HIGHLIGHTS
SUMMARY
Despite the popularity of text embedding models, an important question to ask is: how good are the text representations, given a research task? Specifically, the authors can treat what an embedding model needs to encode as the construct of interest, the embedding model as the measure, and the embeddings as the measurements. In Sect 4, the authors apply the adapted construct validity testing framework to a concrete application, where the authors show how to evaluate the quality of embedding models when they are used to encode survey questions. To apply the idea of construct . . .
If you want to have access to all the content you need to log in!
Thanks :)
If you don't have an account, you can create one here.