To get the Roar dataset .csvs, view releases
It was necesary to create an artificial dataset to train three (3) essential Roar Network components:
- the classifier of roars
- the recommenders of roars
- the sentiment analizer of roars
Roar dataset is based on Yahoo Answers Topic. Each roar belongs to one of the following categories:
- Society & Culture
- Science & Math
- Health
- Education
- Computer & Internet
- Sports
- Business & Finance
- Entretainment
- Family
- Governemnt & Politic
A roar is a short text with a maximun of 512 characters.
Then, as step number one, we take as a roar each title + answer of Yahoo Answer Topic, but just those with a maximun of 512 characters.
In addition,it was created a preference probabilistic graph, which represent the probability of user A which like category m, like category n, where n != m.
So, we generated 5000 users. A user is created by adding an array of random preferences k with different main categories.
Then, the categories that the user likes the most are selected, and they select random roar and they are rated.
The image below represent a visualization of users preference in two dimension using no-supervised algorithm t-distributed Stochastic Neighbor Embedding.