Session 14: Treasure hunt

Your goal is to examine the “experimental” models in the Sandbox or RStudio Server and build a forensic case for the specific properties or limitations in their preparation. Feel free to chat as a group and share notes and ideas together. Try looking at clusters, run a few queries (especially comparing results between these models and the “WWO Full Corpus” model), and test some of the operations, such as additions, subtractions, and analogies. You can even run some of our validation scenarios in RStudio Server, if you’re feeling adventurous.

You have two options for your exploration:

• Option one: This option focuses on finding traces of the choices that were made in corpus preparation. Look at the “experimental corpus” models: what can you determine about the selection and preparation of the texts used to train the models? What are the specific clues you can find by exploring the models? How are the choices made in text selection and preparation impacting your results when you query the models?
• Option two: This option focuses on finding traces of the decisions made in training the model (i.e. the parameter settings), looking at the “experimental preparation” models. How does the number of dimensions seem to be impacting your results? What happens when a model is trained on a very small corpus? What changes when the window size is decreased?

For either option, make a list of the evidence you can find about data preparation or model training, and develop some notes about how these choices seem to be impacting the models.

Word Vectors: Hands-on Practice and Group Work, Intensive Pedagogy-focused, slide 9 of 11

© 2019 Syd Bauman, Julia Flanders, Sarah Connell, and the Women Writers Project This TEI-encoded XML file is available under the terms of the Creative Commons Attribution-ShareAlike 3.0 (Unported) license.