Recommending the best anime for myself

source: https://media.giphy.com/media/3ohc0PsZLDN5Dgcf5e/giphy.gif

source:https://media.giphy.com/media/rCdzKS756yiGs/giphy.gif

This is one of my visualizations from my data visualization class subjected to the problems I, an anime fan personally encountered when I tried to watch anime these days. There is one annoying thing about watching anime is that sometimes we can not decide which anime should we watch and randomly watch often come with disappointment.

Since there are so many anime out there trying to select an anime that match our likings sometimes can be troublesome. Therefore, I try to ease this problem by using various methods to find an anime list that will match my liking.

Data

I used data is collected from myanimelist.com which is the internet anime and otaku community that is available in kaggle.com. The data is involved around the anime and the user in the myanimelist.com community.

Rating or Score

I plot the number of votes against the score. The result I obtained is that there is a positive correlation between votes and score the higher meaning the higher the number of votes results in higher rating or score. In addition, there is also a case where an anime gets a very high rating with only a few voters.

For example,

Anime: Naruto received 7.8 ratings with 50000 voters

Anime: Naruto Shippuden received 9 ratings with 3voters

which of the two do you think is better? Are you certain?

Thus, the rating is not a very useful tool to measure which anime is better. Consequently, selecting an anime to watch based on rating alone is not an appropriate method.

Weighted Rating

Therefore, to determine which anime is the better we have to apply a new rating that takes into account the number of voters and the rating scores. Adopting the IMDB weighted rating to determine which is the best anime.

weighted rating (WR) = (v ÷ (v+m)) × R + (m ÷ (v+m)) × C

Where:

• R = average for the anime (mean) = (Rating)
• v = number of votes for the anime = (votes)
• m = minimum votes required to be listed in the Top
• C = the mean vote across the whole report

In this calculation for minimum votes are required to be listed in the Top(m). I will use 90th percent as a minimum criterion.

These are the results.

These are the absolute best anime measure by the new ratings. This list can be useful for someone who starting to watch anime and should go for the absolute best ones.

However, it didn’t really solve my problem. In fact, what kind of an anime fan are you if you haven’t watched those best anime of all time, and not only that personally, I choose to watch anime based on my liking, not the rating.

Applying Weighted Rating

Therefore, in the next step, I will apply the new rating to my personal liking, the anime genre which is action and comedy.

The results here are more useful than previous findings since I can use the new rating that is not biased to choose the anime from my favorite genre to watch.

Similarity search

This method is for some individuals who are willing to take the anime recommendation to the next level. This method will be quite different from the methods up above. This method is used to find anime that is similar to the anime that you like by taking into account the major factors which are the title of the anime, the studio, genre, and the producers.

1. This method required you to choose the references anime that you like to be the prototype
2. Use combined the four major factors of each anime into a string in the other words into a single sentence.
3. Use the cosine similarity to find the correlation between the prototype and each anime. The anime that has the highest correlation with the prototype is the anime that is closest to the prototype.

Step 1: Since bleach is one of my favorite anime I chose bleach as my prototype.

source: https://media.giphy.com/media/4fCFRodZ5tJu0/giphy.gif

Step 2: combining the 4 factors

Step 3: Read the results

The result shows that Naruto Shippuden is the closest anime to the prototype which is bleach. In the right column, it shows the correlation between the anime and the prototype meaning the higher the correlation the closer anime to the prototype.

For some who would like to try out my code, you can click here and choose an anime recommendation file and can try to play with it.

Conclusion

To sum it up, I think the optimal way to use this method for me is to use the weighted rating with my favorite genre and select my most favorite anime and use the similarity search to find similar anime. In contrast, for someone who is completely new to anime you should go straight for the absolute best anime first then follow my methods accordingly.

That is all for my anime recommendation methods I hoped you enjoy it. I wish that shortly I will have opportunities to present you guys with an interesting topic again. Thank you

references :

Movie Recommender Systems

MyAnimeList Dataset