobs and hobbies in the world to fill it.

Too much exploration and we won’t:

• See our skills improve over time.
• Gain a deeper understanding of how our activities relate to the larger world.
• Benefit from the efficiency gains of specialization in our careers.

Too much exploitation and we won’t:

• Learn principles from other fields which can inform the methods we use for our primary one.
• Be able to adapt when new technologies or changing labor needs threaten our specialization.
• Have fallback hobbies if we experience diminishing returns on our exploited ones.

Ultimately, a happy balance between exploration and exploitation is required for us to thrive.