A ‘good’ YouTube video but a bad learning material

YouTube is full of learning materials. A lot of amateurs, professional educators, as well as studios produce content daily.

In this example, we will take a look at a YouTube video that is highly favored by the algorithm but is probably not that educational. It is also a well produced video, that is not that common.

Crash Course is a great series of educational videos that are very dynamic (as required by YouTube standards). This particular video about derivatives seems to be too dynamic and might not be able to bring the point across.

Here it is:

Its performance on YouTube is remarkable:

However, let’s look at the top comments. While there are positive comments (you can check them by yourself on YouTube), there is a lot of critique from people who claim to be domain experts. For instance:



and many other people question whether this video can be actually effective for teaching the mathematical concepts.

To sum up. It is a super successful video by YouTube standards: a lot of likes and views, favored by the recommendation series, appears near the top when you search for the keywords. But at the same time it is arguably not a good video to learn about derivatives, and I personally have to agree completely with the comments quoted above.

Now, is there a metric that we can derive from this video to automatically label its efficiency as questionable? One of the ways to do it is to mine comments.

We can summarize the top comments using BERT, and then ask the neural network to complete this sentence “This is a * youtube video.” using everything it has learned from the comments. For this particular video the top two words are:

stupid 0.10022516548633575

bad 0.038492351770401