#### Model

This first model use all differents feature with a weight quite balanced between all of them. The visualization is based on the

t-SNE result in 3D with parameters set in order to minimze the KL-divergence.
As a result, all points are homogeneously distributed in the space.

#### Rendering

The rendering is done in WebGL using

Three.js. Every movie is an Icosahedron (a sphere is too expensive in term of memory to be rendered with this quantity of movies). The closer 2 points are, the more similar they are.

#### Model

This second model use also all differents feature with the same weight as model 1 but the visualization is based on the

DOSNES. This algorithm is design to force points to be placed on a sphere with respect of distances as it is with the t-SNE. As a result, all points are homogeneously distributed in the space.

#### Rendering

The rendering is also in WebGL using

Three.js. Nevertheless, in order to reduce performances, very movie is a point of a texture on a Sphere. The closer 2 points are, the more similar they are. In addition, a color is applied on every points based on a score given with number of votes and average score.

#### Model

The third model use the same pairwise distance matrix as the 2 previous model. The idea was to try to make a representation also in the sphere of a 2D

t-SNE. To do so, a

Stereographic projection has been applied on the best result of the algorithm. In order to no pack to mush differents value on poles, the projection is only on 180 degrees.

#### Rendering

The rendering is also in WebGL using

Three.js. Nevertheless, in order to reduce performances, every movie is a point of a texture on a Sphere. The closer 2 points are, the more similar they are. In addition, a color is applied on every points based on a score given with number of votes and average score.

#### Model

This fourth model use also the

DOSNES algorithm but with differents weights. Based on feedbacks, the idea was to increase a lot more points based on their type (action, Drama, ...). This creates kind of galaxies per types.

#### Rendering

The rendering is in WebGL using

Three.js. Similarly to the model 2 and 3, the rendering is also points on a Sphere but now the camera is inside the sphere and it's like if you are looking at stars during the night. The brighter the star is, the better is the movie.

#### Model

This last model use the same model as the model 1 (

t-SNE 3D) but with the matrix having more weight per type of movies. Unfortunately, their is a lot of bad predictions due to a missing type of movie (For example Thor is far from Thor: Ragnarok) as they don't share all styles in the data.

#### Rendering

The rendering is in WebGL using

Three.js. The code used is exactly the same as model 1 with same geometry and parameters. For performances purposes, there is also the restriction of number of movies by default.