Home>

Beginning this summer, AI beginners began to study artificial intelligence by purchasing reference books.
The other day, I thought that it would be interesting if there was a typing game where artificial intelligence learned the weakness of the player and raised a problem that could be overcome, but what is the learning method of artificial intelligence in this case? I do n’t understand, but I ’m curious.
I also want to make it if I can make it myself, and I would like to make a reference book purchase.

Game image

As for the image of the game, not only weak keys etc. are presented as important issues, but typing with one key followed by another key has many mistakes and humans are difficult to grasp It's like finding weak points.
Artificial intelligence learns weakness and finds words that contain poor letters and enumeration from a dictionary and gives you questions.
The typing game "Typing Bolt" (https://www.typingbolt.com/) feels like a Japanese version.

My guess, unsupervised learning?

In this typing game, there is no clear answer to the question to overcome. I think.
I thought that it would be possible for reinforcement learning to be considered as a reward for improving the skills of the player by taking the questions, but I settled on unsupervised learning when considering using typo statistics.

Thanks for your understanding.

  • Answer # 1

    It's hard to bring up machine learning, and there is no guarantee that machine learning will fit your purpose well. If your goal is to study machine learning, we recommend that you solve problems that you know are suitable for machine learning.

    Machine learning doesn't matter, so if you want to make a typing game, let's make it without taking out machine learning first.

    It ’s super simple.

    Put a lot of problems

    Determine probabilities and scores for each problem and answer questions accordingly

    Problems with high error rate are increased

    Rather than measuring the probability in units of problems, if you do well using the per n-gram character, `` Of course, the key that you are not good at will be presented as a problem. "Typing with another key followed a lot of mistakes" can be reflected.

    As for the logic of questions, how much the error rate is reflected in the probability of questions, measure not only the error rate but also the input time, use something, control the output interval well, etc. If you think, you can make various things.

    What happens when you use machine learning? This may be interesting, but suddenly going to machine learning tends to be a brave foot.

  • Answer # 2

    Although it is not an answer, Atok tells you the tendency of the total number of characters and conversion errors every month.

    Understand your own mistake tendency -ATOK Monthly Report-

    I think typing software is full of cocoons, so I think it will be helpful to consider from a different angle.