Text & computational biases workshop by Ramin Soleymani

In this workshop, framed in The II Interface Politics Reserach Congress, we are going to discuss how algorithms that process text, reflect their inherent biases of the text sources they are biased on.

After a theoretical part where we look at examples of biased text-based services and their implications we are going to use a software tool, that allows us to create and explore Word Embedding.

Word Embedding are at the core of all digital services that process and produce text (translation, conversational/voice-interfaces, recommender systems that use reviews and names and advertising). They are built by large amounts of text (or other symbols) and can be understood as a map of words, where semantically or functional similar words are close to each other. They have shown to be a powerful structural element, for neural networks to process text and allow new applications, that seemed impossible or extremely hard a few years ago. It has been shown that word embeddings contain biases and prejudices (e.g. racial, gender biases) that are contained in the source text and amplify them when not discovered and used in applications. Google translate for example until today incorporates gender stereotypes.

Date: November 29th
Schedule: from 10:00 to 13:00 3h
Place: Hangar
Free workshop with prior registration

(*) Participants are advised to bring their computer and are very welcome to bring their own datasets of text.

Categories: Training Program |

Cookies: We use cookies on this site to enhance your user experience. If you continue to browse you are giving your consent to the acceptance of the aforementioned cookies and acceptance of our cookie policy. ACEPTAR

Aviso de cookies