| This model can measure semantic similarity between pairs of texts containing figurative language. As far as we know, | |
| this model works slightly better than sup-simCSE-roberta-base. For example : | |
| **sentence 1**: I have been in seventh heaven since Harry entered my life . | |
| **sentence 2**: I have been in very happy since Harry entered my life. | |
| the cosin score of simcse: 0.897 | |
| the cosin score of us: 0.897 | |
| ------------------------------------------------------------------- | |
| **sentence 1**: I have been in seventh heaven since Harry entered my life . | |
| **sentence 2**: I have been in pain since Harry entered my life . | |
| the cosin score of simcse: 0.846 | |
| the cosin score of us: 0.753 | |
| -------------------------------------------------- | |
| It's still a big challenge for us to measure semantic similarity of figurative language from the sentence embedding perspective. | |
| unsupvised models may useless as the key is to infer the literal meaning of the figurative expression, since the annotated is rare. | |