nreimers
commented
4 years ago Hi @ironllamagirl I think out of the box the models don't capture it.
If you have training data, you could of course fine-tune the sentence embedding to have such properties. You would need to have sentence pairs and the label (0...1) that indicates whether both sentences talk both about a risk or both talk about an opportunity.
Best Nils Reimers
Hi!
Thank you very much for this package. Very helpful! I have a question. Do you think it is possible to implement this in a way that distinguishes between 'positive' and 'negative' sentiment as well? I understand this would be too much to ask from model and that this could be done using sentiment analysis. But I was wondering if there is a way to build it so that if the query has a 'negative' connotation, such as a risk, then I would get results containing risks/opportunity only, so sentences containing negative/positive connotation. Can the embeddings capture this type of sentiment?
Best,