Using Posters and Deep Learning to Recommend Anime & Mangas - - PowerPoint PPT Presentation

using posters and deep learning to recommend anime mangas
SMART_READER_LITE
LIVE PREVIEW

Using Posters and Deep Learning to Recommend Anime & Mangas - - PowerPoint PPT Presentation

Using Posters and Deep Learning to Recommend Anime & Mangas Jill-Jnn Vie, PhD RIKEN Center for AIP, Tokyo July 5, 2018 Outline Mangaki Recommender system for anime & manga Why nonprofit? Research The geometry of mangas


slide-1
SLIDE 1

Using Posters and Deep Learning to Recommend Anime & Mangas

Jill-Jênn Vie, PhD RIKEN Center for AIP, Tokyo July 5, 2018

slide-2
SLIDE 2

Outline

Mangaki

◮ Recommender system for anime & manga ◮ Why nonprofit?

Research

◮ The geometry of mangas ◮ Deep learning: learning from massive data (it’s all blocks) ◮ Our new algorithm (name will come, wait for it)

Future

◮ ???

slide-3
SLIDE 3

Mangaki, recommendations of anime/manga

Rate anime/manga and receive recommendations 2,000 users, 10,000 anime/manga, 350,000 ratings ◮ myAnimeList (RIP their API) ◮ AniDB ◮ AniList ◮ (soon) TVtropes

slide-4
SLIDE 4

Mangaki

slide-5
SLIDE 5

Build a profile

slide-6
SLIDE 6

Mangaki prioritizes your watchlist

slide-7
SLIDE 7

Browse the rankings: top works

slide-8
SLIDE 8

Why nonprofit?

◮ Why should blockbusters get all the fun/clicks/money? ◮ Maybe there is one precious, unknown anime for you

◮ and we can help you find it

◮ Driven by passion, not money Everything is open source: github.com/mangaki (Python, Vue.js) Awards: Microsoft Prize (2014) Japan Foundation (2016)

slide-9
SLIDE 9

Browse the rankings: precious pearls

slide-10
SLIDE 10

Embedding you in the geometry of mangas

slide-11
SLIDE 11

Many blocks for recommendation

KNN K-Nearest Neighbors ALS Alternating Least Squares FMA Factorization Machines Input: ratings Output: representation of users and items (geometry used for reordering & recommendation)

slide-12
SLIDE 12

A new block appears!

Illustration2Vec (by Masaki Saito & Yusuke Matsui, 2015)

slide-13
SLIDE 13

We have (many) posters

slide-14
SLIDE 14

Existing algorithms

ALS Take ratings → recommendations Does not work when no ratings are available (ex. new work) LASSO Take tags/ratings → recommendations Does not take care of people with similar taste I2V Take posters → tags

slide-15
SLIDE 15

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ

Blended Alternating Least Squares with Explanation

slide-16
SLIDE 16

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ γ = 0

Blended Alternating Least Squares with Explanation

slide-17
SLIDE 17

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ γ → ∞

Blended Alternating Least Squares with Explanation

slide-18
SLIDE 18

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ γ = 0.27654

Blended Alternating Least Squares with Explanation

slide-19
SLIDE 19

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ γ = 1.48759

Blended Alternating Least Squares with Explanation

slide-20
SLIDE 20

Our new algorithm

Combine blocks Use machine learning to select the best model

posters Illustration2Vec tags LASSO ALS ratings β, γ Steins;Gate

Blended Alternating Least Squares with Explanation

slide-21
SLIDE 21

BALSE

slide-22
SLIDE 22

Our paper BALSE was accepted at the MANPU workshop!

Genuine excerpt from the reviewers

Two models individually are existing methods, but this work presents a novel fusion method called Steins gate to integrate results given by two models.

slide-23
SLIDE 23

Make your neural network look at the manga

Extract frames from episodes Cowboy Bebop EP 23 “Brain Scratch”, Sunrise

slide-24
SLIDE 24

Watching assistant

slide-25
SLIDE 25

Take home message

◮ Using machine learning, you can predict the future ◮ Machine learning is made of many blocks put together ◮ But they need data ◮ To get data, interoperability is super important

◮ Let’s work together to pair myAnimeList / AniDB / TVtropes! ◮ Contact us to know more

slide-26
SLIDE 26

Still, there are things we can never predict

slide-27
SLIDE 27

Thanks! Do you have any questions? jj@mangaki.fr

mangaki.fr/map github.com/mangaki