Pancake is the name of the tool developed during the Data-Shack program, a collaboration among designers, informatics engineers and data analysts from Harvard University and Polimi. The goal of this tool is to make readable the distance and the relation between some words selected from a dataset.
The project has been commissioned by a cosmetics brand, orienting their production on social network reactions. They used to base their social analysis on English posts, using an algorithm capable of understanding the meaning of each word and its user’s sentiment. Because of the internationality of the comments about the products, they asked for a method considering different languages from English.
A new mathematics model was built for this purpose. It learns the meaning of the foreign words working on their position in the sentence and on the context of the conversation. As designer, I worked to find out a way to visualize the algorithm result. The main problem was related to the amount of the space dimensions considered by the model, which cannot be visualized on a flat surface, like a web page.
For this reason, the tool interface is composed by two levels of visualizations focusing on the same information considered from different points of view. The grid flats (or pancakes) the position of each word on a 2D surface. Because of the inaccuracy due to the simplification from many dimensions to two, gradient reveals the level of the accuracy of each word. The darker the dot is, the more accurate its position is. The position of the light dots is not trustable; they seem close to the words whose meaning is actually different. Because of this misunderstanding a line bar is built. It flats the result in one dimension to highlight the real distance among the words, so that, the user can read the real value of influence.