Generative poster prototype v1

Making a software generating poster using different machine learning algorithms.

First an algorithm generates poems after learning to write using the recueil Poèmes Saturniens from Paul Verlaine. Once generated, it chose one of the words generated randomly and parse it into two word2vec algorithms, these consist of creating a dictionary of word based on text.

Thanks to the iteration of word in different context, Word2vec can link each word of a text to a vector data following its “meaning”. Close word would have close vector.

representation of the program

So, one of the Word2vec algorithm was trained on all of the Poemes Saturniens and the other was trained on the full Wikipedia (The Wikipedia models was found online

Through all these algorithms, I question notion of system made by these algorithms to elaborate a generation and a classification of word. Also it emphases the different interpretation of each word: both input the same data but interpret it in a way different way.

Once it found the term in the word2vec dictonnary. The program generate a poster using the generated text, both of the word2vec outputs and make a representation of it (through a 3d generated form).  

I also want to represent what’s “under the hood”. Machine Learning algorithms tend to enwrap their result in a “black box” opaque way. Trying to show the vector part

Following Ferdinand de Saussure vector part show the “signifié”: the abstraction/concept of the term whereas the “signifiant” is the term/word.

Overlaying both representation of the same word using two different word2vec algorithm shows the difference and common point of the same word. Shows how the computer can interpret word, and how input data change that: the importance of the dataset.