Data without so much context?

[Evelynhuang]

Hi,

Thank you for this very interesting paper and important research question and contribution. I just have a question. Is this model more suitable for long text data/documents and not so suitable for short texts data or data without much context (ex. very short open-ended survey answers)?

Best,
Evelyn

[ArthurSpirling]

@Evelynhuang -- thanks for your query. @prodriguezsosa put together some code for looking specifically at open-ended survey answers, actually. See this part of the guide

Assuming you already have some reasonable pretrained embeddings, we found that you can get reasonable ALC embeddings from just a single instance of a term (and its surrounding ~12 words in context)---see the trump/Trump example in the paper. But I'm not sure if that's the sort of thing you're asking about.

If you clarify your use-case a little more, I will try to say something more definitive.