manual

Recommendation System

The recommendation system helps you find new things to read or watch. It suggests content that matches your reading level when logged in.

The best way to find out what it is exactly, is by trying it out:

Frequently Asked Questions

Have questions about how the recommendation system works? Here are some quick answers to help you understand its features.

How does it know my reading level?

When you create an account on Kimchi Reader, you will start at 0 known word. Kimchi Reader's main feature is a popup dictionary to quickly look up the meaning of words. As you read content, you can mark words as known. The recommendation system will then know exactly what words you know and so can suggest content that matches your level.

How is the complexity score calculated?

Because the recommendation system can be used without an account, it has to order content by a metric. That's why I came up with this new complexity score metric that is independent of the user.

Here are the steps to calculate it:

  1. For a given media item, take all unique words.
  2. For each word, take their frequency rank from the frequency list.
  3. Take the 95th percentile. That's the complexity score.

The rationale is that we can conclude from that number that we can expect some pretty rare words in the content if the number is high.

Although the ranking created from this ordering will be very far from perfect, it was the metric I choose for the first iteration. I expect to come up with a better one in the future. Possibly with a more complex strategy. Feel free to share your ideas if you do have any.

How is the comprehension level calculated?

comprehension % = known word / (known word + seen word + unknown word)

The algorithm used to solve ambiguity is the optimistic one where if one word is found, then we consider it known. If there is at least a seen word, then we consider it seen, otherwise it's unknown.

This ordering is only available when you are connected with an account. Once you reach 100 known words, within 24h it should compute your stats against the entire database of content.

When does the refresh happen?

About every 24h for active users. To be considered active you need to have changed a status or mined a new card.

You can see when the last run occurred in the header section of the recommendation system. This is only visible if you had a run in the past.